YouTube2Text

Guide complet 2026 pour utiliser l’intelligence artificielle dans son entreprise — Transcript

by Le Déclic par Alec Henry · 22,814 words · 3,321 segments · language en · Watch on YouTube

Full transcript

  1. 0:00This AI revolution proves that it's
  2. 0:01much more of a revolution in usage than
  3. 0:03in technology. 90%of French people use
  4. 0:05AI every day for two main use cases:
  5. 0:06researching on ChatGPT and writing
  6. 0:08emails. Those are exactly the two use
  7. 0:10cases that I advise against doing.
  8. 0:11There is an AI topic that is going to
  9. 0:12be huge. You won't be able to bill for
  10. 0:13a service the way you billed for it
  11. 0:14yesterday. That's over.
  12. 0:21The first thing when integrating AI is
  13. 0:22to avoid projects that are too big. We
  14. 0:24are missing out on 80%of what ChatGPT
  15. 0:26really is. And so we must start
  16. 0:28learning how to use all these tools
  17. 0:29properly. Start by identifying all your
  18. 0:31office tasks, do them with AI, and you
  19. 0:33will all save an hour and a half per
  20. 0:35day. Service companies are going to be
  21. 0:40transformed by AI. If you are
  22. 0:42well-equipped with AI, you either
  23. 0:43skyrocket your margins or you lower
  24. 0:44your prices compared to competitors and
  25. 0:46grow in volume. Today, we have a
  26. 0:47technology that makes people dream and
  27. 0:49promises incredible things, but I think
  28. 0:51they are completely mistaken. Thank you
  29. 0:56for joining us in ever-increasing
  30. 0:57numbers to listen to Le Déclic. If you
  31. 0:59haven't already, I invite you to
  32. 1:00subscribe, turn on notifications, and
  33. 1:02share a quick comment with us. It
  34. 1:04allows us to have more and more
  35. 1:05inspiring guests each time. And if you
  36. 1:07prefer listening to the audio format,
  37. 1:09know that it's also possible during
  38. 1:10your commute, work sessions, or even
  39. 1:12while exercising. You can find us on
  40. 1:15all listening platforms according to
  41. 1:17your preferences, as well as exclusive
  42. 1:18episodes or just the one you're
  43. 1:20currently listening to. Thanks again,
  44. 1:22it really allows us to keep evolving
  45. 1:25every time. And now, let's get on with
  46. 1:27today's episode. Hello and welcome to
  47. 1:28this new episode of the Le Déclic
  48. 1:30podcast. I am very well accompanied
  49. 1:31today, as always, with Martin Pavanello
  50. 1:33. How are you doing?
  51. 1:34I'm doing great, thank you. Thanks for
  52. 1:35the invitation.
  53. 1:36With great pleasure. Thanks for being
  54. 1:37here. We've been wanting to record this
  55. 1:39episode for a while. You have already
  56. 1:41had the opportunity to speak at one of
  57. 1:42our immersions on entrepreneurs.com.
  58. 1:44You have a superb background. There is
  59. 1:46an enormous amount to talk about
  60. 1:47regarding a subject that is currently
  61. 1:49very popular: artificial intelligence.
  62. 1:52We are going to talk about it at length
  63. 1:54through your background, through
  64. 1:55current events, and especially through
  65. 1:57how to implement it concretely in a
  66. 1:58company, whatever it may be. You have
  67. 2:01been the co-founder and CEO of Mysteria
  68. 2:03since October 2023, an agency
  69. 2:04specializing in generative AI
  70. 2:05integration. Your goal is to
  71. 2:08democratize the use of generative
  72. 2:10artificial intelligence for everyone,
  73. 2:12making AI a driver of competitiveness
  74. 2:14and transformation for professions and
  75. 2:16businesses. You have now trained over
  76. 2:1925,000 people in AI in France, with 250
  77. 2:21corporate training sessions per month
  78. 2:23and over 100 employees in just 2 years;
  79. 2:26that is quite exceptional. AI is
  80. 2:28undoubtedly the most popular topic, but
  81. 2:31perhaps also one of the most
  82. 2:33misunderstood today. We want to
  83. 2:36understand with you how to use it
  84. 2:37concretely in business and in life to
  85. 2:39make it a real performance lever.
  86. 2:41Welcome, Martin, to Le Déclic.
  87. 2:43Thank you very much, a great
  88. 2:44introduction. You've already said a lot
  89. 2:45of things.
  90. 2:45Thank you very much, my pleasure. And
  91. 2:47there is still a lot more to say,
  92. 2:48believe me. You were in real estate at
  93. 2:50the start, and in a few months, you
  94. 2:52became one of the faces of AI in France
  95. 2:54, particularly in real estate. But not
  96. 2:57only that—one of the most visible and
  97. 2:59well-known. What made you make this
  98. 3:01pivot?
  99. 3:02Yeah, that pivot—actually, I created
  100. 3:04my first company in real estate when I
  101. 3:06was in school, which is why I left
  102. 3:08school before getting my degree, and we
  103. 3:10were thinking about how to monetize the
  104. 3:12base we had created. In fact, we
  105. 3:14created a newsletter for real estate
  106. 3:16that addressed real estate agents and
  107. 3:17independent representatives. 20,000 to
  108. 3:2030,000 people followed it; we did a lot
  109. 3:22of advertising, so we monetized that
  110. 3:23way, and at one point, we wondered if
  111. 3:25there was another way to monetize. And
  112. 3:28actually, at that moment, the need to
  113. 3:29monetize the business differently
  114. 3:31intersected with the arrival of AI. So
  115. 3:34when I asked myself that question, it
  116. 3:35was two months before the release of
  117. 3:36ChatGPT. So you see, it was all quite
  118. 3:38linked. I had done a bit of AI in
  119. 3:39school; I did a business and
  120. 3:40engineering curriculum. So, the need to
  121. 3:42monetize, AI arriving, and secondly, we
  122. 3:45also felt that beyond AI arriving,
  123. 3:47there was a need to be supported
  124. 3:49because we saw everything going very,
  125. 3:51very fast. And the fourth point that
  126. 3:53tipped the scales was this need to
  127. 3:55build a slightly bigger company, if you
  128. 3:56will. I had this real estate media
  129. 3:59outlet; it was a great company that
  130. 4:01made a lot of money—and still exists
  131. 4:02today—but you can't scale it because
  132. 4:04we had 40 clients who were the 40
  133. 4:06companies in the ecosystem. Uh, so we
  134. 4:09were saying it's great, there are two
  135. 4:10of us, my partner and I, the company is
  136. 4:12doing well, we're making a very good
  137. 4:13living, but I can't turn it into a
  138. 4:15company that makes 10, 15, or 100
  139. 4:16million in revenue. And so, when you
  140. 4:18put these four ingredients together, we
  141. 4:20thought, well, the arrival of AI comes
  142. 4:21at just the right time. It reminds me
  143. 4:23of what I used to do in class. It
  144. 4:24really matches the need to perhaps
  145. 4:25train real estate agents, and so there,
  146. 4:27we can build a big company. So, all of
  147. 4:29that together. We really started in
  148. 4:30March '23, not as a formal business.
  149. 4:33That's why we tend to say October,
  150. 4:34because first we launched a newsletter
  151. 4:35about AI.
  152. 4:36thinking, well, we have this base of
  153. 4:3830,000 real estate agents, let's
  154. 4:39explain to them how they can use a bit
  155. 4:41of AI. So we...I originally created
  156. 4:43Misteria for that, which was a
  157. 4:44newsletter that talked a bit about news
  158. 4:45every week, and actually still exists
  159. 4:47today. And in fact, well, it
  160. 4:48overwhelmed us, because not only did
  161. 4:50those real estate agents join the
  162. 4:51newsletter, but many more people joined
  163. 4:53the newsletter. And so, we quickly grew
  164. 4:55to 40, 45,000 people on the AI
  165. 4:56newsletter, which meant that between
  166. 4:58March and September, we asked ourselves
  167. 4:59: "What do we do with this base we
  168. 5:01created around AI?" And in October, we
  169. 5:02turned it into a real company. But
  170. 5:04that's, in a way, how the transition
  171. 5:05happens: from "I have a real estate
  172. 5:06media" to, actually, I'm switching to
  173. 5:07an AI service company.
  174. 5:08We'll come back to the timeline of
  175. 5:10events and what followed. But before
  176. 5:12that, what made you dive into
  177. 5:13entrepreneurship, this adventure?
  178. 5:15school
  179. 5:15well, I also think it's a cocktail of
  180. 5:17many things. I was in business school,
  181. 5:19I was at HEC, and well, I have parents
  182. 5:20who are entrepreneurs, my brother is an
  183. 5:22entrepreneur, my grandparents were
  184. 5:23entrepreneurs. So, I'm quite convinced
  185. 5:25that intuitively and without really
  186. 5:27meaning to, you're a bit conditioned to
  187. 5:28do that, because you don't have parents
  188. 5:30who come home in the evening saying, "
  189. 5:32My boss did this, my boss did that."
  190. 5:33They talk more about their problems
  191. 5:35with their employees, about how they
  192. 5:36can sometimes go on vacation thanks to
  193. 5:38what they earn, about their
  194. 5:39entrepreneurial challenges. So all of
  195. 5:40that puts you in a mindset from the age
  196. 5:42of 10 where you think, "That looks like
  197. 5:43a pretty interesting job." So I think,
  198. 5:45subconsciously, I've wanted to do this
  199. 5:47for a long time, but I didn't want to
  200. 5:49skip any steps. In school, I did a
  201. 5:52classic prep course and a classic
  202. 5:54business school to get a standard
  203. 5:55management background. So, I really
  204. 5:58didn't want to rush things. But the
  205. 6:00real trigger—which fell on fertile
  206. 6:02ground—was during my gap year; I did
  207. 6:04a 6-month internship in real estate
  208. 6:06investment while at HEC and I loved it.
  209. 6:10You buy big assets, you resell them,
  210. 6:11you restructure them; it was quite
  211. 6:12fascinating, but I was an intern and I
  212. 6:14couldn't stand that. h
  213. 6:15In fact, you make slides all day, you
  214. 6:17work on Excel sheets all day, and you
  215. 6:18see your colleague next to you who is
  216. 6:2010 years older, and he's actually
  217. 6:21having fun. And I told myself, "Well, I
  218. 6:23want to do what he's doing; I don't
  219. 6:24want to be the one making the slides."
  220. 6:26So that's when I stopped at the end of
  221. 6:27my internship. I said, "I'm starting a
  222. 6:29company for the second half of my gap
  223. 6:30year, I'm going to try to build a
  224. 6:31business because that's what I want to
  225. 6:33do." And I want to have fun, I want to
  226. 6:34see clients, I want to be out in the
  227. 6:36field; I don't want to be in a suit
  228. 6:37behind my slides and never see anyone.
  229. 6:39So that’s how the click happened and
  230. 6:40that’s when I started the media. OK,
  231. 6:42clear. So you see this market
  232. 6:44opportunity, you see a need, and you
  233. 6:46create Mysteria A as a newsletter. It
  234. 6:48starts to take off. At that moment, AI
  235. 6:50wasn't necessarily as popular as it is
  236. 6:52today. I feel like we're talking about
  237. 6:5410 years ago, but in reality, it's only
  238. 6:55been 2 years. An enormous amount has
  239. 6:57been done. It’s a topic everyone is
  240. 6:58talking about. Where do you go to find
  241. 7:01information so you can quickly master
  242. 7:03the subject, be able to simplify it,
  243. 7:05pass it on, and, above all, help those
  244. 7:07who need to integrate it into their
  245. 7:09daily lives? That's a very good
  246. 7:13question because, as I said, I did a
  247. 7:15bit of AI in school; I studied AI in
  248. 7:17early 2022, so when I was still in the
  249. 7:19program, ChatGPT didn't exist. So, we
  250. 7:22talk about AI, we talk a bit about
  251. 7:23generative AI, as we mentioned earlier.
  252. 7:25Generative AI technology is much older
  253. 7:26than 2020, well, it's older than 2022,
  254. 7:28but there was no ChatGPT yet, so it
  255. 7:29hadn't been democratized. So what we
  256. 7:31see in school is quite theoretical,
  257. 7:32which is why that's not really what I
  258. 7:34highlighted later in the newsletter.
  259. 7:35And actually, my main source of
  260. 7:37information is the US. I subscribe to
  261. 7:40AI newsletters on Twitter, I follow the
  262. 7:4215 biggest AI creators, I've been doing
  263. 7:44that from the start. And I've already,
  264. 7:46just by telling yourself, by the way, I
  265. 7:47think for those who want to start
  266. 7:48companies, it's a great thing you can
  267. 7:50duplicate in any industry: tell
  268. 7:51yourself "I'll look at what's happening
  269. 7:52in the US" and. In France, we're all
  270. 7:54about 6 months behind, so it's a good
  271. 7:55business. So I say: "I'll look at what
  272. 7:57they're doing, I'll learn from what
  273. 7:58they're doing and I'll see if I can't
  274. 7:59do the same thing in France." So, I
  275. 8:01feed a lot, a lot on American content.
  276. 8:03And it's a big issue with AI, because
  277. 8:05you have many experts who have been
  278. 8:06telling us for 3 or 4 years: "Yes, but
  279. 8:08you come from real estate, so you know
  280. 8:10nothing,""Yes, but you became an AI
  281. 8:11expert even though you didn't go to
  282. 8:13engineering school for 5 years.""Yes,
  283. 8:15you don't know how to develop a machine
  284. 8:16learning tool, so you're not an AI
  285. 8:18expert." But I think they are
  286. 8:19completely wrong. In fact, I think this
  287. 8:21AI revolution, what it proves to us, is
  288. 8:22much more of a revolution in usage than
  289. 8:24in technology. For the tech, there are
  290. 8:26four companies in the world creating
  291. 8:27incredible models that are very complex
  292. 8:29, you know, requiring billions in
  293. 8:30investments. But that's an ecosystem of
  294. 8:3310,000, 20,000, 50,000 people worldwide
  295. 8:35. The users of ChatGPT are billions of
  296. 8:37people worldwide. You see, I personally
  297. 8:39decided I wanted to understand use
  298. 8:41cases and implications for the general
  299. 8:42public much more than claiming to be an
  300. 8:44AI expert who knows how to develop
  301. 8:45machine learning algorithms. So that, I
  302. 8:47don't know how to do. Uh
  303. 8:48And so that has really been—it's
  304. 8:49still my source of info, by the way—
  305. 8:51is to go and see a bit what companies
  306. 8:53are doing. So, drawing inspiration from
  307. 8:54the US and, secondly, seeing what
  308. 8:55companies are doing on the ground. My
  309. 8:57communication today, when you look at
  310. 8:58it, isn't about teaching people what
  311. 9:00machine learning is. It's about telling
  312. 9:01them, here is a bit of what's happening
  313. 9:02in the companies we support. Here are a
  314. 9:04few use cases for ChatGPT that you can
  315. 9:05use. Here are the few ChatGPT
  316. 9:07competitors that are interesting. Here
  317. 9:08are the processes that you can put in
  318. 9:10place. And that, actually, is just
  319. 9:11field experience and a bit of
  320. 9:12monitoring what's happening in the US.
  321. 9:14But that’s the big shift: AI went
  322. 9:16from being in the hands of experts for
  323. 9:1810 years to now being something very
  324. 9:19democratized. You just have to be a
  325. 9:21little interested in what companies are
  326. 9:22doing with it.
  327. 9:22Completely.
  328. 9:24And precisely to understand the
  329. 9:25revolution, do you have a few use cases
  330. 9:27like that to share with us that can
  331. 9:28resonate immediately with everyone
  332. 9:29listening?
  333. 9:30Yeah, absolutely. There are plenty. And
  334. 9:32the main takeaway is that we see this
  335. 9:33as a very office-based revolution.
  336. 9:35That’s also what needs to be
  337. 9:36understood. AI is first and foremost an
  338. 9:38office revolution. So today, all your
  339. 9:40office tasks are being transformed by
  340. 9:42AI. The task of making slides. I make
  341. 9:44slides today using a tool called James
  342. 9:46Spark. It’s a tool that makes slides
  343. 9:48in the Boston Consulting Group style in
  344. 9:503 minutes. The guys, the guys at BCG,
  345. 9:52my classmates who are at BCG, I think
  346. 9:54when they see that, they pull their
  347. 9:55hair out. You can use the same colors,
  348. 9:57the same logos, you get the same
  349. 9:58magnificent charts. Except the info
  350. 9:59comes from an AI that did it for you in
  351. 10:012 minutes. So, slide tools are quite,
  352. 10:03quite incredible. Meeting summary tools
  353. 10:05, and here too, these are very simple
  354. 10:06things. But I use Nouta a lot. At first
  355. 10:09, you tell yourself, well, do I
  356. 10:10actually make meeting summaries? I
  357. 10:11never did them. I never did meeting
  358. 10:13minutes. So I always told myself it’s
  359. 10:14not necessarily a use case for me.
  360. 10:15Actually, I realize now that it has
  361. 10:17applications much further than just
  362. 10:19making a meeting summary. I’ll give
  363. 10:21you an example. In every video call I
  364. 10:22have, Nouta is my assistant; it comes
  365. 10:24with me into the call. At the end, I
  366. 10:26don’t look at the report. Well, the
  367. 10:27report, I don’t even look at it.
  368. 10:28However, two days later, I have a
  369. 10:30commercial proposal to make. I take
  370. 10:31Nuta's transcript, I take their report,
  371. 10:33I send it to my slide tool, and it
  372. 10:34makes 10 slides at once, you see. h
  373. 10:37And here, you go from a tool that just
  374. 10:38makes meeting content to one that
  375. 10:40actually helps you create a highly
  376. 10:42personalized sales proposal. Well, it
  377. 10:44makes a slide for me saying, "Hey,
  378. 10:45watch out, the client mentioned their
  379. 10:47training need is more advanced, not
  380. 10:49really for beginners," and so on; so
  381. 10:50that's a tool, for example, that I use
  382. 10:52every day now. So everything regarding
  383. 10:53slides, everything regarding reports,
  384. 10:55and then standard ChatGPT. I use
  385. 10:57ChatGPT a lot for myself; I use it for
  386. 10:59questions here, perhaps to change up
  387. 11:00what we’re used to. Everyone tells
  388. 11:02you it helps you do all the
  389. 11:03low-value-added tasks. It’s true,
  390. 11:05there are plenty of low-value-added
  391. 11:06tasks where AI is very strong. It’s
  392. 11:08also very strong for high-value-added
  393. 11:10tasks. I have the latest model, GPT-5
  394. 11:12Pro. When I'm recruiting, I use it for
  395. 11:15brainstorming. Meaning, I’ll tell it:
  396. 11:17"I’m hiring a salesperson, here is my
  397. 11:19revenue, here is the margin I want to
  398. 11:21make, here is the number of calls they
  399. 11:22can make per day, here is the type of
  400. 11:24client I’m going to put in their
  401. 11:25hands, etc.""Give me the revenue
  402. 11:27targets I should give them to close
  403. 11:28deals.""And then make me a sensitivity
  404. 11:30table based on what they sign, and how
  405. 11:31much margin I’ll make." It’s a job
  406. 11:33that would take me hours to put on
  407. 11:35paper and so on. And here, it reasons
  408. 11:37for 5 minutes, where it really helps me
  409. 11:39with a strategic question. So you can
  410. 11:41really combine the two: quick wins,
  411. 11:42fast things, but also strategic
  412. 11:44reflections where AI becomes very, very
  413. 11:45strong.
  414. 11:46And this ability to use AI daily,
  415. 11:49creating a new reflex, is a real asset.
  416. 11:54And that is why many, ultimately, don't
  417. 11:55realize how much it can change an
  418. 11:56entrepreneur's life, whether it's for
  419. 11:58handling emails, office work, or using
  420. 11:59it in recruitment. But we could—we
  421. 12:01could—we could mention it for every
  422. 12:03single point. You could even tell
  423. 12:05yourself, "I'll take this breakthrough
  424. 12:07episode, push it into the AI, get a
  425. 12:08transcript, a concrete summary of all
  426. 12:10the clear actions I need to implement
  427. 12:11in my business, and I'll do that every
  428. 12:13time for every episode." Uh, do you
  429. 12:17have a trick, a specific point that you
  430. 12:19tend to share, even within the company
  431. 12:21or with people you train, for
  432. 12:22integrating artificial intelligence as
  433. 12:25a habit? The first thing when
  434. 12:28integrating AI for me is to avoid
  435. 12:30projects that are too big. Actually,
  436. 12:32you end up missing out on 80%of what
  437. 12:34ChatGPT is, for example. There are many
  438. 12:36companies that tell me, "Okay, we've
  439. 12:38deployed 100,000 euros on a project,
  440. 12:40we're going to try to develop a tool
  441. 12:41that will automate 100%of our processes
  442. 12:43." I say, "Do your employees even know
  443. 12:45how to use ChatGPT or Claude or an
  444. 12:46equivalent yet?" No. So let's start
  445. 12:48with that. That's a quick win. It’s
  446. 12:50actually when you train quickly and get
  447. 12:52a few licenses that you can already
  448. 12:53achieve major productivity gains. So we
  449. 12:55really insist on that. There are plenty
  450. 12:58of things to go after, as the Americans
  451. 12:59say, "low-hanging fruit," you see.
  452. 13:01There are lots of easy things to go
  453. 13:02after like that. And so you have to
  454. 13:04start by really learning how to use all
  455. 13:06these tools. So I would advise them to
  456. 13:08do that instead. Start by identifying
  457. 13:10all your office tasks, doing them with
  458. 13:12AI, and right there, you'll all save an
  459. 13:14hour and a half a day. That is really
  460. 13:17the big topic. And after that, we'll
  461. 13:19get into more advanced automation. Are
  462. 13:20there processes we can automate? Are
  463. 13:22there customer services where we can
  464. 13:23add more AI, where we go a bit further,
  465. 13:25and so on? But in any case, I think the
  466. 13:26biggest part of integrating AI is not
  467. 13:28to go too fast. There's an MIT study
  468. 13:31that came out two weeks ago saying that
  469. 13:3395%of AI projects don't finish and have
  470. 13:35no ROI for companies. And when you dig
  471. 13:38deeper—that's the headline on TF1,
  472. 13:39published everywhere, Les Échos,
  473. 13:41everyone is happy. But actually, when
  474. 13:42you dig deeper, you realize that all
  475. 13:44the projects mentioned were, first of
  476. 13:45all, projects done by large
  477. 13:46corporations. It's all development and
  478. 13:48automation projects and AI agents.
  479. 13:51These three things are not yet mature.
  480. 13:53And likewise, when you ask those three
  481. 13:54companies, have you trained your
  482. 13:56employees on Copilot, GPT, or Claude?
  483. 13:57No. So today, we have technology that
  484. 13:59makes us dream, promising us incredible
  485. 14:01things, which are sometimes a bit
  486. 14:02disappointing because, well, the agents
  487. 14:04aren't very mature. So there's no point
  488. 14:06in wasting too much time and money on
  489. 14:07that, which makes us miss out on what
  490. 14:09SMEs are succeeding at. SMEs are
  491. 14:10actually succeeding very, very well at
  492. 14:12mastering one or two AI tools perfectly
  493. 14:14, and you'll already save a crazy
  494. 14:15amount of time, you know. So that would
  495. 14:18be my recommendation on how to approach
  496. 14:20, or rather, what steps to take to
  497. 14:22implement AI. I would say don't look
  498. 14:25for a miracle cure, but rather go learn
  499. 14:28how to use it. Experts say we only use
  500. 14:3015-20%of Chat GPT's capacity.
  501. 14:32Completely.
  502. 14:33It's a shame. It's still a tool that we
  503. 14:36struggle to grasp. I think I saw that
  504. 14:3990%of the French people who use AI
  505. 14:41every day have two main use cases.
  506. 14:43Doing research on Chat GPT and writing
  507. 14:44emails. Those are exactly the two use
  508. 14:46cases that I advise against. Well,
  509. 14:47emails maybe, but doing research, no.
  510. 14:50So you see, we're still there. We have
  511. 14:51the Google reflex, it's a tool we don't
  512. 14:53know well enough, but honestly, they
  513. 14:54didn't develop this tool and they
  514. 14:56didn't just raise 500 billion recently
  515. 14:58so we could write emails and do
  516. 14:59searches with it. Otherwise, there's
  517. 15:01Google. So it's meant to go way, way
  518. 15:02beyond being just a simple tool. So I
  519. 15:04would recommend that: few tools, find
  520. 15:06concrete use cases, and already handle
  521. 15:08that office work before moving on to
  522. 15:09larger projects.
  523. 15:10Very clear. We'll talk more about usage
  524. 15:12and concrete cases in a moment. But
  525. 15:14before that, let's go back to the
  526. 15:15timeline of Mysteria. So October 2023,
  527. 15:18you launch Misteria after doing this
  528. 15:21newsletter. Uh, what are the first
  529. 15:23steps? How does it happen? Because
  530. 15:25quickly you have some success, quickly
  531. 15:26it scales quite hard, you're going to
  532. 15:28have to staff up, you're going to have
  533. 15:29to structure. Uh, we're also going to
  534. 15:32talk about revenue and results, the
  535. 15:33number of clients you've been able to
  536. 15:34support. It's pretty crazy. And then
  537. 15:36there’s also the door left open at
  538. 15:38some point to, I imagine, the siren
  539. 15:39song, because everyone is asking for AI
  540. 15:40. At first, you start in real estate,
  541. 15:43and then, well, you readjust. It might
  542. 15:45be interesting for you to tell us that
  543. 15:46story.
  544. 15:46Yeah, absolutely. So, we launched in
  545. 15:49October 2023; actually, during the
  546. 15:50summer, we realized too many requests
  547. 15:52were coming in from the newsletter for
  548. 15:53training and consulting. We weren't
  549. 15:55answering them, so there was a need, we
  550. 15:56had to go for it. So we created those
  551. 15:58two services. We hired quite quickly.
  552. 15:59We hired two or three people in October
  553. 16:02, mostly consultants. So we raised a
  554. 16:04bit of money. If you look online, we
  555. 16:05raised a little bit of money from
  556. 16:07business angels because, well,
  557. 16:09consultants coming out of top business
  558. 16:10schools cost you quite a bit—several
  559. 16:12tens of thousands of euros a year,
  560. 16:14sorry, a month, all included. Plus your
  561. 16:16clients, the projects last 2 or 3
  562. 16:18months, so they pay you after 4 months.
  563. 16:20So, in short, you have a working
  564. 16:20capital requirement of a few months. So
  565. 16:22we figured we needed to hire a bit
  566. 16:23right away, so we raised a little money
  567. 16:24.
  568. 16:25How much did you raise?
  569. 16:26We raised 800,000 euros.
  570. 16:27OK. We had a little less than a million
  571. 16:29euros, we had said a million because
  572. 16:30there was a mix of equity and debt, but
  573. 16:32we didn't end up taking debt, so a bit
  574. 16:33less in equity. The company gave up a
  575. 16:36little over 10%of its capital, fairly
  576. 16:38standard for a round with business
  577. 16:40angels. So we did that round, we hired
  578. 16:42the first three or four people. It was
  579. 16:45a bit of a struggle at first. As you
  580. 16:47said, we started a lot in real estate.
  581. 16:48We have a pretty strong real estate
  582. 16:49network with my partner because we have
  583. 16:51this company together and he had
  584. 16:52another company too. So it's my brother
  585. 16:54Vincent who is my partner. He has
  586. 16:55another real estate company that does
  587. 16:57consulting, which, for its part,
  588. 16:58doesn't target real estate agents but
  589. 16:59large real estate firms. So we started
  590. 17:01by going to see them at first. We
  591. 17:02looked through that real estate lens.
  592. 17:04So we started with that. As you said
  593. 17:06earlier, it's an ecosystem that isn't
  594. 17:07mature at all. You arrive at a moment
  595. 17:09where everyone wants AI, but when you
  596. 17:10start saying, "OK, it’s going to cost
  597. 17:12you X amount just to lift a finger."
  598. 17:14Everyone says, "Actually, I don't know
  599. 17:15what I need." I want to do some AI, but
  600. 17:16I don't know what. I don't want to use
  601. 17:17ChatGPT. GPT is too dangerous. Get
  602. 17:19training, okay, but on what tool?
  603. 17:20Copilot didn't exist. Develop a tool?
  604. 17:22Well, no, I don't want to put €
  605. 17:2450,000 on the table. So there it is, we
  606. 17:25arrive at the right time, it's the very
  607. 17:27beginning, but a time when companies
  608. 17:28aren't ready. So for a few months, we
  609. 17:31do some small training sessions, we do
  610. 17:33some small audit missions fairly
  611. 17:34quickly. It doesn't move much for 6
  612. 17:37months. I would say that for 6 months,
  613. 17:39we recruit, well it doesn't move much,
  614. 17:41we aren't recruiting, we're doing our
  615. 17:42little missions on our side, and we
  616. 17:44have a first shift that arrives in June
  617. 17:45'24 with a first training course that
  618. 17:47comes out for real estate agents where,
  619. 17:49in fact, immediately, we have thousands
  620. 17:50of people who buy the training, and
  621. 17:52there, we find ourselves with the need
  622. 17:53to staff trainers and recruit trainers.
  623. 17:55That was the first big step, I'd say,
  624. 17:57for Misteria, to have to say, actually,
  625. 17:59I need 10 or 15 trainers all at once.
  626. 18:02So there, that's a bit in June '24. We
  627. 18:04had this first momentum of recruiting,
  628. 18:06and with a model that quickly turned
  629. 18:08toward freelancers because, in fact,
  630. 18:10you are in a business where, even today
  631. 18:12if you take the 100 people, there are
  632. 18:1450 employees and 50 freelancers,
  633. 18:15because doing corporate training is a
  634. 18:17freelance business. In fact, when you
  635. 18:19have employees, and you tell Louis, "
  636. 18:20You are going to go train in Toulouse
  637. 18:21tomorrow," and Arthur, you tell him, "
  638. 18:23You are paid the same salary, but
  639. 18:24tomorrow you are going to go to Paris."
  640. 18:25Well, actually, they look at you at the
  641. 18:27end of the day and say, "Why me
  642. 18:27Toulouse and him Paris?"
  643. 18:28And that is not manageable. Everyone
  644. 18:30tells you, "Ah, freelancers must cost
  645. 18:32less.""No, the freelancer costs you
  646. 18:34more in daily rate, but you have a lot
  647. 18:35less hassle afterward, because in fact,
  648. 18:37with a freelancer, you say: 'I have a
  649. 18:38training in Toulouse, who wants to go?'
  650. 18:40And you always have someone who says: '
  651. 18:41Well, I'm not far, I'm up for it, I
  652. 18:43like it, I know this ecosystem well,
  653. 18:44etc.'"» So very quickly, we told
  654. 18:46ourselves we needed a lot of
  655. 18:47freelancers to provide our training
  656. 18:48services and a little bit of consulting
  657. 18:50too; for consulting, we are using
  658. 18:51freelancers more and more. So, the
  659. 18:53company was structured like that pretty
  660. 18:55quickly, and I would say with real
  661. 18:56momentum starting in June when we
  662. 18:58already integrated 15 to 20 employees
  663. 19:00in June '24; that was the first wave,
  664. 19:02and there, Mysteria went from 6 or 7
  665. 19:03employees to quickly 30 or 35 all of a
  666. 19:05sudden like that.
  667. 19:06That was a bit of a first step. And so,
  668. 19:08right away with this desire to
  669. 19:10structure it more or less 50/50 between
  670. 19:11freelancers and internal staff.
  671. 19:13Yes, quite quickly. Yeah.
  672. 19:14What functions did you keep in-house?
  673. 19:16And we understand that freelancers are
  674. 19:18mainly for the training and support
  675. 19:19side.
  676. 19:20Yeah, exactly. Internally, you keep
  677. 19:22what we call support functions. HR part
  678. 19:24. The HR part in a training company is
  679. 19:26essential because, in fact, you are
  680. 19:27staffing all day long; we were talking
  681. 19:28about 250 training sessions. Well,
  682. 19:30every month, you have to find 250 days
  683. 19:33available from trainers. So you are
  684. 19:35staffing everywhere left and right. So
  685. 19:36you need a strong team. Uh, that's
  686. 19:38important. And the same goes for the
  687. 19:40pedagogy side; in fact, you have to
  688. 19:41prepare the training. You have to
  689. 19:43provide feedback to the companies. So
  690. 19:46you need to have support for that. So
  691. 19:48pedagogy, HR, uh CFO and finance, we
  692. 19:50also need that because there are many
  693. 19:52small line items and so on, so we need
  694. 19:55to have fairly strong financial teams.
  695. 19:57After that, we have all the partners
  696. 19:59in-house. Partner/manager, these are
  697. 20:01all the people who will manage the
  698. 20:02consulting assignments. We have four of
  699. 20:04them today. They see 100%of the
  700. 20:05consulting assignments, they
  701. 20:06participate in 100%of the consulting
  702. 20:08assignments. In fact, when you do
  703. 20:09consulting, you still have these needs
  704. 20:11in a company for long-term support, and
  705. 20:12your client doesn't want to talk only
  706. 20:14to a freelancer, and I totally
  707. 20:15understand that. So, I have a partner
  708. 20:17who will manage the client, uh, the
  709. 20:19client Vinci. The partner will manage
  710. 20:22Vinci, and when they need to, they will
  711. 20:23put a freelancer on a training session,
  712. 20:24a freelancer on a bit of consulting,
  713. 20:26etc., but they are the one who follows
  714. 20:27the whole mission, who sees the client
  715. 20:28regularly, and so on. So our four or
  716. 20:31five partners are in-house, and then
  717. 20:32there are sales reps, a few sales reps,
  718. 20:33including on the B side as well.
  719. 20:35Excuse me, I interrupted you when you
  720. 20:36said "partner." Uh, are they associates
  721. 20:38? Are they associates at Misteria or
  722. 20:39not? or standard start-up stock options
  723. 20:41, but there are two main partners at
  724. 20:43Misteria, Vincent and me. So my partner
  725. 20:45and I, my brother and I, we own most of
  726. 20:47the com—the majority of the company,
  727. 20:49and then there are stock options for,
  728. 20:50indeed, what we call partners and some
  729. 20:52of the management positions at our firm
  730. 20:53.
  731. 20:55So it will soon be almost two years to
  732. 20:57the day since it was officially
  733. 20:59launched, beyond the newsletter. Uh, in
  734. 21:02terms of revenue over two years and
  735. 21:04growth, uh, where do you stand? So it's
  736. 21:07a company that made € 100,000 in
  737. 21:09revenue its first year. Well, the first
  738. 21:11year is September to December, really,
  739. 21:12or October to December, even. Uh, since
  740. 21:14we started in October, last year we
  741. 21:16closed at around 344k in revenue. This
  742. 21:18year, we should do a little over 7
  743. 21:19million in revenue with a profitable
  744. 21:21business model. So, uh, it's hard
  745. 21:23because doubling growth while
  746. 21:24maintaining a margin in B2B like that
  747. 21:26is quite difficult. But the company is
  748. 21:29very profitable. As a service company,
  749. 21:31you generally make between 15 and 30%in
  750. 21:33EBITDA margin. So we find ourselves
  751. 21:35somewhere between those two figures. Uh
  752. 21:36, so, uh, rather good growth, uh,
  753. 21:38prices that are trending downwards,
  754. 21:40which is also increasingly interesting.
  755. 21:42You see, the market is becoming
  756. 21:43structured, competitors are arriving,
  757. 21:45so you can't sell your training day for
  758. 21:46€ 2,500 like you did at the beginning
  759. 21:48. So we are pushing prices down a bit;
  760. 21:50I think that's fairly standard. But, uh
  761. 21:53, yeah, that's how it is today. The
  762. 21:55goals are quite high, you know. It's a
  763. 21:56company we want to take very, very far.
  764. 21:58Uh, the goal is to reach 100 million in
  765. 21:59revenue by 2026. That's pretty, pretty
  766. 22:01clear for us. And then, after that, to
  767. 22:03go further, do some M&A, we're starting
  768. 22:04to think about international expansion.
  769. 22:06Because we are training in eight
  770. 22:07countries now.
  771. 22:08Uh, so, rather very small teams. I have
  772. 22:10clients, uh, clients like Clarence, for
  773. 22:12example, who is a big client of ours,
  774. 22:14who tells us, "Well, it's the France
  775. 22:15team we're training, and now we're also
  776. 22:17going to train Asia, China—Asia, the
  777. 22:19Middle East, and the US." So for now,
  778. 22:21it's mostly our French clients taking
  779. 22:23us abroad, but we clearly want to
  780. 22:24expand internationally too. So, that's
  781. 22:28where we're at today, segmented into 70
  782. 22:31%B2B and 30%B2C. We also do some B2C
  783. 22:35training. More and more. So, that's how
  784. 22:38we're structured, and I think the
  785. 22:40company is destined to be bigger in B2B
  786. 22:42than B2C, even if we remain very humble
  787. 22:45in our approach. The ecosystem is so
  788. 22:47new that we really have no idea what
  789. 22:49the company will look like in five
  790. 22:51years. Will our clients need more
  791. 22:53training or more consulting? Today it's
  792. 22:5550/50, but what it will look like in
  793. 22:57five years is really hard to say. So,
  794. 22:59that's the current state of affairs.
  795. 23:01But trying to predict where we'll be in
  796. 23:03five years is very difficult.
  797. 23:04Absolutely. There are so many things
  798. 23:06you just mentioned that I'd like to
  799. 23:08touch on. You mentioned, quite rightly,
  800. 23:10that it's challenging to double your
  801. 23:12growth from one year to the next. Even
  802. 23:15early on, from 100k to 4 million. So I
  803. 23:18imagine it involved a lot of disruption
  804. 23:21, tough decisions, and needing to
  805. 23:23structure quickly without making
  806. 23:26mistakes. What have been the biggest
  807. 23:29hyper-growth challenges you've faced in
  808. 23:30such a short time so far?
  809. 23:32Yeah, that's a good question. The
  810. 23:34biggest growth challenge is structuring
  811. 23:36, that's for sure. The thing is, I'm 27
  812. 23:39, I've never had a CFO, and I've never
  813. 23:41done the HR or CFO job myself. So at
  814. 23:43some point, you have to trust people
  815. 23:45whose jobs you don't actually know.
  816. 23:46That's hard. You're forced to think,
  817. 23:48well, I hope the CFO is good and the HR
  818. 23:49manager is strong. It's been tough to
  819. 23:51accept that you no longer control
  820. 23:53everything happening in the company;
  821. 23:54even if you try to follow it all, you
  822. 23:56have to trust people. That was the
  823. 23:57first main challenge. The second
  824. 23:59challenge is the business model. Maybe
  825. 24:01it's a mystery, I don't know. We face a
  826. 24:03major business model challenge,
  827. 24:05especially regarding consulting. It's
  828. 24:07very hard to build a consulting firm
  829. 24:09that scales a bit like we do, because
  830. 24:11everyone talks to you about daily rates
  831. 24:13in consulting, how much you pay your
  832. 24:15guy, how many days you manage to place
  833. 24:16them, and so on. That works when your
  834. 24:19company has plenty of clients, like BNP
  835. 24:21spending 1 million euros with you every
  836. 24:23year for 10 years. So you know exactly
  837. 24:25where you're placing your consultants
  838. 24:26and so on. When you have clients who
  839. 24:28spend € 7,000 on consulting, then the
  840. 24:30next one spends € 50,000, and one
  841. 24:32month you have two consulting projects,
  842. 24:33and the next month you have 13,
  843. 24:34that’s when it becomes a nightmare. H
  844. 24:37So, in the beginning, we, all our...all
  845. 24:38our projects were done only by
  846. 24:40employees, only salaried consultants.
  847. 24:41And we had a major business model shift
  848. 24:43, telling ourselves: "Actually, this
  849. 24:44isn't sustainable." With the growth
  850. 24:45we're having, we won't be able to hire
  851. 24:47employees. Because, likewise, you sign
  852. 24:48three times as many projects in a month
  853. 24:50, so what do you think? Should I hire
  854. 24:51two consultants? Except that when these
  855. 24:53consultants arrive, well...you find
  856. 24:55them, and so on, and the projects might
  857. 24:57already be finished. So, in fact,
  858. 24:58potentially, they won't be working for
  859. 25:002 months. Except that you're hiring
  860. 25:01good profiles, so it costs a lot, so
  861. 25:03that's really a struggle. So that’s a
  862. 25:05big challenge: figuring out how to
  863. 25:06pivot our business model to absorb the
  864. 25:08growth of the Consulting BU, in
  865. 25:10particular. Training is easier. For
  866. 25:11training,
  867. 25:13tomorrow, if I launch a trainer
  868. 25:14recruitment drive in my newsletter, I
  869. 25:16get about 200 people applying every
  870. 25:17time. So when there's a need, you see,
  871. 25:20we can onboard 15. I think if we feel
  872. 25:23that we're selling a lot in September,
  873. 25:24in October I'll onboard 15. That,
  874. 25:26quote-unquote, is a bit easier. In
  875. 25:28consulting, it's much harder. So we
  876. 25:30shifted; we went from a model where all
  877. 25:32training and all projects were done
  878. 25:34exclusively by employees. to a model
  879. 25:36where all projects are done exclusively
  880. 25:38by freelancers who are managed by
  881. 25:40employees. But the guy who manages a
  882. 25:42project can manage 5, he can manage 10,
  883. 25:43and if he ever has to manage 15 in a
  884. 25:45month, he'll sleep less, but he'll be
  885. 25:46able to do it.
  886. 25:47because behind the scenes he uses
  887. 25:49freelancers who help him out. But guys
  888. 25:50who produce 5, 10, or 15 in a month,
  889. 25:52well no, normally when you're
  890. 25:53delivering on a project, you can't do
  891. 25:55more than is reasonable. So that was
  892. 25:58the big challenge we had, I would say,
  893. 26:01managing to really, really pivot the
  894. 26:03business model. Then, there's a
  895. 26:06financial challenge I didn't suspect
  896. 26:08either. It's one thing to run a company
  897. 26:11that makes 100k, 200k, a million, or 2
  898. 26:13million. At 7, thank goodness there's a
  899. 26:15CFO. It's a good thing there's a CFO.
  900. 26:16Especially in a business where when you
  901. 26:18have 7 million in revenue and sell
  902. 26:20projects for € 300k, it’s, quote
  903. 26:22unquote, fine, you can keep up. We sell
  904. 26:24daily rates of 2,000. Uh, well now we
  905. 26:26have bigger, much bigger tickets,
  906. 26:28it’s improving, but when we were even
  907. 26:30doing 3, 4, 5 million in revenue, they
  908. 26:32were all tickets between € 500 and
  909. 26:332,000. Uh.
  910. 26:34So you're managing a huge number of
  911. 26:35small line items. Plus, we have a part
  912. 26:37of the business that is state-funded.
  913. 26:39People can get their training
  914. 26:41reimbursed. So it's endless paperwork.
  915. 26:44So financially, you get to the end of
  916. 26:45the year and you say, "Oh wait, I
  917. 26:46actually have massive budget
  918. 26:47discrepancies." So we had to reconcile
  919. 26:49everything, and so on. But that’s a
  920. 26:50challenge where, if I start a company
  921. 26:52again tomorrow, the CFO aspect is
  922. 26:54something to take seriously. Now, it
  923. 26:56depends on the business, but it really
  924. 26:58must be taken seriously. Uh, for us,
  925. 27:00that was the issue. So, there you go,
  926. 27:02I’d say financial, business model,
  927. 27:03and sales too, because when you sell
  928. 27:05like we do, you aren't selling a
  929. 27:06product. You aren't selling a product,
  930. 27:08and, uh, I'd have a 4th point, actually
  931. 27:10, I'll mention it, but you aren't
  932. 27:11selling a product. So every morning you
  933. 27:13wake up and think, if I want to sign
  934. 27:15500k or 600k in services this month, I
  935. 27:17have to go find X number of new clients
  936. 27:20. That's hard when you're selling a
  937. 27:22subscription product. In December, when
  938. 27:24you wake up on December 1st, you
  939. 27:25already know the minimum revenue you're
  940. 27:27going to make that month. When we wake
  941. 27:29up, we don't know the revenue we're
  942. 27:30going to make for the month. So that's
  943. 27:31tough. And also, when the company grows
  944. 27:33slowly, it's fine. When it grows super
  945. 27:35fast, it's hard because regarding your
  946. 27:37investment choices—and this ties into
  947. 27:39my fourth point—the big, big, big
  948. 27:41challenge is building a profitable
  949. 27:42company while growing. Your investment
  950. 27:45choices have to be calculated down to
  951. 27:46the millimeter. You ask yourself, "
  952. 27:48Should I spend 5,000 euros here or not?
  953. 27:49" Because we absolutely want to build a
  954. 27:51company that makes a 15, 20, or 25%
  955. 27:52margin. That is really something we
  956. 27:54want to maintain. We don't want to say,
  957. 27:55"Okay, we accept making 5%or we accept
  958. 27:57losing money for the sake of growth."
  959. 27:59For now, we don't really want that. So
  960. 28:00balancing the two is hard. It means
  961. 28:02being really tight on all your
  962. 28:04financial line items. Why did we pay
  963. 28:06that much for this trade show? Why
  964. 28:07shouldn't we pay less for rent? All our
  965. 28:09teams negotiate all day long. I tell
  966. 28:10them: if you're buying something, you
  967. 28:12negotiate. Because actually, if we all
  968. 28:13negotiate 10%off, that's 10%less in
  969. 28:15expenses. For a company with our
  970. 28:16revenue, that saves hundreds of
  971. 28:18thousands of euros a year, even if it
  972. 28:19comes from small 10%cuts here and there
  973. 28:21on subscriptions, etc. So that has been
  974. 28:23—and still is—a big challenge. It's
  975. 28:25about how we manage to grow while
  976. 28:27keeping our employees in the mindset
  977. 28:29that while we are a tech startup to
  978. 28:31some, in terms of structure, we are
  979. 28:34actually an SMB. So an SMB has to
  980. 28:36invest 1 euro, and that 1 invested euro
  981. 28:38must bring back 2, 3, 4, or 5, but we
  982. 28:40can't just say, "Let's recruit
  983. 28:41infinitely." You see, with hiring,
  984. 28:43sometimes your sales director comes to
  985. 28:44see you and says, "Yeah, I want to hire
  986. 28:46two or three salespeople.""I think it
  987. 28:47would be good." Oh sure, if I had my
  988. 28:49investors 'money, okay. If I'm spending
  989. 28:51money that isn't mine, why not? But
  990. 28:53here, I tell him, "No, tell me why you
  991. 28:54want one, what is it going to bring you
  992. 28:56?""What is it going to cost us?"
  993. 28:57That’s a great challenge
  994. 28:59for us and for the managers because we
  995. 29:00spend a bit more time on this, but the
  996. 29:02manager is less experienced, our guys
  997. 29:03are young and all. So that’s a good
  998. 29:05challenge. How do I grow, how do I not
  999. 29:08spend at any cost in order to stay
  1000. 29:10profitable?
  1001. 29:10It’s a real topic, optimizing growth
  1002. 29:12and EBITDA in the end.
  1003. 29:14People often say it's not necessarily
  1004. 29:15possible to do both, and the challenge
  1005. 29:16here is to succeed in doing both as
  1006. 29:18much as possible.
  1007. 29:18Yeah, exactly. I know a lot of people
  1008. 29:21say you need, I think it's 30%, they
  1009. 29:23tell you that if your EBITDA margin
  1010. 29:25plus your growth is over 30%, your
  1011. 29:27service company is in very good health.
  1012. 29:30So we're well above that, so that's
  1013. 29:32good for now. I think that's it, 30 or
  1014. 29:3440%. So you see, we're trying, and
  1015. 29:36we'll get there. We won't have 100%
  1016. 29:37growth every year, and we won't have 25
  1017. 29:39%margins every year either. We're quite
  1018. 29:41realistic. The more CAC 40 companies
  1019. 29:42you take on as clients, the more you'll
  1020. 29:44get squeezed on price, trade-offs, and
  1021. 29:46so on. So we'll end up like a classic
  1022. 29:48consulting or training firm, hovering
  1023. 29:50between 15 and 20%margin. That's pretty
  1024. 29:51much certain. So that means we have to
  1025. 29:53keep growing by 10, 15, or 20%. But
  1026. 29:54yeah, it's very, very hard. Very, very
  1027. 29:56hard. And it really stands out. No, no,
  1028. 29:58I talk about it a lot, but that's where
  1029. 30:00the French Tech scene is good. I love
  1030. 30:02the French Tech scene, it's cool and
  1031. 30:04all, but building a tech startup where
  1032. 30:05you raise 100 million is so different
  1033. 30:07from building a body shop or a bakery
  1034. 30:09where you only invest a euro once
  1035. 30:10you've actually earned it. Obviously, I
  1036. 30:13respect them, but I admire the baker so
  1037. 30:15much more, who has to pay salaries and
  1038. 30:17thinks: "How do I manage to open three
  1039. 30:19bakeries?" Well, I have to make sure my
  1040. 30:21EBITDA margin on bakery number one is
  1041. 30:23higher so I can invest, but what I'm
  1042. 30:24investing is a risk. So that's what
  1043. 30:26fascinates me, and we want to run our
  1044. 30:28company like that. So that is often
  1045. 30:30underestimated. Everyone says, "Oh
  1046. 30:32great, they raised 100 million, they're
  1047. 30:34so strong, look, they have huge growth.
  1048. 30:36" I always say growth is one thing, but
  1049. 30:37at what price? Growing with other
  1050. 30:39people's money is not the same as
  1051. 30:40growing with the money you generate
  1052. 30:42yourself, and you have a business model
  1053. 30:43quite similar to mine. So, you see what
  1054. 30:45I mean? That's quite difficult. So it's
  1055. 30:47the biggest challenge.
  1056. 30:49Completely. What makes you take this
  1057. 30:51decision to want to scale the company
  1058. 30:53this way? Because, ultimately, uh,
  1059. 30:55there's a culture that's quite, uh,
  1060. 30:56quite young, quite tech-oriented, quite
  1061. 30:58US-style, quite start-up overall,
  1062. 31:00whether it's the AI, the average age of
  1063. 31:02the employees, the hyperscale at
  1064. 31:03certain moments, the fact that at the
  1065. 31:05start you raise 100 million euros—
  1066. 31:07that's not trivial either, it's
  1067. 31:09strategic, 1 million euros but on a
  1068. 31:10valuation of 10, excuse me—it's, it's
  1069. 31:12, it's strategic, it's well thought out
  1070. 31:14. You could have also told yourselves "
  1071. 31:16Okay, well, we're going to keep wanting
  1072. 31:17to do that to flood the market and ride
  1073. 31:18the wave." However, you have a
  1074. 31:21philosophy that I completely share,
  1075. 31:23which is to invest every euro
  1076. 31:25intelligently and build something solid
  1077. 31:27, even with major growth, much like the
  1078. 31:29small businesses you mention.
  1079. 31:33There are plenty of reasons, I think.
  1080. 31:34The first is a bit about the market
  1081. 31:35structure, actually. I think in a world
  1082. 31:37where VC is doing much worse in France,
  1083. 31:38you don't raise money like that anymore
  1084. 31:39. Now, we have a specific theme, so
  1085. 31:41that helps, all the same. But it's
  1086. 31:43harder to raise VC money today, so we
  1087. 31:44arrived at that point, and we said to
  1088. 31:46ourselves it's not necessarily the
  1089. 31:47track we want to follow. We sell
  1090. 31:49services. There are few service
  1091. 31:50companies today that get VC funding.
  1092. 31:52Hm.
  1093. 31:52So, uh, we told ourselves, look, we're
  1094. 31:55in our place, we make money from day
  1095. 31:57one, so we might as well stay the
  1096. 31:58course. Yeah, that was, that was pretty
  1097. 32:01much it at the start, and besides, that
  1098. 32:03doesn't stop you from growing. And, by
  1099. 32:05the way, we see all the companies that
  1100. 32:06have experienced growth. I have many,
  1101. 32:07many friends, and especially Vincent,
  1102. 32:09my partner, who is a bit older. He grew
  1103. 32:11up, he was 25 or 24 years old, between
  1104. 32:1320 and 25, right in the middle of the
  1105. 32:15French Tech boom. Uh, I was that age
  1106. 32:18when French Tech was already doing a
  1107. 32:19little worse. So he had, uh, 50 friends
  1108. 32:21who started companies and all raised 10
  1109. 32:23or 15 million euros. Not a single one
  1110. 32:24of them ended up taking any cash home.
  1111. 32:25Either the companies crashed, or they
  1112. 32:27ended up having an exit below their
  1113. 32:28previous funding rounds, so they didn't
  1114. 32:29make any money, and so on. Not for
  1115. 32:31everyone, obviously, but for many. So,
  1116. 32:33I, I kind of grew up with that too.
  1117. 32:35That didn't excite me, the idea of
  1118. 32:37saying, "Okay, I'm going to raise and
  1119. 32:38raise and raise, and I won't touch a
  1120. 32:39single dime for a while." Uh, I'd need
  1121. 32:43the macro and the economic climate to
  1122. 32:45be good in 5 or 10 years if I want to
  1123. 32:47make money by selling. I told myself,
  1124. 32:49that excites me. And then, when I said
  1125. 32:51everyone in my family was a bit of an
  1126. 32:53entrepreneur, it was always with very
  1127. 32:54classic SMEs. So we also told ourselves
  1128. 32:56, look, this is a healthy business,
  1129. 32:57there are a lot of things to do in AI
  1130. 32:59here. That doesn't stop us from growing
  1131. 33:01, it doesn't stop us from building a
  1132. 33:02company; One Point is worth a billion,
  1133. 33:03you see, our competitor, in quotes—I
  1134. 33:05say quotes because they are much bigger
  1135. 33:06than us and have existed for 20 years.
  1136. 33:08But they are an example for us.
  1137. 33:09Artefact, One Point, these are AI
  1138. 33:11consulting firms that are very big.
  1139. 33:13Well, they've been around for 10 or 20
  1140. 33:15years and the companies are worth
  1141. 33:16billions, so if there's a path to do it
  1142. 33:18that way while making money along the
  1143. 33:20way, well, that's what interested us,
  1144. 33:22and we're going to continue down that
  1145. 33:23path. Now, when you're making money,
  1146. 33:26you don't raise from VCs anymore. But
  1147. 33:27there you go. And another point,
  1148. 33:29perhaps, is that I don't know if we're
  1149. 33:31in an ecosystem where raising 10
  1150. 33:33million euros would really help us grow
  1151. 33:35. You see, when you sell a product or a
  1152. 33:38tool, you go all-in on communication,
  1153. 33:41you hire a bunch of salespeople to make
  1154. 33:43calls all day, and you grow fast. For
  1155. 33:46us, it's so much of an execution-based
  1156. 33:48business—where you need strong
  1157. 33:50execution to be effective—that if you
  1158. 33:51gave me 5 million euros tomorrow, I
  1159. 33:53wouldn't know what to do with it. If I
  1160. 33:55put 5 million into ads, I wouldn't be
  1161. 33:57able to handle the commercial follow-up
  1162. 33:58, have the right trainers, or monitor
  1163. 34:00the quality of the training. I think
  1164. 34:02it's very hard, much harder than with a
  1165. 34:04product where things scale easily. So,
  1166. 34:05all things considered, we prefer to go
  1167. 34:07a bit more step-by-step. Besides, our
  1168. 34:09competitors are doing much the same. We
  1169. 34:11haven't seen any training providers
  1170. 34:12raise huge amounts of money like that;
  1171. 34:14they all have this same problem of
  1172. 34:15having to execute behind the scenes. So
  1173. 34:17, it's completely tough. You mentioned
  1174. 34:20several clients, notably Vinci, and
  1175. 34:21Clarin as well, who are therefore not
  1176. 34:23in real estate. At what point did you
  1177. 34:26make the switch from focus to a gradual
  1178. 34:28diversification of your client typology
  1179. 34:30, including B2B and B2C? I'm asking you
  1180. 34:33this because it's a real issue for many
  1181. 34:35entrepreneurs: should I stay focused on
  1182. 34:37a very specific niche, or allow myself
  1183. 34:39to open up to a wider niche to
  1184. 34:41potentially address a larger pool? It's
  1185. 34:44a super interesting question, and I
  1186. 34:45like to describe it this way. Actually,
  1187. 34:47we stayed niche in real estate for 6
  1188. 34:49months, I would say. Sure, you had
  1189. 34:51clients here and there coming from
  1190. 34:52elsewhere, but we focused on real
  1191. 34:54estate for 6 to 8 months. The point of
  1192. 34:57that, I think it's very important to do
  1193. 34:58it that way, because when you focus on
  1194. 35:00a niche, you become an expert in that
  1195. 35:02niche, and all your prospects know
  1196. 35:03you're an expert, so they sign with you
  1197. 35:05faster. So, we were real estate experts
  1198. 35:08: developers, social landlords, real
  1199. 35:10estate agents, notaries. Those were our
  1200. 35:12four main areas originally, so very
  1201. 35:13quickly all the notaries want to work
  1202. 35:15with you, all the social landlords want
  1203. 35:16to work with you, and all the
  1204. 35:17developers, and so on. So you grow much
  1205. 35:19faster than our competitors who were
  1206. 35:21generalists. And since you grow fast, I
  1207. 35:23always use this example: we reached 35
  1208. 35:25to 40 employees when our competitors
  1209. 35:27were still at 6, 7, 8, or 9, and
  1210. 35:28honestly, I don't know if we're better,
  1211. 35:30I'm not sure. They are very good; I
  1212. 35:33know a few who are very good. Do we
  1213. 35:34have a better brand than them? I'm not
  1214. 35:36sure either. Well, you can have a much
  1215. 35:38better product than the other, so you
  1216. 35:40grow faster. But we are good, I think,
  1217. 35:42but I think they are very good too, so
  1218. 35:44that's not what made us go from 35 to
  1219. 35:4670. It's that we were very niched and
  1220. 35:47said, "We won't diversify, we will
  1221. 35:49execute what we do well and grow," but
  1222. 35:51then you reach that moment where you're
  1223. 35:5230 or 35, and the market starts to
  1224. 35:54become a bit more mature. So then you
  1225. 35:56start having larger companies
  1226. 35:58interested in AI that ask themselves
  1227. 35:59who they should have as a partner. And
  1228. 36:01there are two choices on the market:
  1229. 36:02companies with 7 people or companies
  1230. 36:03with 35. And so, what did we get in our
  1231. 36:06first clients? Agribusiness companies
  1232. 36:07started coming to us and asking, "Well,
  1233. 36:09don't you do agribusiness?" I would say
  1234. 36:10, "Yes, we don't do agribusiness."
  1235. 36:11However, we have 40 people at Misteria,
  1236. 36:13while our competitors have 6. So if we
  1237. 36:14work with you, you're a bit large, I
  1238. 36:16have the capacity to provide you with
  1239. 36:1730 days of training a month starting
  1240. 36:18tomorrow. Our competitors can't do that
  1241. 36:20. And so then they say, "Oh, yeah,
  1242. 36:21that's not bad." So you say that to one
  1243. 36:23agribusiness firm, then two, then three
  1244. 36:25, and boom, you open a new vertical,
  1245. 36:26and in fact, you grow that vertical
  1246. 36:27quickly and soon you hit 60 employees
  1247. 36:29while your competitors are at 12. And
  1248. 36:32then you have Vinci or Clarins arrive
  1249. 36:33and say, "Well, I need to train 600
  1250. 36:35people all over the world at once, but
  1251. 36:36I need a huge volume of trainers." I
  1252. 36:38say, "Well, we have 60 people." So, in
  1253. 36:40fact, your niche allows you to grow
  1254. 36:41very quickly, to reach a sort of
  1255. 36:43critical mass which then allows you to
  1256. 36:44branch out. And for us, that's been the
  1257. 36:47main strategy, which we're still doing
  1258. 36:49in real estate, which is still 60%of
  1259. 36:51our business.
  1260. 36:53And we aren't going to let that go,
  1261. 36:54because we like it, it's a huge market,
  1262. 36:55and we're good at it, but also because
  1263. 36:56it allows us to grow, grow, grow our
  1264. 36:58company and become more structured. You
  1265. 37:00see, we wouldn't have had a CFO or an
  1266. 37:02HR director if we hadn't grown so fast.
  1267. 37:04But having a CFO, an HR director, and
  1268. 37:06so on means that new prospects tell us,
  1269. 37:07"Oh yes, you are a serious company, you
  1270. 37:09are structured." So even if I'm in
  1271. 37:10aerospace and it's not your specialty,
  1272. 37:12I know you'll assign someone to me who
  1273. 37:14knows aerospace well and who will
  1274. 37:16follow me from A to Z and so on. So
  1275. 37:18that's a big piece of advice that I
  1276. 37:19give. I would say be niched at the
  1277. 37:21beginning, grow to be bigger than your
  1278. 37:23competitors because you are niched, and
  1279. 37:25then you have all the weapons to, well,
  1280. 37:27I don't know, to go and attack other
  1281. 37:29markets. But I think you can see it in
  1282. 37:31plenty of other areas. It has to be
  1283. 37:33like that, you know? You you build a
  1284. 37:35team a bit on a on a specific vertical
  1285. 37:37and in fact you reach a critical size
  1286. 37:38that allows you to go elsewhere, you
  1287. 37:40know. So that that was the strategy we
  1288. 37:42used and that we are going to continue
  1289. 37:43to use even if today, for that matter,
  1290. 37:45we no longer present ourselves at all
  1291. 37:46as a specialized real estate firm. We
  1292. 37:48have specialized verticals. However,
  1293. 37:49Mystoria is very generalist and we do
  1294. 37:51everything there. And actually, that
  1295. 37:53will allow me to ask you a question
  1296. 37:54about the acquisition side. Because
  1297. 37:56since it's very diversified, both in
  1298. 37:59the range of offers and in the target
  1299. 38:01panel, because you won't prospect or
  1300. 38:03close a deal the same way with Vinci or
  1301. 38:05Clarence as with an individual or a
  1302. 38:07business owner, a fishmonger who wants
  1303. 38:09to train in AI. How do you structure
  1304. 38:13your offers? What exactly do you offer?
  1305. 38:16Do you offer it directly, just
  1306. 38:18front-end on the site, and you go, you
  1307. 38:19choose, you click, and you start an
  1308. 38:21adventure? Or does it happen
  1309. 38:23differently?
  1310. 38:24So first of all, our company structure
  1311. 38:25is in three parts. It's B2B consulting,
  1312. 38:27B2B training, B2C training. Those are
  1313. 38:30really the three products of Mysteria.
  1314. 38:31And how do we sell it? A lot of
  1315. 38:33webinars, a lot a lot of webinars. I do
  1316. 38:35webinars every week but with zero paid
  1317. 38:37acquisition. Actually, our newsletters
  1318. 38:39—yes, newsletters, because we have
  1319. 38:41Mysteria, we have the one for real
  1320. 38:42estate agents, and we have the one from
  1321. 38:44my partner's company, which by the way
  1322. 38:45has now been acquired by Mysteria,
  1323. 38:47which also reaches 30,000 people in the
  1324. 38:49real estate sector. So, we have our
  1325. 38:51three newsletters that draw about
  1326. 38:53100,000 people that we invite to
  1327. 38:54webinars very regularly, and in fact,
  1328. 38:56you realize that people send their
  1329. 38:57friends, or sometimes we do little
  1330. 38:59promos, we tell them: "Well, if you
  1331. 39:00invite three people who don't know
  1332. 39:02Mysteria, at the end we give something
  1333. 39:03away," etc. So, in fact, you pull in
  1334. 39:05lots of people who don't even read the
  1335. 39:07newsletter. So we do a lot of free
  1336. 39:09webinars, we present our strategy, and
  1337. 39:11we're going to keep that up—we give
  1338. 39:13you free info. So we do an hour of
  1339. 39:15content on AI, and at the end we tell
  1340. 39:17you if you want to be accompanied, you
  1341. 39:18can ask us, and we'll be happy to
  1342. 39:20accompany you. That's the big growth
  1343. 39:21strategy. First, second growth strategy
  1344. 39:23is to be very niche by profession. Even
  1345. 39:26now. Today, we have 6 or 7 professions
  1346. 39:28where we are still very strong. And
  1347. 39:29niche by trade means, for us, attending
  1348. 39:31the trade shows that matter. Misteria,
  1349. 39:32for example, doesn't go to AI trade
  1350. 39:34shows. All my competitors do AI trade
  1351. 39:36shows. I, however, go to the public
  1352. 39:37housing trade show. At the public
  1353. 39:38housing show, everyone says, "But what
  1354. 39:39are you doing here?" Well, actually,
  1355. 39:41there are 600 to 1,000 public housing
  1356. 39:43firms in France with 100 to 1,000
  1357. 39:44employees that need AI just like
  1358. 39:46everyone else. So when I go there, I am
  1359. 39:48the AI expert for the public housing
  1360. 39:50sector. Same for notaries: we attend
  1361. 39:52the Notaries' show, we do real estate
  1362. 39:54developers 'shows, we do the Mayors'
  1363. 39:55show for local authorities, but we
  1364. 39:57don't do Big Data shows, we don't do
  1365. 39:58Vivatech—though we should, and we
  1366. 40:00probably will one day. But given the
  1367. 40:01choice today, we invest much more in
  1368. 40:03these other shows. So we see a lot of
  1369. 40:06growth from acquisition happening this
  1370. 40:08way. Trade shows, webinars, word of
  1371. 40:11mouth—it’s nearly half of our
  1372. 40:12acquisition. One out of every two
  1373. 40:14people says, "My friend works with you,
  1374. 40:16he told me to reach out, so I'm working
  1375. 40:17with you."
  1376. 40:18Because the ecosystem is still small,
  1377. 40:20we have that advantage. It feels like
  1378. 40:22everyone is doing AI, but in reality,
  1379. 40:24companies are just starting out, so
  1380. 40:25naturally, whenever they have questions
  1381. 40:26, they look around and ask, "Do you
  1382. 40:28know someone who helped you?" And
  1383. 40:29Misteria comes up a lot. SEO as well.
  1384. 40:32That's starting to pick up; we're
  1385. 40:34beginning to rank well. Those are our
  1386. 40:36main strategies. But the acquisition
  1387. 40:38cost is virtually zero. We launched our
  1388. 40:40first Google ad at the beginning of
  1389. 40:41September. It's been 9 days of running
  1390. 40:44Google ads for small amounts.
  1391. 40:46We also made our first paid acquisition
  1392. 40:48for a webinar this month. So before
  1393. 40:50that, we had zero paid acquisition.
  1394. 40:52I’m speaking primarily about B2B here
  1395. 40:53. In B2C, we started doing paid
  1396. 40:56acquisition a little earlier, but it's
  1397. 40:58still quite small for us. Those are the
  1398. 41:00main channels. And well, LinkedIn—I
  1399. 41:02have about 27,000 followers, I think;
  1400. 41:04I've been posting every day for 2 years
  1401. 41:06, and it still drives a lot of traffic.
  1402. 41:09So, you see, we have a real organic
  1403. 41:11acquisition strategy for now. But trade
  1404. 41:13shows, you still have to pay for them,
  1405. 41:14they aren't free, even if the amounts
  1406. 41:15are small. So, trade shows plus organic
  1407. 41:17, that's the bread and butter for us
  1408. 41:19today. And often, and here I'm talking
  1409. 41:21more B2B than B2C, the person arrives,
  1410. 41:23"I want training," well, that's perfect
  1411. 41:25, we're CPF eligible, so we train a lot
  1412. 41:27of them using CPF. We're hitting some
  1413. 41:28pretty crazy numbers right now. Which
  1414. 41:30is great, the CPF allows for that.
  1415. 41:31We're pushing the government to steer
  1416. 41:33CPF funds more toward AI because it's
  1417. 41:35really important for the country. So
  1418. 41:37B2C is pretty standard. They watch the
  1419. 41:39webinar, they ask, they use their CPF,
  1420. 41:40and they train with us. On the B2B side
  1421. 41:42, the request often comes in saying, "I
  1422. 41:44have no idea where to start with this
  1423. 41:46AI topic." Companies tell me that. They
  1424. 41:48see us in a webinar, at a trade show,
  1425. 41:49or where I mentioned, and they ask, "
  1426. 41:51Where do I start to get a little bit
  1427. 41:53going?" So, what we push very often is
  1428. 41:55a quick audit to start. We say, "We'll
  1429. 41:58audit you, see what you need, create a
  1430. 41:59roadmap, and then roll out training,
  1431. 42:00some custom development, and so on."
  1432. 42:02But today, that's mainly it. You have
  1433. 42:05very few companies that come to you
  1434. 42:06saying, "Okay, I have an AI training
  1435. 42:07plan in mind, I'm looking for a
  1436. 42:08provider to do this, that, and the
  1437. 42:09other." That's quite rare. Most are
  1438. 42:11like, "Guys, we need help, we don't
  1439. 42:12know how to approach this thing."
  1440. 42:14Yeah, we'll get to that. But before
  1441. 42:16that, I want to jump on a point you
  1442. 42:17raised that's super interesting, which
  1443. 42:19is the trade show aspect. Uh, it's true
  1444. 42:21you might think, "Well, Vivatech, the
  1445. 42:23AI show, etcetera," and so on. However,
  1446. 42:26this strategy is better than being a
  1447. 42:28shrimp in the ocean; you become a whale
  1448. 42:30in a pond. It's a bit like blue ocean,
  1449. 42:32red ocean. Uh, you show up there,
  1450. 42:34you're the only AI player, talking
  1451. 42:35about the HLM trade show or other shows
  1452. 42:36you mentioned. It's pretty crazy, and I
  1453. 42:39think many should take inspiration from
  1454. 42:40this type of strategy if, in fact,
  1455. 42:42you're a service provider who can help
  1456. 42:43one or more business sectors rather
  1457. 42:45than just going to your own trade show.
  1458. 42:47Clearly. Clearly. And people really
  1459. 42:49like that too, you know. Me, whether
  1460. 42:50it's at trade shows or even outside of
  1461. 42:52trade shows, I spend all my time out in
  1462. 42:53the field. Actually, I'm not in the
  1463. 42:55office enough; that's what I blame
  1464. 42:56myself for. I would like to stay there
  1465. 42:58more. You see, yesterday I was in Niort
  1466. 43:00, tomorrow I'm in Toulouse. The day
  1467. 43:02after tomorrow, I'm in Toulouse. I'm
  1468. 43:03moving around all the time. Not to have
  1469. 43:05fun, but to see clients. Actually, they
  1470. 43:06really appreciate it and I find that
  1471. 43:07normal. They are tired of the little
  1472. 43:09Parisian who explains life from his
  1473. 43:11Paris office, telling people what to do
  1474. 43:13. No, I tell them I'm not going to do
  1475. 43:14that, I'm coming to see you in person
  1476. 43:16and we'll talk. And actually, when you
  1477. 43:17do that, people are so much nicer, they
  1478. 43:18put you in touch with a lot of people.
  1479. 43:20That is our real strategy. We really
  1480. 43:22try to avoid the thing where I'm a bit
  1481. 43:24of an AI star in Paris, which I'm not,
  1482. 43:26but I could very well act like it, you
  1483. 43:28see. And no, we try to avoid that
  1484. 43:29absolutely. We go into the field, we go
  1485. 43:31to trade shows, we meet people, we get
  1486. 43:33to know them well, and we don't
  1487. 43:34hesitate, however, to tell them: "Look,
  1488. 43:35things are going very well between us."
  1489. 43:37You're going to say: "I'm, well, I tell
  1490. 43:38the person, hey, if you can put me in
  1491. 43:39touch with one of your colleagues, that
  1492. 43:41would be great, we're looking to expand
  1493. 43:42here a bit," and everyone does it. When
  1494. 43:43things are going well with a client and
  1495. 43:45you are nice, everyone accepts you. And
  1496. 43:47the conversion rate is 100%. With a
  1497. 43:50personal connection, the conversion
  1498. 43:51rate really is 100%. So we rely on that
  1499. 43:53a lot and I think that, indeed, that
  1500. 43:54might be a mistake of the French Tech.
  1501. 43:56I saw recently, I think it's Payfit or
  1502. 43:58Qonto, I don't remember, maybe Payfit,
  1503. 44:00the CEO who said the company is almost
  1504. 44:0210 years old, created in 2015-2016 or
  1505. 44:04is about to be 10. And the CEO says, a
  1506. 44:06year ago, "Okay, let's go, for the
  1507. 44:08first time, I'm going to do a tour of
  1508. 44:09France to visit my clients." And I said
  1509. 44:11to myself, but what have you been doing
  1510. 44:12for 9 years? You see, obviously, go see
  1511. 44:14the clients. Now, it's easier to say
  1512. 44:16that from my sofa than anything else,
  1513. 44:17right, I'm not criticizing either. But
  1514. 44:18it's quite interesting and, besides, I
  1515. 44:20think in his place, I would have done
  1516. 44:21the same. I didn't really go see my
  1517. 44:23clients for 10 years because the thing
  1518. 44:25was, we all raise a lot of money, we
  1519. 44:26spend heavily on ads, and we close.
  1520. 44:29It's fine, but I feel like there's a
  1521. 44:30bit of a shift today where people want
  1522. 44:32to see you, and we have plenty of
  1523. 44:33competitors. So if companies don't see
  1524. 44:35me, I'm no better at ads, I'm no better
  1525. 44:37at whatever else. So they need to see
  1526. 44:39me so they can trust me. One, it's
  1527. 44:41better for selling, and two, for
  1528. 44:43improving your services. When you're
  1529. 44:45selling services. I tell them, I ask
  1530. 44:46them concretely, what do your employees
  1531. 44:47actually do? You know, what's their
  1532. 44:49daily problem? I go into the teams, I
  1533. 44:51go to their offices very often, and I
  1534. 44:53actually dig up gold mines for my
  1535. 44:54consultants when I bring them tons of
  1536. 44:56info. And so even they have that rule.
  1537. 44:58I tell them, when a client wants to see
  1538. 45:00you in Brest, you go to Brest. I don't
  1539. 45:02care if we lose time, we go see them in
  1540. 45:03Brest. And you have a trade show, a
  1541. 45:06notary fair, you know, it's not the
  1542. 45:07sexy thing in Paris or whatever. It's
  1543. 45:10sometimes in lesser-known cities,
  1544. 45:11sometimes it's not very big. Well, we
  1545. 45:13go, we don't care. We see our clients,
  1546. 45:14they're happy to see us, we're happy to
  1547. 45:16see them, and I find it's a sales
  1548. 45:17strategy that also works very, very
  1549. 45:18well.
  1550. 45:19And it's one that's sincere, you see.
  1551. 45:21It's not just ADS anymore, it's the ADS
  1552. 45:23sales strategy. You might tell yourself
  1553. 45:25you're trying to reach people's
  1554. 45:26emotions and so on, and that's the game
  1555. 45:27. Why not? But here, when you go to see
  1556. 45:29them, it's sincere, you talk properly
  1557. 45:31with them, you say, "Look, I can do
  1558. 45:32this for you," does that work or not?
  1559. 45:33We make a deal. It's more of a, let's
  1560. 45:35say, grassroots business approach. I
  1561. 45:37like it completely. You also say that
  1562. 45:40you're in a very competitive market.
  1563. 45:42Yeah, which is true. Today, anyone,
  1564. 45:44like you said, could just sit on their
  1565. 45:47couch and criticize. But ultimately,
  1566. 45:49today, anyone on their couch, in their
  1567. 45:51room, or anywhere else with a computer
  1568. 45:52and an internet connection—wherever
  1569. 45:54they are in the world, regardless of
  1570. 45:56their age or background—can create a
  1571. 45:58website and start an AI agency with AI
  1572. 45:59agents or just a few people. So indeed,
  1573. 46:01there are an enormous number of players
  1574. 46:02in the market. You also mentioned
  1575. 46:04LinkedIn, where there are tons of
  1576. 46:06experts or pseudo-experts present. The
  1577. 46:08idea isn't to criticize or point
  1578. 46:10fingers, but rather: beyond this
  1579. 46:12strategy you mentioned, which is
  1580. 46:14hyper-interesting—the ground-level,
  1581. 46:16down-to-earth side, the contact with
  1582. 46:19the client—how do you differentiate
  1583. 46:21yourself in an extremely competitive
  1584. 46:23market with low barriers to entry,
  1585. 46:25relatively low margins, and
  1586. 46:27extraordinary growth?
  1587. 46:29It’s tough, it’s tough. So, the big
  1588. 46:31pillars: if I had to start over in
  1589. 46:33another ecosystem, I’d tell myself:
  1590. 46:35you need to see people because in
  1591. 46:37consulting, they hire you for a project
  1592. 46:39because they know you, not just because
  1593. 46:41you're the best.
  1594. 46:42So you have to see people,
  1595. 46:44that’s the first thing. And you have
  1596. 46:47a scale issue in agencies, regardless.
  1597. 46:50Agencies with two, three, or even ten
  1598. 46:53people can handle B2B for small
  1599. 46:55companies. But when you’re dealing
  1600. 46:57with a company of a thousand people, I
  1601. 46:58don't know if they're right or not, but
  1602. 46:59I think so. They want to have a very
  1603. 47:01structured company in front of them, a
  1604. 47:02very structured provider. So, there is
  1605. 47:04still a race to scale, which we are
  1606. 47:05currently winning against some
  1607. 47:07competitors. So that really helps us.
  1608. 47:10Our very pragmatic approach—I'll
  1609. 47:13speak more about AI—is very popular.
  1610. 47:16I have plenty of competitors saying, "
  1611. 47:18We’re going to create agents for you,
  1612. 47:19we’ll automate 70%of your business."
  1613. 47:21I tell them, we’ll set up ChatGPT,
  1614. 47:23and you’ll all save an hour a day.
  1615. 47:24Even though we obviously do more than
  1616. 47:25just that. That’s more of the promise
  1617. 47:27, and it’s much more down-to-earth.
  1618. 47:29And I tell them if we start now, in a
  1619. 47:30month, your 1,000 employees will be
  1620. 47:31trained. Now, that’s starting to get
  1621. 47:32a bit concrete. So, we try to be
  1622. 47:34concrete and maintain a pragmatic
  1623. 47:35approach. That’s more in relation to
  1624. 47:38our big competitors, because compared
  1625. 47:40to the small ones, I play the size and
  1626. 47:42support card. I say, look, if you train
  1627. 47:43with a freelancer, they won't follow
  1628. 47:45you for the next two years. For us,
  1629. 47:47you’re a 100-person company. You're
  1630. 47:49hesitating between us, with our 100
  1631. 47:50employees, and a 10-person firm. Well,
  1632. 47:53since we're 100, we'll handle your
  1633. 47:54training, your consulting, and give you
  1634. 47:55a packaged offer. And since people find
  1635. 47:57it hard to imagine the business needs
  1636. 47:59of tomorrow regarding AI, we tell them
  1637. 48:01we'll be on the front line no matter
  1638. 48:03what to address them. If tomorrow we
  1639. 48:06feel you actually need to do some
  1640. 48:07AI-driven SEO, or need some advanced
  1641. 48:09training or a bit of consulting here
  1642. 48:10and there, we know you so well and have
  1643. 48:12already supported you on so many things
  1644. 48:14that we'll be able to support you on
  1645. 48:15that too, because we're big. So against
  1646. 48:18smaller competitors, so to speak, I
  1647. 48:19play that card, and against the very
  1648. 48:21big ones like OnePoint, Artefact, and
  1649. 48:22the like, we also try to hold our own,
  1650. 48:24and there we play the pragmatism card.
  1651. 48:26We say, actually, that BCG strategic
  1652. 48:27mission at € 500,000 to tell you that
  1653. 48:29AI will help you write meeting minutes
  1654. 48:31...At Nouvel, it costs € 20,000 and
  1655. 48:33we provide something very pragmatic.
  1656. 48:35That was really the focus from the
  1657. 48:37start, deciding we would do practical,
  1658. 48:39hands-on missions. I've seen many
  1659. 48:41missions from big consulting firms pass
  1660. 48:43through our clients 'offices. First off
  1661. 48:45, they sell it for a fortune because
  1662. 48:46they don't get out of bed for €
  1663. 48:4710,000. So they sell it for a fortune
  1664. 48:49and it's theoretical: there will be
  1665. 48:51productivity gains here, there will be
  1666. 48:52a reorganization of the company there,
  1667. 48:54and so on. And my clients tell me, "I
  1668. 48:56received this from X consulting firm,
  1669. 48:58what am I supposed to do with it?"
  1670. 49:00Whereas I, at the end of the mission,
  1671. 49:01tell them, "Look, we've audited your
  1672. 49:03development team." For your development
  1673. 49:04team, we found 30 use cases. For each
  1674. 49:06use case, we made a demo video of how
  1675. 49:08they could implement it. We identified
  1676. 49:09the tool or tools that could be used.
  1677. 49:11How much it would cost you in licenses
  1678. 49:12if you took one, if you took five. By
  1679. 49:14the way, we also told you how many
  1680. 49:15licenses you actually need. And there,
  1681. 49:17actually, at the end of it, the company
  1682. 49:18has something super concrete. So that's
  1683. 49:20the card we play against the big
  1684. 49:21players. It's when we're dealing with
  1685. 49:23big corporate clients, where it's
  1686. 49:24either us, or a OnePoint, or an
  1687. 49:26Artefact, or an EY facing them, that we
  1688. 49:27tell them: "Well, with us, you won't
  1689. 49:29get the Artefact or OnePoint stamp,
  1690. 49:30you'll get a MisterIA stamp that's
  1691. 49:32worth much less on the market, but
  1692. 49:33you'll get something extremely concrete
  1693. 49:34." So that still appeals to many, to
  1694. 49:37many big companies. That's how we try
  1695. 49:39to differentiate ourselves a bit
  1696. 49:40between the two, between the very small
  1697. 49:42freelancer training companies and the
  1698. 49:43very large firms doing massive
  1699. 49:44development plans.
  1700. 49:45Very clear. Earlier, you mentioned that
  1701. 49:48many companies don't really understand
  1702. 49:50how to implement AI in their workplace,
  1703. 49:53and you said, "Well, click here, get an
  1704. 49:55audit, and we'll explain it to you."
  1705. 49:58Let's imagine that the company watching
  1706. 50:00this episode doesn't have the immediate
  1707. 50:02capacity to call us or get an audit.
  1708. 50:05What would be the first major general
  1709. 50:07pieces of advice you'd share with any
  1710. 50:09company, say with a turnover between
  1711. 50:11500k and 100 million euros, that would
  1712. 50:13like to implement AI in their daily
  1713. 50:15routine? Well, the first reflex for
  1714. 50:18those who don't necessarily have the
  1715. 50:20budget for consulting is to go ahead
  1716. 50:21and get ChatGPT licenses for your
  1717. 50:22employees. Be transparent about AI
  1718. 50:25usage, meaning authorize your employees
  1719. 50:27to use it, but understand what they're
  1720. 50:29doing with it. I think that's quite
  1721. 50:31important. There are still issues with
  1722. 50:33data, etc. So, be careful. So, get them
  1723. 50:35licenses, and take a look at what
  1724. 50:36they're doing. Set up a small AI
  1725. 50:38committee—many are already doing it
  1726. 50:40—with 4 or 5 people in the company
  1727. 50:42who are interested in the topic, and
  1728. 50:43meet once a month for a morning to
  1729. 50:45reflect on what can be implemented, and
  1730. 50:47so on. You don't need to do big things,
  1731. 50:49but at least small use cases will
  1732. 50:50emerge. That works well. Large
  1733. 50:52corporations do this a lot. For example
  1734. 50:54, we have an AI ambassador support
  1735. 50:55service; many large groups tell us, "
  1736. 50:57I've created 50 AI ambassadors, and
  1737. 50:58that's their role, to do what I just
  1738. 51:00described." However, they are supported
  1739. 51:01by MisterIA because the stakes are high
  1740. 51:03, and so on. But that works well. So,
  1741. 51:05set up a small AI committee. After that
  1742. 51:07, consume content on YouTube, but
  1743. 51:09there's so much content, it's hard.
  1744. 51:11Well, it's the classic advice, but it's
  1745. 51:14hard to really know what to do just by
  1746. 51:16consuming content on YouTube. Um,
  1747. 51:20otherwise, you have coaching programs
  1748. 51:23that we know well, which work quite
  1749. 51:24effectively. I have quite a few CEOs
  1750. 51:27who tell me they don't want to spend 10
  1751. 51:29, 20, or 30k on an audit. However, they
  1752. 51:31do want to be guided for 3 or 4 months
  1753. 51:33to really understand the subject. So we
  1754. 51:35do a lot of coaching like that, where
  1755. 51:36you're one-on-one with your coach and
  1756. 51:38you just get things moving. So these
  1757. 51:40are things that work well. So, overall,
  1758. 51:43that would be the advice I'd give to
  1759. 51:44people who don't have too much budget:
  1760. 51:46start with a short training day, which
  1761. 51:48isn't very expensive and already allows
  1762. 51:49you to cover a lot of ground. The idea
  1763. 51:52of a committee is really interesting.
  1764. 51:53It effectively allows you to empower
  1765. 51:55one, two, or three people in your
  1766. 51:57company and tell them, "Stay informed
  1767. 51:59on the subject once a month." You meet,
  1768. 52:01and based on the daily use cases we
  1769. 52:03have, we see how we can gradually
  1770. 52:04implement it. In my experience, it
  1771. 52:06might seem a bit abstract to some who
  1772. 52:09haven't adopted AI yet. As crazy as it
  1773. 52:11might seem to you or me, since we are
  1774. 52:13immersed in AI every day and we talk
  1775. 52:15about it a lot on entrepreneurance.com.
  1776. 52:18Can you explain these famous concrete
  1777. 52:20use cases where you say, "You're going
  1778. 52:22to save an hour a week per employee,
  1779. 52:24maybe even an hour a day per employee
  1780. 52:26at certain times"? Sometimes it's 15%,
  1781. 52:3020%, 30%more productivity, not to
  1782. 52:32mention the costs you save because,
  1783. 52:34ultimately, ChatGPT is only 20 dollars
  1784. 52:37a month, which is nothing, but you pair
  1785. 52:39it with other tools. You mentioned
  1786. 52:43Jasper, just to name one, but there are
  1787. 52:45so many others that we can use. We'll
  1788. 52:47talk later about all the tools and how
  1789. 52:49not to fall into the frenzy of using
  1790. 52:511,500 of them to no avail. But
  1791. 52:53basically, in concrete terms, how do we
  1792. 52:55manage to save even just one hour a
  1793. 52:57week for each employee?
  1794. 52:58Yeah. Well, in terms of use cases that
  1795. 53:01work well, for the entire sales side of
  1796. 53:02companies, there's a lot to be done.
  1797. 53:05All our sales proposals are done in one
  1798. 53:07click and they are personalized, they
  1799. 53:09are interactive, they are beautiful,
  1800. 53:11they are detailed, and that is
  1801. 53:12impossible to do without AI. So, a bit
  1802. 53:15like I was telling you, I take my
  1803. 53:16outreach, I send it to my to my Claude
  1804. 53:17—I use Claude a lot—which creates a
  1805. 53:19sales proposal for me in the form of a
  1806. 53:20website. You send that to the client,
  1807. 53:22nobody else is sending them that. There
  1808. 53:24is a huge "wow" factor, and they have
  1809. 53:25all the information. So, just with that
  1810. 53:27, you’ve saved a lot of time. Meeting
  1811. 53:29minutes. So, I was talking about video
  1812. 53:30calls, but even in-person meetings with
  1813. 53:31a prospect. Me, as soon as I finish a
  1814. 53:33physical meeting, I leave, take out my
  1815. 53:35phone, and I dictate a summary. And for
  1816. 53:37the summary, I tell it, "Here is
  1817. 53:38everything that happened, write a recap
  1818. 53:39email and a WhatsApp recap for my
  1819. 53:41partner."
  1820. 53:41It's simple, but it's done in a minute
  1821. 53:43and a lot of information gets shared.
  1822. 53:45So that saves a lot of time in the long
  1823. 53:46run. Then, as for all your analyses, it
  1824. 53:49really depends on the profession, you
  1825. 53:51know. The use cases are all very
  1826. 53:53job-specific, but for all your document
  1827. 53:55analysis, any document I want to
  1828. 53:57analyze, I send it to an AI—to Claude
  1829. 53:59or ChatGPT—and I tell it, "Summarize
  1830. 54:00this for me, analyze this, compare
  1831. 54:02these two docs." You see, if you have
  1832. 54:04if you work with contractors, send five
  1833. 54:06quotes from contractors to Claude and
  1834. 54:08say, "Make me a summary table." Those
  1835. 54:10are things that take up a lot of time.
  1836. 54:12Uh, sometimes all the email management,
  1837. 54:13for me 100%of emails are answered with
  1838. 54:15AI, 100%. I have a ChatGPT shortcut, I
  1839. 54:18talk to it, and actually I see the
  1840. 54:19email, it sees it too. I I click
  1841. 54:21Option-Space, and it brings up a tiny
  1842. 54:24ChatGPT window. I click the microphone,
  1843. 54:26I say, "Okay, reply to Alex to tell him
  1844. 54:28I'll be there tomorrow, that I'll
  1845. 54:29arrive at 11:00 AM, and that and that I
  1846. 54:31want to have some tea." And in fact, it
  1847. 54:33writes the email right before my eyes,
  1848. 54:35I copy, I paste, I send. It saves me a
  1849. 54:36crazy amount of time. Just that alone
  1850. 54:37is definitely 40 minutes a day. So you
  1851. 54:40have a few cases like that, email,
  1852. 54:41sales teams, well finance teams, you
  1853. 54:43have a bit fewer actually, you see, I
  1854. 54:45have fewer use cases in mind for the
  1855. 54:47finance side. Uh HR teams, every
  1856. 54:50company has HR teams, preparing your
  1857. 54:51interviews, writing your job
  1858. 54:53descriptions, comparing a resume with a
  1859. 54:55job description. I send both and I say,
  1860. 54:57I send the job description and I say,
  1861. 54:59out of these five resumes, which ones
  1862. 55:00best match the job description and why?
  1863. 55:02Uh it does it for us all the time, you
  1864. 55:03see. It allows us to have a single HR
  1865. 55:05manager who does daily interviews,
  1866. 55:07which she could never do otherwise. But
  1867. 55:09actually, she uses AI that helps her do
  1868. 55:11all of that. She, for example, gives
  1869. 55:12the job description and resume to the
  1870. 55:13AI, and tells it, "Prepare the 15
  1871. 55:15interview questions I need to ask."
  1872. 55:16H
  1873. 55:17It’s unbeatable. You have plenty of
  1874. 55:19little use cases like that. And in all
  1875. 55:20of that, even just picking one or two
  1876. 55:22and doing them well, you’ll save an
  1877. 55:23hour a day. 100%per day. That’s
  1878. 55:25really the
  1879. 55:26equivalent. And come on, let’s
  1880. 55:28include the whole communication side
  1881. 55:29which works very well. And
  1882. 55:30communication is important. Careful, "
  1883. 55:33I’ll write my LinkedIn post with AI,"
  1884. 55:35it’s good, it can work, but you
  1885. 55:36always have to go further in the use
  1886. 55:38case. I often say, for example,
  1887. 55:40regarding the LinkedIn post case,
  1888. 55:42it’s good to write it with AI, but
  1889. 55:44instead tell your ChatGPT: "I want to
  1890. 55:45write this LinkedIn post, I have this
  1891. 55:47idea in mind, help me turn it into a
  1892. 55:49series of 10 LinkedIn posts.""I want my
  1893. 55:5110 LinkedIn posts to have a common
  1894. 55:53thread.""I want three of my 10 LinkedIn
  1895. 55:55posts to be aimed at my prospects,
  1896. 55:56three to attract the curiosity of my
  1897. 55:58clients, and three to attract the
  1898. 56:00curiosity of my future employees for
  1899. 56:02employer branding.""And in these 10
  1900. 56:04LinkedIn posts, I want you to give me
  1901. 56:05little examples each time, and in each
  1902. 56:07post, tease the one for the next day a
  1903. 56:08little bit." You see, you can play with
  1904. 56:10it like that and you’ll get it to
  1905. 56:12create something much more structured,
  1906. 56:13much more powerful, rather than just
  1907. 56:15telling it: "I have a post idea, write
  1908. 56:17it for me." It works. I write
  1909. 56:18everything with it, so it works, but we
  1910. 56:20can take it further. You see, today
  1911. 56:22these use cases are somewhat well-known
  1912. 56:23. However, I don't see anyone using it
  1913. 56:24the way I do, like, "Okay, I want 10 of
  1914. 56:26these." I want each LinkedIn post to
  1915. 56:28link to a Twitter post I'll make the
  1916. 56:30same day as my newsletter; you have it,
  1917. 56:32honestly, we have a newsletter. I
  1918. 56:34always tell him, "Take my newsletter,
  1919. 56:36break it into a bunch of posts,
  1920. 56:37announce something, dig a little deeper
  1921. 56:39with it," and in terms of communication
  1922. 56:41, you'll save a lot of time, too. Why
  1923. 56:43is it that many entrepreneurs lack the
  1924. 56:45desire to dig deeper into the subject
  1925. 56:47or push things further? Because it
  1926. 56:49seems obvious when you explain it, just
  1927. 56:51like saying, "Well, first ask one
  1928. 56:53question at a time, and then dig deeper
  1929. 56:55, develop it," and. And it's true that
  1930. 56:58you have much more advanced tools,
  1931. 57:01while still staying on ChatGPT, just
  1932. 57:03within its "deep search" features or
  1933. 57:05others, where I've already had the case
  1934. 57:07: I push it so much to the limit that
  1935. 57:10it takes me several days to get the
  1936. 57:12full summary out. And I have to follow
  1937. 57:15up every 2 hours so I can say, "So, are
  1938. 57:15you done thinking?" And boom, and boom,
  1939. 57:16and boom, it spits out a document up to
  1940. 57:1780 pages long. That's how I built the
  1941. 57:19vision for entreprenance.com for the
  1942. 57:21next 5 years. For example, 80 pages
  1943. 57:22that were written between ChatGPT and
  1944. 57:24me, where it challenged me, and then I
  1945. 57:26just submitted it to different people
  1946. 57:28around me to see if it was compelling.
  1947. 57:30But that would have never been possible
  1948. 57:31. It would have taken me 4 months to
  1949. 57:33write that document. Completely,
  1950. 57:34absolutely. Why don't people dig deeper
  1951. 57:36? Is it a lack of time? I think that's
  1952. 57:38still the problem. Wait, service firms
  1953. 57:40exist largely because people don't have
  1954. 57:41the time. It's always the same; you're
  1955. 57:43buying two things from them. Either you
  1956. 57:45buy their expertise, or you buy the
  1957. 57:46time of junior consultants that you
  1958. 57:47don't have yourself. I think that's a
  1959. 57:49bit true in every business, and so we
  1960. 57:51obviously bring the expertise a bit
  1961. 57:52because not everyone knows all the
  1962. 57:53tools, and so on. And also, I have
  1963. 57:55plenty of entrepreneurs who tell me if
  1964. 57:57I don't have someone with me, if I
  1965. 57:59don't block out a day-long slot and
  1966. 58:01have someone with me to explain it,
  1967. 58:03I'll never do it. And I understand them
  1968. 58:06, and I’m the first to admit it;
  1969. 58:07sometimes I get coached on things where
  1970. 58:08I could tell myself, "Look, I have
  1971. 58:10enough perspective to go on YouTube,
  1972. 58:11sit down for an hour, look up a bit of
  1973. 58:12GPT, a bit of all that," and learn, but
  1974. 58:13I’ll never do it. So I don't think
  1975. 58:16it's out of ignorance, it’s just that
  1976. 58:17they don't have the time. So you have
  1977. 58:19to find the time, compel yourself,
  1978. 58:20force yourself. Uh, that's the first
  1979. 58:22thing. Secondly, there’s a big bias
  1980. 58:24with AI: you save time very quickly.
  1981. 58:27You save a little time very quickly,
  1982. 58:28and so you often stop there. You do
  1983. 58:30your research on ChatGPT, you write a
  1984. 58:32few emails, and you’re already saving
  1985. 58:34a lot of time. So you often tell
  1986. 58:35yourself, "Well, it's good, I already
  1987. 58:36have a good handle on it." And you
  1988. 58:38never make the effort to say, "Oh wait,
  1989. 58:39actually, maybe I could save not just
  1990. 58:41one hour, but 3 or 4 hours." That is a
  1991. 58:42big, big problem we see in companies,
  1992. 58:44for example, where I have people tell
  1993. 58:45me, "No, it's fine, my employees all
  1994. 58:47use ChatGPT, it's great." Or others who
  1995. 58:49...I insist a lot on saying, "GPT saves
  1996. 58:51you time, yes, but it also provides a
  1997. 58:53lot of quality." And it's really hard
  1998. 58:55to make employees understand that it
  1999. 58:57will improve quality, because the time
  2000. 58:59savings are just too visible, and it's
  2001. 59:01too easy to save time. So that’s also
  2002. 59:02a bias I think is somewhat important,
  2003. 59:04which is telling yourself, "Oh yeah,
  2004. 59:05okay, but it does other things too."
  2005. 59:07Kind of like, I don't know, the Office
  2006. 59:08suite; you see, we could say that with
  2007. 59:10Outlook, we all just write our emails
  2008. 59:12on it. I’m quite certain that we
  2009. 59:14could push productivity much further in
  2010. 59:15Outlook. Actually, it’s so practical
  2011. 59:17for writing emails, and we don't have
  2012. 59:18the time to look into it too much. So
  2013. 59:20we only use that part of the Office
  2014. 59:21suite. Yet, with AI, we could go much
  2015. 59:23further. So that’s, that’s a bit of
  2016. 59:25a hurdle. And by the way, that was the
  2017. 59:27big aspect of ChatGPT. For two years or
  2018. 59:29a year and a half, you couldn't do
  2019. 59:31internet searches on ChatGPT. I used
  2020. 59:33Perplexity a lot, which still works
  2021. 59:34very well.
  2022. 59:36And I'm quite convinced that the fact
  2023. 59:38that ChatGPT can search the internet is
  2024. 59:40a basic feature for them. It's easy to
  2025. 59:43develop. Why didn't they develop it for
  2026. 59:45a year and a half? It’s not because
  2027. 59:46OpenAI isn't good or because they lack
  2028. 59:48engineers; they have everything they
  2029. 59:49need. I think they didn't want us to
  2030. 59:51fall into the trap of turning it into
  2031. 59:52another Google. I think they said: "
  2032. 59:55They're right, we've created a tool so
  2033. 59:57powerful that we can take so far that,
  2034. 59:58if from day one, we let Google search,
  2035. 1:00:00Internet search happen on it, everyone
  2036. 1:00:02will use it like that and stop and
  2037. 1:00:03never dig any deeper." And so for two
  2038. 1:00:06years, in my opinion, they refused to
  2039. 1:00:08add it. And today, now that they’ve
  2040. 1:00:09added it, well, we feel it’s one of
  2041. 1:00:11the most used features, even though,
  2042. 1:00:12heck, sure, you save a bit of time on
  2043. 1:00:14research, but that's not the goal of
  2044. 1:00:15ChatGPT. So that’s a bit of a bias.
  2045. 1:00:17It’s that the first productivity gain
  2046. 1:00:19is so fast, people stop a bit there
  2047. 1:00:21without asking more questions, so we
  2048. 1:00:23forget all the quality it brings. And
  2049. 1:00:25you mentioned the fact of having an HR
  2050. 1:00:28manager who actually uses AI on a daily
  2051. 1:00:30basis to be able to bring quality and
  2052. 1:00:32productivity to all these tasks. Yet
  2053. 1:00:35you are hiring a huge number of people.
  2054. 1:00:37Uh, we mentioned it earlier, now you're
  2055. 1:00:40over 100 people. Uh.
  2056. 1:00:43How many people do you hire on average
  2057. 1:00:45per week? Why am I asking you this
  2058. 1:00:46question? Admittedly, because I imagine
  2059. 1:00:50the figure is interesting due to the
  2060. 1:00:51growth, but also to make those
  2061. 1:00:53listening to us realize how much AI can
  2062. 1:00:55allow one to be much more efficient,
  2063. 1:00:57much more effective, much more
  2064. 1:00:58productive, and on top of that, with
  2065. 1:01:00quality, because ultimately every hire,
  2066. 1:01:02even just for external contributors,
  2067. 1:01:04the freelancers you bring on, well,
  2068. 1:01:06it's capital, because when you're in
  2069. 1:01:08the service industry, it's your
  2070. 1:01:09reputation you're putting in the hands
  2071. 1:01:11of a freelancer.
  2072. 1:01:13Every time for every mission, for every
  2073. 1:01:15client.
  2074. 1:01:15Completely. So we hire, it's not
  2075. 1:01:17regular, we hire more by the quarter, I
  2076. 1:01:20would say. It's not every week, you see
  2077. 1:01:21, it's more by the quarter. Now, we're
  2078. 1:01:23going to hire 15 people per quarter, 15
  2079. 1:01:25to 20 people per quarter, I think,
  2080. 1:01:26roughly.
  2081. 1:01:27It's about that. And indeed, we only
  2082. 1:01:29have one HR director, but that's only 5
  2083. 1:01:31%of her time because she does all the
  2084. 1:01:32staffing, prepares the training, and is
  2085. 1:01:34also the head of pedagogy. So it's a
  2086. 1:01:35lot, and I find that's the most
  2087. 1:01:37striking part; where you're right is
  2088. 1:01:38that service companies will be
  2089. 1:01:40transformed by AI. And you have two
  2090. 1:01:41things. You have the fact that we'll be
  2091. 1:01:44able to do more, you see. Thanks to AI,
  2092. 1:01:46as a service company, back in the day,
  2093. 1:01:48I might have only been a training
  2094. 1:01:49company. Today, I can do some
  2095. 1:01:51consulting, I can do some coaching and
  2096. 1:01:53support because we are very
  2097. 1:01:54well-equipped, and secondly, it's going
  2098. 1:01:56to blow up the margins. That's going to
  2099. 1:01:58be the big topic, though. There's a
  2100. 1:01:59margin issue in the service sector
  2101. 1:02:01that's coming, realizing that, well, we
  2102. 1:02:03see it, we price ourselves twice as
  2103. 1:02:05cheap as EY on equivalent assignments.
  2104. 1:02:07For instance, we had a case on a
  2105. 1:02:09digital audit assignment for a company
  2106. 1:02:10where we were two to three times
  2107. 1:02:12cheaper than EY, and the result, the
  2108. 1:02:13mission is finished. The result is the
  2109. 1:02:15same. And the big issue, well, the
  2110. 1:02:17result, or at least the quality of the
  2111. 1:02:19output, is the same. And the big issue
  2112. 1:02:21is that all these firms are built with
  2113. 1:02:23layers of consultants who are paid to
  2114. 1:02:25make slides, paid to write meeting
  2115. 1:02:27minutes, paid to interview employees,
  2116. 1:02:29and so on. And that is over. And for us
  2117. 1:02:31, for example, by using AI, it means
  2118. 1:02:33that you have two choices. Either you
  2119. 1:02:34skyrocket your margins, which we could
  2120. 1:02:36have done, or, as we did, we lower our
  2121. 1:02:37prices and skyrocket our volume. That's
  2122. 1:02:39the choice we decided to take. And that
  2123. 1:02:41is a major, major issue for service
  2124. 1:02:42companies. If you are well-equipped
  2125. 1:02:43with AI, I'll take a classic consulting
  2126. 1:02:45assignment as an example. Our interview
  2127. 1:02:46grids are made by AI; I say who to
  2128. 1:02:48interview, what the topic is, and so on
  2129. 1:02:50. Interview grid made by AI, we spend
  2130. 1:02:52zero time. When we conduct an interview
  2131. 1:02:54, we record it and then send that plus
  2132. 1:02:56the grid, plus our analysis framework,
  2133. 1:02:58and it gives us a full analysis of the
  2134. 1:03:00person's interview. That's something a
  2135. 1:03:02consultant normally does in a classic
  2136. 1:03:03consulting firm. As for the slides, I
  2137. 1:03:05don't know, maybe we price the cost of
  2138. 1:03:07creating slides at about 5%of a
  2139. 1:03:09mission's cost, whereas it's 30%.
  2140. 1:03:11Usually, in my opinion, it's
  2141. 1:03:12consultants who make all the slides.
  2142. 1:03:14Well, for us, no, it's AI that makes
  2143. 1:03:15the slides. We give it all the content,
  2144. 1:03:17it makes all the slides, and it's like
  2145. 1:03:18that the whole time. You have so, so,
  2146. 1:03:20so many little things done with AI,
  2147. 1:03:21which means that me, my partners—I
  2148. 1:03:23have partners—they manage to handle
  2149. 1:03:2510 projects at € 15,000 per project
  2150. 1:03:27at the same time, you see, or even a
  2151. 1:03:28bit more. And that, normally, you can't
  2152. 1:03:30do that either in a traditional setup.
  2153. 1:03:32So that is a huge, huge, huge topic.
  2154. 1:03:35You have many who say that thanks to AI
  2155. 1:03:36, people won't call on consulting or
  2156. 1:03:38training firms anymore. I don't believe
  2157. 1:03:39it. You will always need an expert; you
  2158. 1:03:41always need to have an intermediary. In
  2159. 1:03:43my opinion, we've been told for a long
  2160. 1:03:45time that intermediaries would die out
  2161. 1:03:46with the internet, but now there's AI,
  2162. 1:03:47and I don't really believe it. On the
  2163. 1:03:49other hand, clients are going to be
  2164. 1:03:50much more demanding, that's for sure.
  2165. 1:03:51And they'll say, "Why are you selling
  2166. 1:03:52me this project for € 20,000 when I
  2167. 1:03:53know you're going to do a lot of it
  2168. 1:03:55with AI?" Well, we say: "Yes, we won't
  2169. 1:03:56sell it to you for 20, we'll sell it to
  2170. 1:03:57you for 10," and, indeed, we'll be very
  2171. 1:03:59well-equipped behind the scenes. So
  2172. 1:04:00that is a big, big, big productivity
  2173. 1:04:02topic. All the companies doing SEO, the
  2174. 1:04:04SEO agencies, there's an AI topic
  2175. 1:04:06that's going to be huge. You won't be
  2176. 1:04:08able to charge for an SEO management
  2177. 1:04:10service today like you charged
  2178. 1:04:11yesterday, because—I won't say you'll
  2179. 1:04:13write your articles with AI, not
  2180. 1:04:14necessarily—but your semantic
  2181. 1:04:16research, keyword research, your
  2182. 1:04:17brainstorming on topics, all that,
  2183. 1:04:18you'll be doing with AI. So either you
  2184. 1:04:22boost your margins, or you lower your
  2185. 1:04:23prices compared to your competitors and
  2186. 1:04:25you'll grow in volume. But it's really,
  2187. 1:04:27really going to transform things today.
  2188. 1:04:29And at the same time, big firms can't
  2189. 1:04:31do that. BCG can't have all its slides
  2190. 1:04:33done by an AI, because BCG—who
  2191. 1:04:34tomorrow stops hiring, who loses a bit
  2192. 1:04:36of quality, you lose a bit of visual
  2193. 1:04:37quality...We, our slides, they aren't
  2194. 1:04:39all the same, for example. But who said
  2195. 1:04:41that consulting firm slides had to be
  2196. 1:04:43...that deck 1 for client 1 must be
  2197. 1:04:44exactly the same as deck 2 for client 2
  2198. 1:04:46? It's huge. Our clients couldn't care
  2199. 1:04:49less. They say it's beautiful, it's
  2200. 1:04:51structured, it's practical; whether the
  2201. 1:04:53logo is here or if for client 2 it's at
  2202. 1:04:55the bottom, we don't care. So we waste
  2203. 1:04:58zero time with that. I used to invest a
  2204. 1:04:59bit of time, I spent my time putting
  2205. 1:05:01the logo in the right place, fixing the
  2206. 1:05:03slides, making drawings and such. Oh
  2207. 1:05:06yes, the slides were beautiful in the
  2208. 1:05:07end, but it cost the time of a junior
  2209. 1:05:09being paid, they're paid € 10,000 a
  2210. 1:05:10month there, or € 7,000 a month, and
  2211. 1:05:12you charge that to your client. So
  2212. 1:05:15a huge, huge disruption for consulting
  2213. 1:05:17firms and service companies in general,
  2214. 1:05:18where we will indeed be able to do much
  2215. 1:05:20more, in my opinion, thanks to AI tools
  2216. 1:05:21.
  2217. 1:05:22I want to jump off that point about
  2218. 1:05:24tools. You mentioned earlier that when
  2219. 1:05:27you leave any kind of meeting, you take
  2220. 1:05:29a voice note and say, "Here's what I
  2221. 1:05:31saw." Quick, quick, quick, quick, quick
  2222. 1:05:33, quick. "Write me an email, do this,
  2223. 1:05:35do that." Uh, what makes you choose to
  2224. 1:05:38use your voice note rather than, for
  2225. 1:05:41example, putting a recorder on the
  2226. 1:05:42table, recording the entire meeting,
  2227. 1:05:45and then using an automated API that
  2228. 1:05:47sends it to ChatGPT, then to Slack, and
  2229. 1:05:49using automations to do it all? My
  2230. 1:05:53underlying question is, how do you
  2231. 1:05:55ultimately choose your daily workflow
  2232. 1:05:57given the plurality of tools we're
  2233. 1:05:59facing? That's a real issue. There are
  2234. 1:06:02maybe 100 AI tools coming out a day
  2235. 1:06:05from companies wanting to launch a SaaS
  2236. 1:06:07, a thing, a gadget, each time with
  2237. 1:06:09completely different or sometimes
  2238. 1:06:11similar features. You can't even tell
  2239. 1:06:14the difference between them, apart from
  2240. 1:06:16the landing page. How do you choose?
  2241. 1:06:19It's quite hard, you mustn't spread
  2242. 1:06:20yourself too thin, that's for sure. Uh,
  2243. 1:06:22you should always start with the
  2244. 1:06:24generalists. Once again, dig into
  2245. 1:06:26ChatGPT or Claude. Pick one or the
  2246. 1:06:28other, it doesn't matter, Claude is
  2247. 1:06:29very good, but explore everything it
  2248. 1:06:31can do. For me, that's my first piece
  2249. 1:06:33of advice. Uh, next, what I look at,
  2250. 1:06:35well, that also depends on B2B versus
  2251. 1:06:37B2C. B2C, have fun, you have a fairly
  2252. 1:06:40wide range of tools. In B2B, we are
  2253. 1:06:42constrained by tool security, that's
  2254. 1:06:44clear.
  2255. 1:06:44I would never put an unknown tool in
  2256. 1:06:46the hands of a company. It's impossible
  2257. 1:06:47. It's too dangerous. Tomorrow, if you
  2258. 1:06:50want to make a competitor to, I don't
  2259. 1:06:52know, what's a nice tool? Like Uta,
  2260. 1:06:54which is my meeting tool. You want to
  2261. 1:06:56create a Uta just as good, just as
  2262. 1:06:57powerful? If you're a 13-year-old in
  2263. 1:07:00India on your couch in Bangalore, you
  2264. 1:07:02can create it thanks to AI. In fact,
  2265. 1:07:04you'll create the same tool. So it's
  2266. 1:07:06very dangerous for companies to be
  2267. 1:07:07faced with many tools that are
  2268. 1:07:08beautiful, powerful, and look serious.
  2269. 1:07:10Some are French companies, others are
  2270. 1:07:1213-year-old kids on their couch in
  2271. 1:07:14their underwear in Bangalore. So there
  2272. 1:07:16is one you want to trust more than the
  2273. 1:07:18other. Except that companies don't know
  2274. 1:07:19who is behind the tools, who is
  2275. 1:07:20actually behind them. So we spend a lot
  2276. 1:07:21of time telling them, "OK, let's take
  2277. 1:07:23the example of meeting minutes, it's a
  2278. 1:07:24need you have, you need to implement a
  2279. 1:07:25tool. Well, we are going to screen all
  2280. 1:07:27the tools for you and see which one is
  2281. 1:07:29really secure because be careful not to
  2282. 1:07:30send all your data to things that
  2283. 1:07:31aren't secure. So, in business, we do
  2284. 1:07:34it like this: we look at all the use
  2285. 1:07:36cases, we tell ourselves there are 6 or
  2286. 1:07:377 tools that might be interesting for
  2287. 1:07:39you, we look at the budget, and
  2288. 1:07:40depending on the budget, we say:" Well,
  2289. 1:07:41we'll have to focus on two or three
  2290. 1:07:43tools, or at least two or three
  2291. 1:07:44categories of tools, "and for those, we
  2292. 1:07:46pick the most secure one for you.
  2293. 1:07:48That's a bit of our B2B approach. In
  2294. 1:07:51B2C, obviously, you are a bit freer
  2295. 1:07:52because you're dealing with less
  2296. 1:07:54confidential information and so on.
  2297. 1:07:56Well, on that note, I would rather tell
  2298. 1:07:58you to keep yourself a bit updated on
  2299. 1:07:59the latest tools that come out through
  2300. 1:08:01newsletters; in Misteriad, we often
  2301. 1:08:02talk about the latest tools. Now, there
  2302. 1:08:04are some random lists that I advise
  2303. 1:08:05against. They make lists of tools that
  2304. 1:08:07—well, not us, but people make lists
  2305. 1:08:09of tools that aren't very useful. Look
  2306. 1:08:11instead at the tools that are raising
  2307. 1:08:12money, it's that simple. But for me,
  2308. 1:08:14it's a topic; when I see a great slide
  2309. 1:08:17generation tool, I go and think, should
  2310. 1:08:20I use it or not? I'll type into the
  2311. 1:08:23internet what are the best AI slide
  2312. 1:08:24generation tools. I look at those who
  2313. 1:08:25have raised money. Well, that means
  2314. 1:08:27they're more serious, that they're here
  2315. 1:08:28to last, and so on. I try to avoid the
  2316. 1:08:29little unknown Chinese tool. You see,
  2317. 1:08:31with James Spark, we hesitated a lot at
  2318. 1:08:32first. We saw that it was serious.
  2319. 1:08:33There are big companies using it. We
  2320. 1:08:35saw that they had raised a lot of money
  2321. 1:08:37. Well, the probability that everyone
  2322. 1:08:38is wrong is also a bit low. So for us,
  2323. 1:08:40I look at it a lot like that, and
  2324. 1:08:42indeed, pick two tools, three tools
  2325. 1:08:43maximum. Look at where you have the
  2326. 1:08:45most needs. You see, for example, for
  2327. 1:08:47me, reports, I do them with ChatGPT,
  2328. 1:08:49but reports because I do few reports
  2329. 1:08:51where I can put a card like PL. So I
  2330. 1:08:53don't use that, and I don't use Slack,
  2331. 1:08:56I don't need it to send automatic
  2332. 1:08:58emails and so on. So in this case, I
  2333. 1:09:01tell myself my use case is more about
  2334. 1:09:03making little voice notes and getting
  2335. 1:09:04short reports that I send via
  2336. 1:09:05copy-paste directly. So, I look at the
  2337. 1:09:08use case like that, and depending on it
  2338. 1:09:09, I pick two or three tools, but that's
  2339. 1:09:11enough. Perplexity stays free, stay on
  2340. 1:09:14free by the way, for everything you can
  2341. 1:09:16use for free, keep Perplexity free. I
  2342. 1:09:18would take a paid ChatGPT, a free
  2343. 1:09:19Perplexity, maybe a paid James Spark,
  2344. 1:09:21it's still a super powerful tool. James
  2345. 1:09:23Spark or the equivalent, Manus, it's
  2346. 1:09:25kind of the future of AI, so it lets
  2347. 1:09:26you see something. When you already
  2348. 1:09:29have those three, 99%of the whole world
  2349. 1:09:31doesn't use any of the three, right? So
  2350. 1:09:34already with those three, that's 3 x 20
  2351. 1:09:35€ a month, or even Perplexity is free
  2352. 1:09:37, so more like 2 x 20 € a month.
  2353. 1:09:39You're already doing a lot of things,
  2354. 1:09:41completely. You're talking about
  2355. 1:09:42ChatGPT where you do your little
  2356. 1:09:43reports. How do you use it yourself? Is
  2357. 1:09:46it ChatGPT, just a new chat, or did you
  2358. 1:09:48create a GPT or an AI agent? I've
  2359. 1:09:50created few GPTs, or none at all on my
  2360. 1:09:52end. Then again, I have a somewhat
  2361. 1:09:55special job, where I'm in the field
  2362. 1:09:58almost all day. I'm giving conferences
  2363. 1:10:00all day, so I have fewer office use
  2364. 1:10:01cases, so I haven't created any GPTs,
  2365. 1:10:03well, I've created plenty, but I don't
  2366. 1:10:05use them daily.
  2367. 1:10:06For those listening, can you explain
  2368. 1:10:08what a GPT is first? a GPT, sorry. The
  2369. 1:10:10problem with ChatGPT is that every time
  2370. 1:10:13I want to use it, I have to give it a
  2371. 1:10:15new instruction, a new prompt; it
  2372. 1:10:17doesn't remember me. Now, there is
  2373. 1:10:19memory and so on, it's improving, but
  2374. 1:10:21let's assume it starts from a blank
  2375. 1:10:23slate. So, I might want to, for example
  2376. 1:10:26—let's take a use case—I'm a real
  2377. 1:10:28estate agent, I do viewings every day,
  2378. 1:10:31and I have a visit report template that
  2379. 1:10:34I want ChatGPT to follow. I don't want
  2380. 1:10:36to explain the visit report template to
  2381. 1:10:38it every single time. Plus, I only do
  2382. 1:10:40viewings in English. So, I want my
  2383. 1:10:42report to be translated into English. I
  2384. 1:10:43don't want to explain how to do it and
  2385. 1:10:45the template and the English every
  2386. 1:10:46single time. So, I tell myself, I'll
  2387. 1:10:48create a little robot that I teach to
  2388. 1:10:49follow my template and speak in English
  2389. 1:10:51. So, on ChatGPT, you can—it's called
  2390. 1:10:53creating your own robot, creating your
  2391. 1:10:55GPT—you give it an instruction. So,
  2392. 1:10:57for me, the instruction would be to
  2393. 1:10:58tell it that every time I trigger you,
  2394. 1:11:00you will follow this interview template
  2395. 1:11:01, you will respond in English, and so
  2396. 1:11:03on. You can give it documents that it
  2397. 1:11:06can use for inspiration. So I would
  2398. 1:11:09tell it, look at the attachment, I've
  2399. 1:11:10put in some English template documents
  2400. 1:11:12that you can use as a reference. I give
  2401. 1:11:13it that once and for all. I call my
  2402. 1:11:15robot" Visit Report. "And now, every
  2403. 1:11:18time I go to ChatGPT, instead of going
  2404. 1:11:20back to ChatGPT and telling it," You
  2405. 1:11:22are a visit expert, you will follow my
  2406. 1:11:24template, etc., "I go to the robot I
  2407. 1:11:26called" Visit Report "and there, I just
  2408. 1:11:28tell it," I just visited a property,
  2409. 1:11:30etc., "and automatically, it remembers
  2410. 1:11:32that it has to follow my template, use
  2411. 1:11:34my documents for reference, and do it
  2412. 1:11:36in English. And so, this way, you can
  2413. 1:11:38provide what are called
  2414. 1:11:39pre-instructions, reference documents
  2415. 1:11:40that you don't need to give it every
  2416. 1:11:42time. That works well. We see many HR
  2417. 1:11:44chatbots in companies. Actually, they
  2418. 1:11:46put all their HR procedures behind a
  2419. 1:11:48robot. They publish it on ChatGPT. The
  2420. 1:11:51robot is called" HR Assistant "and as
  2421. 1:11:52soon as someone joins the company, they
  2422. 1:11:54talk to the HR Assistant. They say,"
  2423. 1:11:55When can I take my vacation? "When do I
  2424. 1:11:57do this?" And actually, the HR
  2425. 1:11:58assistant on ChatGPT, well, it will
  2426. 1:12:00query the famous document that the HR
  2427. 1:12:02teams gave it 2 years ago, 2 months ago
  2428. 1:12:03, and uh that allows for providing
  2429. 1:12:05pre-instructions that way. So there you
  2430. 1:12:07go, that's a bot, it works well, you
  2431. 1:12:08can make many GPTs. But me, I don't
  2432. 1:12:11really, I don't really have any...and
  2433. 1:12:13yet, I think I query GPT 30 times a day
  2434. 1:12:16, 20, 30 times a day about all sorts of
  2435. 1:12:18things. I still prepare my meetings
  2436. 1:12:20with it.
  2437. 1:12:22GPT/Perplexity, I prepare my meetings
  2438. 1:12:24with that. I go see someone, I say give
  2439. 1:12:26me the three current events there. Give
  2440. 1:12:27me those three news items. Here is what
  2441. 1:12:29I sold to a small competitor. Tell me a
  2442. 1:12:30bit how, based on these news items.
  2443. 1:12:34That, that helps me a lot. Preparing,
  2444. 1:12:36uh, what do we call that? Sales pitches
  2445. 1:12:39. I use it a little bit but not GPTs
  2446. 1:12:41yet. I need to, I need to find one that
  2447. 1:12:42would really help me. And it changes so
  2448. 1:12:44much, actually. For my part, my job is
  2449. 1:12:46also to test as many tools as possible
  2450. 1:12:48constantly because the nightmare of a
  2451. 1:12:50consultant is showing up to a client
  2452. 1:12:51who knows more or knows a tool you
  2453. 1:12:53don't know. When it's tools not known
  2454. 1:12:55at all, it's fine, it's not a big deal.
  2455. 1:12:57I tell him, listen, we don't open 100 a
  2456. 1:12:59day. However, for the 10, 15, 20, 30
  2457. 1:13:01that are quite serious, that are
  2458. 1:13:03evolving a bit, I am constantly testing
  2459. 1:13:05them.
  2460. 1:13:05H
  2461. 1:13:06which means I spend less time on
  2462. 1:13:08ChatGPT for example and more on Gemini,
  2463. 1:13:10Claude every day, I think I use a bit
  2464. 1:13:12of Gemini, a bit of Manus, a bit of
  2465. 1:13:14Claude, NTA every day, uh, and a little
  2466. 1:13:16French one every day. So you see, I'm
  2467. 1:13:18more into testing a bit of everything
  2468. 1:13:20rather than having a good,
  2469. 1:13:21well-calibrated process on ChatGPT.
  2470. 1:13:23You're right to point out this, the
  2471. 1:13:24evolution of the tools, which is
  2472. 1:13:26ultra-fast. Uh, already when we see the
  2473. 1:13:28evolution of ChatGPT in 2 years, it's
  2474. 1:13:30just, uh, phenomenal. Before, we
  2475. 1:13:31couldn't even make, make, make images.
  2476. 1:13:33Uh, ChatGPT, the, the very first one,
  2477. 1:13:35what was it? Was it 2.5 or or 3.5
  2478. 1:13:383.5.5.5
  2479. 1:13:40it was pretty crazy. There have been
  2480. 1:13:41several. There have been a dozen
  2481. 1:13:42versions in all. Now only 5 is
  2482. 1:13:44accessible. The evolution of usage is
  2483. 1:13:47insane. I think we’re only at the
  2484. 1:13:48very beginning compared to everything
  2485. 1:13:50that's coming. And, uh, that multiplied
  2486. 1:13:52by, uh, hundreds of tools. Uh, I say
  2487. 1:13:55hundreds. I could say tens of thousands
  2488. 1:13:57, but let’s say hundreds for those
  2489. 1:13:59that are more or less serious, a few
  2490. 1:14:01dozen that are really very serious, and
  2491. 1:14:02then about fifteen, like you said, uh,
  2492. 1:14:04that we can, we can trust and really
  2493. 1:14:05build and move forward with. How do you
  2494. 1:14:08manage to stay constantly up to date
  2495. 1:14:10while still growing a hyper-growth
  2496. 1:14:12company? Because it requires
  2497. 1:14:13competitive intelligence on one hand
  2498. 1:14:14regarding competitors, and on the other
  2499. 1:14:16hand regarding tools, all while staying
  2500. 1:14:18focused on the growth of a company
  2501. 1:14:19that’s doubling every year. How do
  2502. 1:14:21you do it?
  2503. 1:14:22It’s hard, it’s very hard. I spend
  2504. 1:14:23an hour a day staying informed.
  2505. 1:14:24OK.
  2506. 1:14:25One hour a day of staying informed.
  2507. 1:14:26Actually, I force myself to publish,
  2508. 1:14:28which is a pretty good trick. Since I
  2509. 1:14:29force myself to post on LinkedIn every
  2510. 1:14:31morning, I have no choice but to stay
  2511. 1:14:32informed because if I don’t do my
  2512. 1:14:33research, I have nothing to say, so to
  2513. 1:14:35speak. So, uh, I do my research every
  2514. 1:14:36morning from 8 to 9 AM, uh, on AI, and
  2515. 1:14:38I’m on Twitter. I’m big on Twitter,
  2516. 1:14:41Twitter plus a newsletter I like, an
  2517. 1:14:43American one called The Rundown. It’s
  2518. 1:14:46the English equivalent of Mysteria, or
  2519. 1:14:47Mysteria is the French equivalent of
  2520. 1:14:48The Rundown. Put it that way. So, I do
  2521. 1:14:51my research like that for about an hour
  2522. 1:14:54a day, not so much on the tools, by the
  2523. 1:14:56way. I do it a little bit on tools, but
  2524. 1:14:58much more on current events. You see,
  2525. 1:15:00it’s important for me to know, is
  2526. 1:15:01OpenAI developing other things? Did
  2527. 1:15:04Mistral partner with Orange? If so,
  2528. 1:15:06what does that mean? It wasn’t them
  2529. 1:15:09who did it, by the way, it was more
  2530. 1:15:11OpenAI, but you see, I’m trying more
  2531. 1:15:12to understand the macro ecosystem, how
  2532. 1:15:14it evolves within companies. So, I do a
  2533. 1:15:16lot of research on that for my part.
  2534. 1:15:19Plus, we have internal briefings every
  2535. 1:15:20week; all Mysteria trainers and
  2536. 1:15:21consultants are trained by an external
  2537. 1:15:23person who comes to brief us. So, I
  2538. 1:15:25participate in that too, and I listen
  2539. 1:15:26to it. Uh, so that’s the research we
  2540. 1:15:28do internally, which means there’s
  2541. 1:15:29little chance there’s an AI topic I
  2542. 1:15:31don’t master today, you see, it’s
  2543. 1:15:32impossible. Uh, for someone whose
  2544. 1:15:34full-time business isn't this—I don't
  2545. 1:15:36have a choice, it is my full-time
  2546. 1:15:38business and I love doing it. But at a
  2547. 1:15:40minimum, read a newsletter, the one
  2548. 1:15:42from Mysteria, it’s once a week,
  2549. 1:15:44there are five news items, one news
  2550. 1:15:46item—
  2551. 1:15:48if you read that, you know everything
  2552. 1:15:50about, well, you know everything.
  2553. 1:15:51You're on top of a little bit of what's
  2554. 1:15:52happening in AI and testing one small
  2555. 1:15:54tool a week, you know, over the weekend
  2556. 1:15:55, I test a small tool, that's something
  2557. 1:15:57people can do quite easily. Uh, just
  2558. 1:15:59with that, same thing, with that you'll
  2559. 1:16:01be better than 95%of your competitors,
  2560. 1:16:02and it's better to test one or two
  2561. 1:16:04tools and actually deploy them all the
  2562. 1:16:05way. Your competitors aren't doing it,
  2563. 1:16:07so do it, rather than telling yourself
  2564. 1:16:08you'll test two or three a week and
  2565. 1:16:09then in the end, none of them go into
  2566. 1:16:11production. So for me, I'd say read a
  2567. 1:16:13newsletter, Mystery AI or another, and
  2568. 1:16:15uh, and test a tool every, every week,
  2569. 1:16:17a small tool, and you'll already see a
  2570. 1:16:19lot of things. By the way, in The
  2571. 1:16:22Rundown, it's very good because The
  2572. 1:16:23Rundown is a great newsletter—it's in
  2573. 1:16:25English, mind you—but it gives you
  2574. 1:16:26the three or four news items, they
  2575. 1:16:27provide a few business use cases, and
  2576. 1:16:29they describe one tool in a bit more
  2577. 1:16:30depth, saying: "Here is what it does,
  2578. 1:16:31here is what companies can do with it."
  2579. 1:16:34So it's not bad, honestly, just that
  2580. 1:16:36alone is good, and you have to be
  2581. 1:16:37disciplined for those who love AI but
  2582. 1:16:38whose business it isn't, you have to be
  2583. 1:16:40disciplined. not to dig into everything
  2584. 1:16:42, because we see everything, it's
  2585. 1:16:44infinite, and we see everything and
  2586. 1:16:46anything, and
  2587. 1:16:47it goes in every direction, so you
  2588. 1:16:48shouldn't, no, at first we dug into
  2589. 1:16:50everything, we dug into everything, but
  2590. 1:16:51after a while, you just want to jump
  2591. 1:16:52out the window. So you have to be
  2592. 1:16:53disciplined a bit, tell yourself there
  2593. 1:16:55are lots of things in AI, I follow
  2594. 1:16:56Mysteria or Rundown which are very
  2595. 1:16:57serious, I see a bit of what's being
  2596. 1:16:58done, and that'll already be pretty
  2597. 1:17:00good.
  2598. 1:17:00You have mentioned confidentiality and
  2599. 1:17:02data several times, we could even talk
  2600. 1:17:04about GDPR. Uh,
  2601. 1:17:06what about that? Because sometimes some
  2602. 1:17:08people, as we see, say, "No, but AI, we
  2603. 1:17:10don't know what they'll do with our
  2604. 1:17:12data, we could lose everything, it's
  2605. 1:17:14too dangerous," et cetera. "Others say,
  2606. 1:17:17" No, not at all, when you just get the
  2607. 1:17:19subscription, well, you click this box
  2608. 1:17:21and then it's completely secure. "You
  2609. 1:17:23who implement this daily in companies,
  2610. 1:17:25with entrepreneurs, with individuals,
  2611. 1:17:27what is the truth? There is a big issue
  2612. 1:17:30, which is that most AI tools are
  2613. 1:17:32American and they will remain American.
  2614. 1:17:34I don't know if you've seen the latest
  2615. 1:17:36funding rounds, but OpenAI is set to
  2616. 1:17:37raise nearly 100 billion at a 500
  2617. 1:17:39billion valuation. We are talking about
  2618. 1:17:41a private company that will be worth
  2619. 1:17:42500 billion, which had no product 2 or
  2620. 1:17:443 years ago. So that is the first point
  2621. 1:17:46. Their competitor, Anthropic, just
  2622. 1:17:49raised 18 billion or 14 billion at a
  2623. 1:17:51200 billion valuation. So we will not
  2624. 1:17:53catch up with those people. You have to
  2625. 1:17:54forget about Mistral, meanwhile, the
  2626. 1:17:56French one just raised 1 billion at a
  2627. 1:17:5714 billion valuation. 14 versus 180
  2628. 1:17:59versus 500. Americans, as soon as they
  2629. 1:18:01raise their hand, they find 50 or 100
  2630. 1:18:03billion wherever they want, all over
  2631. 1:18:04the world. So I think we won't catch up
  2632. 1:18:06with that. So the big players in AI are
  2633. 1:18:08American and will remain so. So there
  2634. 1:18:11is an issue, which is that the data you
  2635. 1:18:13put into those tools ends up in the
  2636. 1:18:15United States. That much is clear. At
  2637. 1:18:17the very least, because OpenAI is
  2638. 1:18:18opening servers, notably in Europe. At
  2639. 1:18:20the very least, it might perhaps stay
  2640. 1:18:21in Europe with Microsoft Copilot. It
  2641. 1:18:22could also stay in Europe. But on
  2642. 1:18:25servers that belong to Americans. And
  2643. 1:18:27it's quite funny because, very often
  2644. 1:18:29with AI, we hear a lot that it will
  2645. 1:18:31feed the models, it will train...if all
  2646. 1:18:33the data we put in goes to feed the
  2647. 1:18:35American models, etc. And we have many
  2648. 1:18:37clients who tell us," Oh no, I don't
  2649. 1:18:39want to use ChatGPT, it's the big bad
  2650. 1:18:42wolf. "However, I use SharePoint, folks
  2651. 1:18:44. SharePoint, those are servers that
  2652. 1:18:46belong to Microsoft, the same way that
  2653. 1:18:48OpenAI servers are Microsoft servers.
  2654. 1:18:51The problem, however, behind all that,
  2655. 1:18:52is that the Microsoft server is subject
  2656. 1:18:54to the American Cloud Act. The American
  2657. 1:18:56Cloud Act, which explains to us that,
  2658. 1:18:58in fact, the Americans, it is the
  2659. 1:18:59extraterritoriality of American law. At
  2660. 1:19:01any time, the U.S. government or the
  2661. 1:19:04CIA, or at least major U.S. public
  2662. 1:19:06agencies, can get their hands on 100%of
  2663. 1:19:08the data belonging to companies that is
  2664. 1:19:11stored by American firms anywhere in
  2665. 1:19:13the world. It means a Microsoft server,
  2666. 1:19:15whether it's in Mississippi, Saint-
  2667. 1:19:18Étienne, Toulouse, or Poland—if it's
  2668. 1:19:20a Microsoft server, the Americans can
  2669. 1:19:22get their hands on the data at any time
  2670. 1:19:25. So we often say, okay, for businesses
  2671. 1:19:27, we'll use secure tools that don't
  2672. 1:19:29train their models with your data. You
  2673. 1:19:31have to be careful with that; it's a
  2674. 1:19:32shame. Be careful with the free
  2675. 1:19:34versions of these tools, by the way.
  2676. 1:19:35Everything you put on the free cloud,
  2677. 1:19:37free GPT, free Perplexity—they take
  2678. 1:19:38it and train their models with it. The
  2679. 1:19:41probability that you'll find what was
  2680. 1:19:43said tomorrow in a trained model is
  2681. 1:19:44virtually zero. But still, I advise
  2682. 1:19:46against doing that, and we never put
  2683. 1:19:48companies on these free versions. Once
  2684. 1:19:50you've moved to the paid versions of
  2685. 1:19:52these tools, they comply with GDPR and
  2686. 1:19:55they do not train on your data. When
  2687. 1:19:58you're on a paid version—I'll take
  2688. 1:19:59ChatGPT, which is supposedly the least
  2689. 1:20:01secure. When you're on the paid version
  2690. 1:20:03, they don't train their models with
  2691. 1:20:05your data. That's the first thing. It's
  2692. 1:20:08GDPR compliant, so you can delete your
  2693. 1:20:10data and so on. However, yes, it will
  2694. 1:20:12pass through American servers, but
  2695. 1:20:14you're already routing everything
  2696. 1:20:16through American servers anyway. So I
  2697. 1:20:18often say: it's a fake problem that
  2698. 1:20:19we're trying to invent for ourselves
  2699. 1:20:21today. But we probably should have
  2700. 1:20:22asked ourselves that question 10 years
  2701. 1:20:23ago when the cloud was born, because
  2702. 1:20:24the real problem is that we don't have
  2703. 1:20:25a French cloud. And that's the real
  2704. 1:20:28issue: if we had a French cloud, then
  2705. 1:20:29we could say," Okay, OpenAI, as long as
  2706. 1:20:31their cloud isn't with a French
  2707. 1:20:32provider, as long as they aren't using
  2708. 1:20:34OVH—I'll put it that way—then okay,
  2709. 1:20:35we won't use them. "" But we don't have
  2710. 1:20:38a French cloud; we have OVH, but it's
  2711. 1:20:40tiny compared to giants like Microsoft,
  2712. 1:20:42Amazon, and so on. "And so today we are
  2713. 1:20:44forced to use things where everything
  2714. 1:20:46we use daily is stored on servers that
  2715. 1:20:48belong to Americans, and unfortunately,
  2716. 1:20:49that won't change. So, be careful, yes.
  2717. 1:20:52Don't just use any tool; yes, choose
  2718. 1:20:54paid corporate tools rather than free
  2719. 1:20:55ones to prevent them from training
  2720. 1:20:57their models. Yes. But having a general
  2721. 1:21:00psychosis that says we can only use
  2722. 1:21:02Mistral because of the data problem. No
  2723. 1:21:05, if you want to use Mistral for
  2724. 1:21:06sovereignty, because you want to
  2725. 1:21:08promote French companies, okay, that's
  2726. 1:21:10fine, and we should all be a bit more
  2727. 1:21:12patriotic like that, why not. But the
  2728. 1:21:14data argument, you have to be a bit
  2729. 1:21:16careful because data—I've seen
  2730. 1:21:17executive committees tell me they would
  2731. 1:21:18never deploy ChatGPT. And during the
  2732. 1:21:20meeting, she gives her colleague
  2733. 1:21:21something and says," Wait, here, I sent
  2734. 1:21:22you the report via WhatsApp. "" But
  2735. 1:21:24what is WhatsApp? "If it's not an
  2736. 1:21:26insecure tool, 100%of it goes to
  2737. 1:21:28American servers, and it's actually
  2738. 1:21:30stored on those servers. That's not
  2739. 1:21:32ChatGPT at all." So you see, there's a
  2740. 1:21:34big, uh, there's a lot of ignorance
  2741. 1:21:35about this. "It's normal; we haven't
  2742. 1:21:37been very well sensitized to it. But
  2743. 1:21:38when you're not sensitized, rather than
  2744. 1:21:40having grand principles, just say,"
  2745. 1:21:41We'll be careful, we'll keep an eye on
  2746. 1:21:43things, and we won't do anything too
  2747. 1:21:44stupid. "" But the grand principles of'
  2748. 1:21:46We absolutely forbid ChatGPT and the
  2749. 1:21:48like '. "On the other hand, we talk to
  2750. 1:21:50each other on WhatsApp at work; that's
  2751. 1:21:52not very consistent, completely. Uh,
  2752. 1:21:54we've talked a lot about usage in
  2753. 1:21:56companies, large groups or others, uh,
  2754. 1:21:58entrepreneurs, and then you mentioned
  2755. 1:22:00individuals that you also train. Uh,
  2756. 1:22:03what are the uses we can have in our
  2757. 1:22:05personal lives? Whether it's for
  2758. 1:22:07training, learning, maybe managing
  2759. 1:22:09finances better, doing sports, why not.
  2760. 1:22:12Uh, that could be interesting. You've
  2761. 1:22:14already mentioned quite a few that work
  2762. 1:22:15well. Uh, you can use it as a personal
  2763. 1:22:17coach; I like that, I use it as a
  2764. 1:22:19personal coach with a lot of
  2765. 1:22:20perspective. Be careful, there have
  2766. 1:22:22been issues with people who committed
  2767. 1:22:24suicide because of unsecured AIs. So,
  2768. 1:22:25first of all, when you do that, use
  2769. 1:22:28mainstream AIs like ChatGPT. Meaning
  2770. 1:22:30the day it pushes someone toward
  2771. 1:22:32suicide, in a millisecond, uh, there
  2772. 1:22:33are people monitoring everything said
  2773. 1:22:35on ChatGPT, so they are highly watched,
  2774. 1:22:37so they can't do much, whereas there
  2775. 1:22:39are open-source AIs that can make you
  2776. 1:22:40do anything. So, do this instead; I do
  2777. 1:22:42it on ChatGPT, but I have a personal
  2778. 1:22:44coach where I tell it everything,
  2779. 1:22:45essentially, and again, everyone will
  2780. 1:22:47do as they please, I don't know if it's
  2781. 1:22:48the best practice. I give it everything
  2782. 1:22:50. I operate on the principle that my
  2783. 1:22:52health data is managed by Apple today.
  2784. 1:22:53I give it all; I'm using an American
  2785. 1:22:55service. It sends everything to my
  2786. 1:22:56iPhone, which is with Apple. So, my
  2787. 1:22:58health data is already with the
  2788. 1:22:59Americans, as far as I'm concerned. I
  2789. 1:23:00assume they have more to offer me in
  2790. 1:23:01terms of positives than negatives by
  2791. 1:23:03knowing all my data. So, in any case, I
  2792. 1:23:06have a ChatGPT that knows everything
  2793. 1:23:08about me and it's a personal coach.
  2794. 1:23:10Meaning I talk to it two or three times
  2795. 1:23:12a week about what I'm doing, what's
  2796. 1:23:13going well, what's not, and so on. And
  2797. 1:23:16instead of telling it," You're a
  2798. 1:23:17psychologist, though I don't know if I
  2799. 1:23:19need one, "I instead tell it to give me
  2800. 1:23:20actions I can take to improve what you
  2801. 1:23:22feel is lacking in me. Like, I told you
  2802. 1:23:24for three weeks that I didn't have time
  2803. 1:23:26to see my friends because I was working
  2804. 1:23:27too much. Well, when I explain all that
  2805. 1:23:28, without even being very specific, I
  2806. 1:23:30just tell it," Here, I did this today,
  2807. 1:23:31that, and so on and so forth. "Well, it
  2808. 1:23:32will tell me,' Wait, I realize you
  2809. 1:23:34haven't seen your friends in three
  2810. 1:23:35weeks. So, send a message to two or
  2811. 1:23:37three of your friends to see how they
  2812. 1:23:38are; it will get you back in the loop,
  2813. 1:23:40and you can book a breakfast. '" See,
  2814. 1:23:41it's just really simple stuff. It helps
  2815. 1:23:43me a lot with that. Uh, you see, a
  2816. 1:23:44really simple thing about my personal
  2817. 1:23:46life. Uh, I realize I work so hard
  2818. 1:23:47during the week that I end up at 10
  2819. 1:23:49p.m. on Friday, closing my laptop and
  2820. 1:23:51saying, "What am I doing this weekend?"
  2821. 1:23:52Except your friends don't wait until 10
  2822. 1:23:54p.m. to decide what they're doing for
  2823. 1:23:55the weekend. So you end up seeing the
  2824. 1:23:56same people, doing the same things.
  2825. 1:23:58Well, you talk about that with your AI
  2826. 1:24:00for three weeks. Well, I wasn't even
  2827. 1:24:01talking about that, I was talking about
  2828. 1:24:03my life. And he finally told me, "Look,
  2829. 1:24:05here's my advice: let's do one thing—
  2830. 1:24:07take a weekend, sit down, and plan your
  2831. 1:24:08weekends for the next six weeks, at
  2832. 1:24:10least then you won't be in a rush
  2833. 1:24:11looking for this or that.""It's so
  2834. 1:24:12simple, but it's going to change my
  2835. 1:24:13life." I mean, really, I've been doing
  2836. 1:24:14so much better since I started doing
  2837. 1:24:15that. So you see, I have a personal
  2838. 1:24:17coach like that
  2839. 1:24:18who helps me a bit with these types of
  2840. 1:24:19things. Sometimes he asks me, "What
  2841. 1:24:21would you like to improve about
  2842. 1:24:23yourself as an entrepreneur?""Um, so I
  2843. 1:24:25don't know, I tell him,' Well, here's
  2844. 1:24:26what I do. '""He says,' Well, wait, I
  2845. 1:24:27think you'd benefit from being more
  2846. 1:24:29creative. '""So I say,' Well, give me
  2847. 1:24:30some creativity exercises. '""And just
  2848. 1:24:32recently, he gave me a creativity
  2849. 1:24:33exercise that is brilliant." Every
  2850. 1:24:35morning, he sends me—well, I trigger
  2851. 1:24:37it, but he sends me a theme every
  2852. 1:24:39morning, or basically, he gives me a
  2853. 1:24:41theme where I have to come up with 10
  2854. 1:24:43ideas. For example, he'll say, "Imagine
  2855. 1:24:4610 features you could buy on Facebook."
  2856. 1:24:49"The next day, he says,' Imagine 10
  2857. 1:24:51companies that Snapchat could acquire.
  2858. 1:24:52'""And why?""Imagine 10 Tesla models
  2859. 1:24:56that Tesla could create and why?" It's
  2860. 1:24:59a creativity exercise that is super
  2861. 1:25:00interesting. He's the one who gave it
  2862. 1:25:02to me because he told me, "In your job,
  2863. 1:25:04you see the same people a lot, you give
  2864. 1:25:06the same talks often, so your
  2865. 1:25:07creativity is fading a bit.""That's a
  2866. 1:25:09big topic for me, I love the personal
  2867. 1:25:10coach." And obviously, you can input
  2868. 1:25:12all your data, you know. Me, my health
  2869. 1:25:13data, but everyone will do what they
  2870. 1:25:15want, of course. I think that if there
  2871. 1:25:17are some listening to us in the
  2872. 1:25:18comments and so on, not everyone will
  2873. 1:25:20agree, but it finds things in my blood
  2874. 1:25:21tests that my doctor doesn't find. It's
  2875. 1:25:24true. It's, and again, I'm not advising
  2876. 1:25:27you to get treated based on ChatGPT,
  2877. 1:25:28but telling it, "Here is my blood test,
  2878. 1:25:30what do you think?" It knows my
  2879. 1:25:32lifestyle, so very quickly it tells me,
  2880. 1:25:34"Well, you're not getting enough sleep.
  2881. 1:25:35""That's why this thing is high." My
  2882. 1:25:37doctor doesn't. "So those are things
  2883. 1:25:39that can be done." Now, there might be
  2884. 1:25:41more secure AIs than this one to do it.
  2885. 1:25:44Personally, I give it a lot of personas
  2886. 1:25:45, and then obviously everything related
  2887. 1:25:47to workout programs, I do that a lot.
  2888. 1:25:50Uh, what is going to help me learn? Hey
  2889. 1:25:52, I've got this topic. Oh, wait, I
  2890. 1:25:54actually have one I've been using
  2891. 1:25:55lately that I love. When I read the
  2892. 1:25:56newspaper, every morning I read Les
  2893. 1:25:58Échos, and there are plenty of things
  2894. 1:25:59I don't understand. Screenshot, explain
  2895. 1:26:01it to me in simple words. It's
  2896. 1:26:02wonderful. It's really about things
  2897. 1:26:04like, why does the increase in debt in
  2898. 1:26:05such-and-such country have an impact on
  2899. 1:26:07youth employment rates in this other
  2900. 1:26:08country? They explained it quickly
  2901. 1:26:10there, and I didn't get it. Snap a
  2902. 1:26:11photo for GPT, and it explains it to me
  2903. 1:26:13. Actually, you learn so much faster
  2904. 1:26:15with that, and that's a purely personal
  2905. 1:26:16thing, or one that works in
  2906. 1:26:17professional life too, but it's a
  2907. 1:26:18personal thing that works well. And
  2908. 1:26:20another point, a final one that is less
  2909. 1:26:22about learning, but regarding movies: I
  2910. 1:26:25realized recently, I was watching a
  2911. 1:26:27film, *Tenet*or I don't know what, that
  2912. 1:26:29was a bit complex, and there was a
  2913. 1:26:31scene I didn't understand. I tell it,
  2914. 1:26:33the guy is referring to something, but
  2915. 1:26:34I didn't understand what, and I ask
  2916. 1:26:36ChatGPT, it explains it to me, and
  2917. 1:26:37what's more, because it knows the movie
  2918. 1:26:38, it has read a thousand forums where
  2919. 1:26:40people explain it, and it unlocks it
  2920. 1:26:41for you, and that's quite, quite cool.
  2921. 1:26:43You watch the end of the movie and you
  2922. 1:26:44understand why he did that, and there
  2923. 1:26:45are tons of cases like that. There are
  2924. 1:26:47indeed plenty of examples, and you
  2925. 1:26:48mentioned the fact of analyzing blood
  2926. 1:26:50tests, uh,
  2927. 1:26:52in the context of health coaching, etc.
  2928. 1:26:54, or the data that a Whoop or an Apple
  2929. 1:26:55device, whatever, might give you. Those
  2930. 1:26:57are good examples that I've used myself
  2931. 1:26:59and that I also use on a daily basis. I
  2932. 1:27:01use AI. However, there was a rather
  2933. 1:27:04disappointing effect, at least
  2934. 1:27:05perceived by people overall, during the
  2935. 1:27:07release of the new version of ChatGPT,
  2936. 1:27:09GPT-5, where they had mentioned the
  2937. 1:27:11fact that, at a minimum, all answers to
  2938. 1:27:13certain questions would be at a PhD
  2939. 1:27:15level, it was going to be much more
  2940. 1:27:17advanced, and so on. Personally, I
  2941. 1:27:21wasn't necessarily disappointed by this
  2942. 1:27:22new version. In your case, what do you
  2943. 1:27:24think about it? Why was there that
  2944. 1:27:26effect? Is it because it had evolved so
  2945. 1:27:28much in previous times that we were
  2946. 1:27:29expecting something incredible? That
  2947. 1:27:31could be interesting.
  2948. 1:27:32I think it's a good mix, indeed, of all
  2949. 1:27:33that. In my opinion, the model is
  2950. 1:27:35better. The model is better. There's no
  2951. 1:27:37debate. That's for sure, benchmark-wise
  2952. 1:27:39, the model is better.
  2953. 1:27:40The model has evolved. Does it have a
  2954. 1:27:42way of answering you that's a bit
  2955. 1:27:44different? And actually, it's super
  2956. 1:27:47hard for a human eye to say it's better
  2957. 1:27:49or worse. What does a better model even
  2958. 1:27:50mean? In fact, I have a lot of people
  2959. 1:27:52telling me that since it answers
  2960. 1:27:53differently, it's worse. Well, not
  2961. 1:27:55necessarily. So it's not worse. It
  2962. 1:27:58answers in a different way, that's for
  2963. 1:27:59sure. There's a big bias too, as you
  2964. 1:28:02said, since it evolved so much between
  2965. 1:28:04versions 3 and 4, and between 4 and 1,
  2966. 1:28:06that now the improvements are more
  2967. 1:28:08marginal, that's certain. So we were
  2968. 1:28:11expecting a huge jump. No, the jump is
  2969. 1:28:13smaller, that's certain too. And then,
  2970. 1:28:15we just get used to it. We get used to
  2971. 1:28:17the model. Recently, I think it was GPT
  2972. 1:28:19-4, there were these charts: they ask
  2973. 1:28:21people all the time what they think
  2974. 1:28:23about performance, and the performance
  2975. 1:28:24curve of GPT-4 dropped in people's
  2976. 1:28:26opinion between March '24 and the end
  2977. 1:28:27of '24. Everyone thought, "That's weird
  2978. 1:28:30, that's not normal." OpenAI said, "
  2979. 1:28:31It's the same model, folks, it hasn't
  2980. 1:28:33changed," so at the very least it's the
  2981. 1:28:35same, it can't be worse. And in fact,
  2982. 1:28:37when you looked deeper, scientists
  2983. 1:28:39realized that as we got used to ChatGPT
  2984. 1:28:41, we were less and less impressed by
  2985. 1:28:43its answers, and so we felt it was
  2986. 1:28:45worse because, back then, you'd tell it
  2987. 1:28:47, "Give me a recipe." It would give you
  2988. 1:28:49a recipe and you'd say, "Wow, it's
  2989. 1:28:50incredible, it works so well." Six
  2990. 1:28:53months later, on the same model, you
  2991. 1:28:55wouldn't just ask for a recipe, you'd
  2992. 1:28:56say, "X grams of this, X milligrams of
  2993. 1:28:57that.""Look at the photo, look at my
  2994. 1:29:00utensils, give me a recipe, milligram
  2995. 1:29:02by milligram." And then, it wasn't
  2996. 1:29:03perfect. So, you'd say, "Ah, it's worse
  2997. 1:29:04than before." No, it's not worse. It's
  2998. 1:29:06just that we've learned how to use it
  2999. 1:29:07and we are increasingly demanding. So,
  3000. 1:29:09I think with GPT-5, the expectations
  3001. 1:29:11have also increased a lot. They hyped
  3002. 1:29:12it up a bit by telling us, "It's going
  3003. 1:29:14to be incredible, it's a model like no
  3004. 1:29:15other." It turns out it is better, but
  3005. 1:29:17not by a massive amount. And so, I
  3006. 1:29:19think there was a bit of disappointment
  3007. 1:29:20as well. But that's mainly it, it's
  3008. 1:29:21very biased, you know. In fact, it's a
  3009. 1:29:23model that is objectively better, that
  3010. 1:29:24reasons better, that writes better,
  3011. 1:29:26that has fewer writing biases in what
  3012. 1:29:27it does. Uh so no, it's a better model
  3013. 1:29:30and there too, people shouldn't
  3014. 1:29:32overthink the model. It's not the model
  3015. 1:29:34that does the work, it's your way of
  3016. 1:29:36using it and the quality of the product
  3017. 1:29:37. Those are two big productivity
  3018. 1:29:39factors for an AI tool. The model today
  3019. 1:29:42, I think 90%of people, if you don't
  3020. 1:29:44write GPT-5, you run 4 EOS behind it or
  3021. 1:29:46Llama 3 or Claude 3.5, they don't know
  3022. 1:29:49who they're talking to. That's normal,
  3023. 1:29:51these models are quite similar. So
  3024. 1:29:53there you go, completely. I have three
  3025. 1:29:55final questions for you, two more
  3026. 1:29:57business-related. The first one is, you
  3027. 1:29:59are two with your brother as partners.
  3028. 1:30:02How do you divide the roles? Well,
  3029. 1:30:05until today, he was very B2C, and I was
  3030. 1:30:07very B2B.
  3031. 1:30:08OK.
  3032. 1:30:09So that was a division that is changing
  3033. 1:30:10a bit in September because the B2B side
  3034. 1:30:12is really exploding for us. So, we're
  3035. 1:30:13both going to be on B2B. We'll split
  3036. 1:30:15the accounts, actually. We'll split the
  3037. 1:30:17accounts. There are two or three
  3038. 1:30:19industries I'm very strong in, so I
  3039. 1:30:20keep those accounts. He is more into
  3040. 1:30:22real estate, actually. He handles real
  3041. 1:30:23estate accounts very well, so he keeps
  3042. 1:30:25the real estate accounts, so we're
  3043. 1:30:26splitting the accounts among ourselves.
  3044. 1:30:29What we call the Misteria partners,
  3045. 1:30:30meaning our four true consultant
  3046. 1:30:32partners plus the three associates—
  3047. 1:30:33there's a third one who joined—the 7
  3048. 1:30:35of us, we split the accounts, actually.
  3049. 1:30:37So, it's more like that. Today, we also
  3050. 1:30:39split the tasks a bit. We bought back
  3051. 1:30:41…so our two former companies have
  3052. 1:30:43been integrated into the business, we
  3053. 1:30:45bought them back. So, he will manage
  3054. 1:30:46those two media outlets in addition; I
  3055. 1:30:48don't manage them at all, I manage the
  3056. 1:30:50recruitment parts a bit more, he
  3057. 1:30:51doesn't do finance at all. We are
  3058. 1:30:53really both splitting things up a bit
  3059. 1:30:54like that, but we try not to do the
  3060. 1:30:56same things. There are few subjects
  3061. 1:30:57that we work on together.
  3062. 1:30:58Which means that, in fact, I trust him
  3063. 1:31:00100%, he trusts me 100%, and that makes
  3064. 1:31:02things go super fast. So that's good.
  3065. 1:31:05Completely. And that answers the other
  3066. 1:31:08question a bit regarding the breakdown
  3067. 1:31:10between B2C and B2B, which is a real
  3068. 1:31:13topic. Many companies say you have to
  3069. 1:31:15stay focused and that you can't fight
  3070. 1:31:17three battles at once, or even two. Yet
  3071. 1:31:20today, that is what you are doing, with
  3072. 1:31:22major accounts on one side, more
  3073. 1:31:23traditional companies on the other, and
  3074. 1:31:25B2C as well. What made you decide to
  3075. 1:31:28take on these three different fronts?
  3076. 1:31:30Is it because there is a global market
  3077. 1:31:32opportunity, while at the same time
  3078. 1:31:34responding to the high demand from
  3079. 1:31:36companies, as you mentioned, by
  3080. 1:31:38recentering resources around B2B? That
  3081. 1:31:41could be interesting.
  3082. 1:31:41I would have given the same advice at
  3083. 1:31:43the beginning, saying don't diversify,
  3084. 1:31:45it's not a good idea. It turns out the
  3085. 1:31:48AI market is so enormous that you take
  3086. 1:31:50a little less risk when you diversify
  3087. 1:31:52in that sector. That's for sure. There
  3088. 1:31:55is also the fact that the two talk to
  3089. 1:31:56each other a lot. Actually, we have a
  3090. 1:31:58huge number of trainees and people
  3091. 1:31:59trained in B2C who come back and say, "
  3092. 1:32:01Okay, now we're going to train my
  3093. 1:32:02company." So for us, that is very
  3094. 1:32:04important. And conversely, I have many
  3095. 1:32:06companies tell me, "We're an SME, we
  3096. 1:32:08can't afford Misteria training."
  3097. 1:32:10However, everyone has a CPF account, so
  3098. 1:32:12come to us and we'll fund it, and
  3099. 1:32:13anyone who wants to be trained can do
  3100. 1:32:15so, because the employer doesn't have
  3101. 1:32:16the right to force you to use your CPF.
  3102. 1:32:18CPF is personal. However, the employer
  3103. 1:32:20can very well say, "Listen, I don't
  3104. 1:32:22have the budget to train you, sorry.""
  3105. 1:32:23If anyone really wants to get trained,
  3106. 1:32:25well, I'll ask Misteria to run a
  3107. 1:32:26session for you.""You get a group of 5
  3108. 1:32:28together, they come here to the company
  3109. 1:32:29, we customize it, and so on," and
  3110. 1:32:30there, they use their CPF. So B2C and
  3111. 1:32:33B2B talk to each other a lot at our
  3112. 1:32:35company, which is why we keep both. We
  3113. 1:32:38brought someone on very quickly;
  3114. 1:32:39someone arrived recently. In our
  3115. 1:32:40company, they manage B2C, which is also
  3116. 1:32:41important for us. I didn't want the two
  3117. 1:32:43founders to be spread too thin between
  3118. 1:32:45the two. We really have someone
  3119. 1:32:46dedicated to that. And there you go,
  3120. 1:32:49and the final point I would make, it
  3121. 1:32:51was also a way for us not to put all
  3122. 1:32:53our eggs in one basket in an ecosystem
  3123. 1:32:56that is certainly huge, but where we
  3124. 1:32:58still have little visibility on AI.
  3125. 1:33:00It's it's always the same, even if I am
  3126. 1:33:03100%convinced of the depth of the tech,
  3127. 1:33:05we don't know how the market will
  3128. 1:33:07evolve in 5 years. When you build a
  3129. 1:33:09SaaS on a classic theme, well, you have
  3130. 1:33:11a bit of visibility and so on, and
  3131. 1:33:13there indeed you wonder, should I do
  3132. 1:33:15B2B or B2C? For us, it's very, very
  3133. 1:33:17hard. So we said let's put some eggs in
  3134. 1:33:18several baskets, let's grow like that,
  3135. 1:33:20let's try to build as many bridges as
  3136. 1:33:22possible between B2B and B2C so that,
  3137. 1:33:23as you said, it's not too scattered.
  3138. 1:33:25And there you go, that was the choice,
  3139. 1:33:26and it happened little by little. It
  3140. 1:33:28wasn't a choice from the beginning, by
  3141. 1:33:29the way, it was opportunities here and
  3142. 1:33:30there. But right now, for instance, we
  3143. 1:33:32are thinking about partnering with
  3144. 1:33:33companies that only do B2C, saying
  3145. 1:33:35maybe Misteria will do a big B2B pole
  3146. 1:33:36and behind that we will have a partner
  3147. 1:33:38who will handle the whole B2C side. We
  3148. 1:33:40are still thinking about that. But yeah
  3149. 1:33:41, that's the big argument behind it.
  3150. 1:33:44Thank you Martin for everything that
  3151. 1:33:46was discussed in this episode. It's a
  3152. 1:33:48vast subject, a big subject, and also
  3153. 1:33:50super interesting to see how things
  3154. 1:33:52grow quickly. Looking forward to what's
  3155. 1:33:54next. Congratulations on what you have
  3156. 1:33:56built in just 2 years and for
  3157. 1:33:58responding to this wave of opportunity.
  3158. 1:34:00Uh, I loved this episode. And if you
  3159. 1:34:02enjoyed this episode as much as I
  3160. 1:34:04enjoyed hosting it, let us know as
  3161. 1:34:06always by sharing this video, leaving a
  3162. 1:34:08comment, or liking it. And then, if you
  3163. 1:34:11are on a streaming platform, your
  3164. 1:34:13favorite streaming platform, don't
  3165. 1:34:14forget to leave 5 stars on Apple
  3166. 1:34:16Podcast or here. Martin, I have one
  3167. 1:34:18last question for you that I ask all
  3168. 1:34:19our guests. Can you share the turning
  3169. 1:34:21point that made all the difference for
  3170. 1:34:23you, whether personally or
  3171. 1:34:25professionally, that you haven't shared
  3172. 1:34:27in this episode or perhaps never in any
  3173. 1:34:29interview? It could be a simple
  3174. 1:34:30sentence, an anecdote, something
  3175. 1:34:32positive or less positive. You have
  3176. 1:34:33free rein for your final words. Not
  3177. 1:34:36easy. Uh, a turning point that changed
  3178. 1:34:39a lot of things. Hmm. I'll have to
  3179. 1:34:44think a bit in terms of
  3180. 1:34:45entrepreneurship. What could have made
  3181. 1:34:47me take the leap? Yes, maybe I did have
  3182. 1:34:51one. I went to the United States, to
  3183. 1:34:53Texas. I went to Texas. So, I I come
  3184. 1:34:57from I come from Saint-Étienne. I'm
  3185. 1:34:59not really I'm not Parisian at all, and
  3186. 1:35:02I think that, especially outside of
  3187. 1:35:04Paris, French people don't stop.
  3188. 1:35:07Everyone says we're not good enough.
  3189. 1:35:08Well, no, that's not it. Actually,
  3190. 1:35:11teachers at school—I’m caricaturing
  3191. 1:35:12it a bit—I find don't give much hope.
  3192. 1:35:14You see, they tell you you'll have a
  3193. 1:35:16small job and that it's just fine. They
  3194. 1:35:19constantly tell you that bosses are
  3195. 1:35:20stupid, they keep saying it. I think
  3196. 1:35:21everything is just a bit negative. And
  3197. 1:35:23when you come from that environment,
  3198. 1:35:24honestly, you don't want to be an
  3199. 1:35:25entrepreneur, you don't want to get
  3200. 1:35:26started. They keep talking about people
  3201. 1:35:29who started companies and crashed, and
  3202. 1:35:30it's seen as shameful. They always talk
  3203. 1:35:32about the guy in your high school
  3204. 1:35:34making videos, and it's seen as
  3205. 1:35:35shameful. That's the feeling, we're
  3206. 1:35:37kind of stuck in that, you know. Paris
  3207. 1:35:38is a little less like that, but
  3208. 1:35:39otherwise in France, it's a bit gloomy.
  3209. 1:35:42And the turning point was when I went
  3210. 1:35:44to the US, I lived in Austin for a year
  3211. 1:35:46, and you actually see the polar
  3212. 1:35:47opposite where you can do whatever you
  3213. 1:35:49want. I get to school, and everyone has
  3214. 1:35:52started a company. When I say a company
  3215. 1:35:54, it’s small Instagram pages with
  3216. 1:35:5510,000 followers sometimes, or 5,000,
  3217. 1:35:57food pages, you know. But everyone has
  3218. 1:35:59a project in class; the teacher speaks,
  3219. 1:36:01and 100%of the students raise their
  3220. 1:36:02hands to participate, whereas in France
  3221. 1:36:04, it’s embarrassing to be the one who
  3222. 1:36:06participates. And you actually feel a
  3223. 1:36:09population that doesn't ask questions,
  3224. 1:36:10that wants to move forward, wants to do
  3225. 1:36:12things, and isn't there thinking: "Yes,
  3226. 1:36:14but is it really fair if I make money
  3227. 1:36:16while my neighbor earns a little less?"
  3228. 1:36:18No, everyone is pulled upward, you see.
  3229. 1:36:20And that really shocked me a bit. And
  3230. 1:36:22when I returned to France, I told
  3231. 1:36:23myself, that's also why I wanted to
  3232. 1:36:24start a business, I think. I told
  3233. 1:36:25myself, that's what I want to do. I
  3234. 1:36:26want to be around ambition. I want to
  3235. 1:36:28be with people who want to change
  3236. 1:36:30things a bit, who aren't ashamed to
  3237. 1:36:31talk about money, who don't ask
  3238. 1:36:33themselves a thousand questions, you
  3239. 1:36:35know, who talk to clients, who take
  3240. 1:36:37part, and so on. So that was a bit of a
  3241. 1:36:38turning point. If I had to give one
  3242. 1:36:40piece of advice, I'd tell everyone to
  3243. 1:36:41go live in the US for a year. There are
  3244. 1:36:42plenty of flaws in that country, but
  3245. 1:36:44they have one quality: they ask
  3246. 1:36:45themselves fewer questions before doing
  3247. 1:36:46things. And I think if we asked
  3248. 1:36:47ourselves fewer questions, if we all
  3249. 1:36:48took ourselves a little less seriously,
  3250. 1:36:50we'd get more done. And that's been a
  3251. 1:36:52big topic, and it bridges the gap with
  3252. 1:36:54school. So I was at HEC, and one day I
  3253. 1:36:56heard someone at the Lyon campus say
  3254. 1:36:58this. Everyone, someone asks in a
  3255. 1:37:00podcast, "What's the big advantage of
  3256. 1:37:02HEC?" Honestly, you never go to class,
  3257. 1:37:04you just party. So well, there's no
  3258. 1:37:07academic advantage, you don't really
  3259. 1:37:09learn much. Sorry if there are any HEC
  3260. 1:37:11people listening, but you don't learn
  3261. 1:37:13anything. You meet cool people, that's
  3262. 1:37:15for sure. Everyone tells you, "The
  3263. 1:37:17network, the network." Yes, but I don't
  3264. 1:37:19know if it's a game changer. However,
  3265. 1:37:21he was saying there is one strong point
  3266. 1:37:23: you're surrounded by ambitious people
  3267. 1:37:24every day, and you're surrounded by
  3268. 1:37:26teachers and a school that makes you—
  3269. 1:37:28I was going to say makes you believe,
  3270. 1:37:29no, makes you realize that anything is
  3271. 1:37:31possible. And without it being BS, they
  3272. 1:37:33tell you, "Oh, you want to start a
  3273. 1:37:34company that will...? Well yes, we have
  3274. 1:37:35students who've done it, so yeah, you
  3275. 1:37:36can do it.""Oh, you want to start a
  3276. 1:37:37company and have employees in the US?""
  3277. 1:37:39Yeah, go ahead, we'll help you with
  3278. 1:37:40some things." And that was a big
  3279. 1:37:41turning point too, that you're in a
  3280. 1:37:43school, an ecosystem where people don't
  3281. 1:37:45put barriers on themselves, you know.
  3282. 1:37:47Whereas I have friends who, you know,
  3283. 1:37:49see things and say, "Damn, that guy has
  3284. 1:37:51a Ferrari, damn, I'll never have that."
  3285. 1:37:54There isn't a single thing in life
  3286. 1:37:55where I tell myself, "I'll never have
  3287. 1:37:56that or I'll never do that." Being
  3288. 1:37:57president, maybe I'll do it, you know,
  3289. 1:37:58I have no idea. In any case, it's
  3290. 1:37:59possible. You say that everything is
  3291. 1:38:01possible and that's what Americans do;
  3292. 1:38:03they say everything is possible for
  3293. 1:38:04them. What a school like a top-tier one
  3294. 1:38:06brings a lot is ambition; they tell you
  3295. 1:38:08that everything is possible, which is
  3296. 1:38:09something we don't instill in young
  3297. 1:38:11people. We tell them, "Oh no, don't
  3298. 1:38:12make too much noise, don't speak too
  3299. 1:38:13loudly." And you see, he has a big
  3300. 1:38:14house. Yeah, but that doesn't interest
  3301. 1:38:16us; we'll never be like that. You see,
  3302. 1:38:17he studied in Paris. Yeah, but you
  3303. 1:38:19don't have that luck, you. People are
  3304. 1:38:21very much like that in France. They
  3305. 1:38:22talk to their kids, their students,
  3306. 1:38:23like that a lot. And, well, I think
  3307. 1:38:25it's a real breakthrough to understand
  3308. 1:38:27that, in fact, everything is possible.
  3309. 1:38:29Yes, building a consulting firm worth
  3310. 1:38:3010 billion is possible. It will be very
  3311. 1:38:31, very, very hard. The probability of
  3312. 1:38:32succeeding is low. Very, very low, but
  3313. 1:38:34it is possible. So you have to try to
  3314. 1:38:36find the solution, you have to try to
  3315. 1:38:37fight, meet people who have done it.
  3316. 1:38:39But there you go, it's a good wake-up
  3317. 1:38:41call to go to the US for a week or 10
  3318. 1:38:43days and see people who have ambition;
  3319. 1:38:44it changes a lot of things.
  3320. 1:38:46Thanks, Martin.
  3321. 1:38:47Great wake-up call.

About this transcript

This page contains the full transcript of Guide complet 2026 pour utiliser l’intelligence artificielle dans son entreprise by Le Déclic par Alec Henry, generated from the public captions YouTube serves with the video. The transcript has 22,814 words across 3,321 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to choose a caption track available on the video.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.