YT YouTube2Text

YouTube2Text → Transcripts → Я сделал х4 в прибыли с помощью Claude на Wildberries: Рассказываю как!

Я сделал х4 в прибыли с помощью Claude на Wildberries: Рассказываю как! — Transcript

by На Маркетплейсах с Лео Шевченко · 10,954 words · 1,774 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00Yes, ladies and gentlemen, hello
0:01everyone.
0:02I’m chilling, relaxing, having some tea
0:05, because today our broadcast is about
0:08how to automate all your work so you can
0:11drink tea while AI agents do the work
0:14for you.
0:15If it seems unrealistic or like all this
0:18is made up, you will see right now that
0:20this is actually already happening.
0:23That’s why today will be a very
0:24interesting broadcast. Its topic is: My
0:26account is managed by agents.
0:28Simply put, these are employees who work
0:31a bit better than your regular live
0:33staff, get paid less or don't get paid
0:35at all, and yet perform their tasks much
0:37more efficiently.
0:41Today I’ll show how the employees I
0:43created yesterday are working for me.
0:47I mean, I already have my own separate
0:49employees working on my platform, but we
0:51specifically created new ones via Claude
0:54yesterday as an experiment, and they
0:56have already started working today as we
0:58took over the accounts.
1:00I will show what they do, what they are
1:01capable of, and how it all looks in
1:03general. And all of this was done in a
1:04day.
1:05Again, if you have an understanding of
1:07how to do this, you can implement it all
1:08just as easily and quickly.
1:11Today, as usual, I’m with you, Lev
1:13Shevchenko.
1:15You all already know about me, so I
1:16won’t say anything.
1:18I will say that I have been trading on
1:21marketplaces for a long time, trading
1:23with varying degrees of success
1:24depending on what is happening in our
1:27market.
1:28So, we continue moving forward, we don’t
1:30give up, and we make sure everything
1:32works as well as possible.
1:34You probably often see in chats that
1:36Claude gets blocked. Let’s address right
1:39away that yes, Claude can be blocked.
1:42But, for instance, this account of ours
1:43, we have already gone through
1:45verification and confirmation.
1:47This account has been working for at
1:49least half a year now. We have one more
1:50account, and another one as well. They
1:53have been working for a month or two,
1:54and they are all still active.
1:56Why do we need so many? Because we use
1:58different ones for different projects.
2:00But what’s the point? Could mine get
2:02blocked? Yes, they could.
2:05At any moment they could, in principle,
2:07and well, I don't believe that it’s
2:08impossible, or perhaps I do believe it.
2:12But in any case, this is no reason not
2:14to learn AI and not to move into this
2:16space.
2:17Why?
2:19Because you can always play it safe and
2:20, for example, connect ChatGPT, because
2:22it gets blocked less often.
2:24But for me, I don't really like it; it
2:27feels dumber compared to Claude, and
2:29even with updates, it feels slower, so,
2:32well, you can connect it. I know that
2:35DeepSeek is currently developing
2:36something too. DeepSeek has also started
2:38working super fast, and they are
2:40developing a version that will support
2:42both code and everything else. And, in
2:44principle, I think it will be possible
2:45to switch to it later.
2:47Therefore, there will always be an
2:49alternative; you can connect different
2:51systems, switch between them, and, well
2:53, not studying AI right now is very
2:55strange.
2:56Actually, for me, when I entered the
2:58marketplace business, why did I start
2:59recording about them, making YouTube
3:01videos, and everything else? Because
3:02what did I like about them? Because
3:04everything on the marketplaces was
3:05constantly being updated. There wasn't a
3:07day, week, or month where things weren't
3:10being updated.
3:11I started selling on Wildberries back in
3:142016, and for 2-3 years everything was
3:16super, well, stable, let's say; changes
3:18happened once a month or every two
3:20months, but then, probably starting in
3:222021, they changed everything every week
3:25, expanded their staff of analysts and
3:27developers, adding more and more.
3:30And it was interesting for me to work
3:31there because every time there was
3:32something new; you are constantly
3:33learning, constantly in motion,
3:34constantly changing, and that was cool.
3:36Now, marketplaces are probably still
3:38making their various changes, but it's
3:40not as interesting anymore.
3:42All that was interesting in terms of
3:44updates is now in AI, because everything
3:46is constantly changing, everything is
3:48new, agents are appearing, code is
3:50appearing, MCPs, and everything else.
3:52And that is truly cool. And what is
3:54important is that AI for marketplaces is
3:56a miracle right now. Why? Because there
3:59is currently too much data on
4:00marketplaces, and it’s unclear—neither
4:03on Ozon, nor on Wildberries, and even on
4:05Yandex Market, it’s hard to manually
4:07gather all the reports to analyze sales
4:09volume or the exact profit for each SKU
4:11.
4:12It’s difficult to do all this; there are
4:14various services and auxiliary systems,
4:16but even they don’t always calculate
4:18everything accurately and correctly, so
4:19this whole thing needs to be kept up to
4:21date.
4:22That is precisely why AI is very
4:24well-suited for this. There is a
4:27question right here: someone had to
4:28register for Claude, and it says
4:30verification is required.
4:33Even with a residence permit from Serbia
4:34, it didn't work.
4:35We have gone through this; I have an
4:37acquaintance with a second passport, and
4:40just two days ago, I easily passed
4:42verification on one of my accounts. It
4:45is done as easily as getting an English
4:47visa; I was recently in England.
4:50You just download an app on your phone,
4:52open it, scan your passport with the
4:54phone, then scan your face, and that's
4:56it, the verification is complete.
4:59So, there is nothing complicated about
5:00it; anyone can do it.
5:02We have staff, for example, some are
5:04from other countries, and some of them
5:06go through verification for us if
5:08necessary.
5:11I think that soon a service will appear
5:13somewhere like Avito services that says
5:15, "We will handle the verification for
5:17your Cloud."
5:19That is also, in principle, possible. So
5:21, you shouldn't be afraid that the Cloud
5:23will be blocked.
5:24You can connect to another system and
5:26work with that instead. In principle,
5:28all data is transferred and copied over
5:30.
5:31Now I will tell you how to store keys,
5:34various accounts, and your skills on
5:37your own separate desktop.
5:40Soon, everything will not even be on
5:42your desktop, but separately in the
5:44cloud. Therefore, in principle, there
5:47absolutely shouldn't be any worries here
5:49. What will I show today?
5:52About the skills—what skills I already
5:54have and what skills I can eventually
5:56give to you, show you, and share with
5:58you.
5:59That's the first thing. Second, about
6:01the employees. I will show you the
6:03employees we created yesterday, how they
6:05are working today, what they can do, and
6:07most importantly, I will show how they
6:09create results together. And then I will
6:11explain how to build such a system for
6:13yourself.
6:14In short, it will be interesting.
6:16And at the end of the stream, of course
6:17, I will give you the skill I promised,
6:18the one for detecting where advertising
6:20budget is being wasted. What does it do
6:21? You can set it to run once a week, or
6:24every two days, or even every day to
6:26check all active campaigns from the last
6:287 days.
6:30"Active" means those that are running
6:31now or have finished within 7 days; it
6:33finds all campaigns with spending but no
6:35orders and ads for products with no
6:37stock, shows you the keywords eating up
6:39your budget, and calculates how much
6:41money can be recovered for each item and
6:43what to do first.
6:45Let's, for instance, run such a skill or
6:47such a command right here live on my
6:49personal account and I'll show you how
6:51it all works.
6:53By the way, for those of you who want to
6:55book a spot in our course, there is a "
6:56book a spot" button.
6:58This is our course on API/automation
7:00that we are launching specifically for
7:01sellers.
7:02I am not a programmer myself, nor am I a
7:04major AI expert, but I know everything
7:06you need to know and do with
7:07marketplaces from that perspective;
7:09we’ve long since moved past that, we’ve
7:11been doing this for six months, we’ve
7:13taken our lumps, gained so much super
7:15experience, and now we know, can do, and
7:17understand it all.
7:18I won't be talking about, I won't be
7:21telling you how to build your own, I
7:23don't know, any kind of software.
7:25I will talk about what needs to be done
7:28so that you understand and automate your
7:31work on marketplaces, which will really
7:34free up a lot of your time and allow you
7:37to give up some analytics services, some
7:40paid integrations, and, of course, even
7:42part of your staff, which is also
7:45something you can do.
7:47We have already moved beyond that point
7:49.
7:49So, let's move on and take a look at
7:52what is necessary.
7:55Here is our Anthropic, which is where we
7:57work on a regular basis, and here is
7:58what we have here.
8:00I won't be explaining the interface
8:01right now. For example, let's take a
8:03look at, uh, I still don't even know how
8:05to pronounce this word correctly.
8:07Or maybe the proper way is somehow, well
8:11, anyway, it’s recurring tasks.
8:15Let's see what kind of recurring tasks I
8:17have.
8:18For instance, I set a task for myself
8:20today at 11:30 PM to buy a black down
8:21jacket, there's one size, I need to pick
8:23it up.
8:24What else is there?
8:26The morning briefing is sent every
8:28morning for both WB and Ozon. Draft
8:31responses—this is in case I suddenly get
8:33one or two stars on my replies, I check
8:35what my AI is automatically sending for
8:37that, because the AI is already writing
8:39the replies, we don't write them anymore
8:41for both WB and Ozon. Well, this isn't a
8:44surprise to anyone, but in principle,
8:45all of this is already being implemented
8:47.
8:48And if, all of a sudden, there are
8:49ratings, or rather reviews with one or
8:51two stars, the AI first asks if it's
8:53suitable or not.
8:54CFO report, meaning a daily report for
8:56the accounts on both Ozon and WB,
8:58reporting on where we stand today
8:59specifically from a financial standpoint
9:01.
9:03Arkashka dump—once a week we get this
9:05kind of report for both WB and Ozon.
9:08Quality for the week—this is a full scan
9:11of reviews.
9:14Margin and prices—it looks at this based
9:15on the financial report, for which items
9:17or groups of items we need to raise
9:19prices.
9:20ABC and stock forecast—the same thing
9:23for analyzing the stock levels required
9:25for Ozon and WB.
9:27Analysis of the top three SKUs against
9:29competitors.
9:32There is a simple analysis here, and I
9:34also have a skill for a deeper analysis
9:36where we compare product listings via
9:38Wildberries to analyze them more
9:40thoroughly.
9:42This is done through MPstats using its
9:44data.
9:46Task repetition in the project, shelf
9:48reports, and all sorts of other tasks we
9:50have.
9:51We also have various entities, projects
9:53, that we work in.
9:54WB account, Ozon account, Yandex Market
9:56account, and Lamoda account as well. And
9:59all these accounts have their own skills
10:00, I'll show them a bit later. For
10:02example, let's take a look at the WB
10:04account. This is what the account looks
10:05like.
10:06There are instructions uploaded here,
10:08along with API keys for Wildberries:
10:10some for reading, others for reading and
10:13editing.
10:14There is context provided here as well.
10:16We have uploaded more than twenty skills
10:18just for WB, I think.
10:20There are recurring tasks. And, for
10:22example, here is what happened today.
10:24There was a morning briefing today,
10:27where a report like this was received.
10:30It is sent every morning.
10:33We look at the morning briefing for
10:3504/04 or 05/01. The whole account, 100
10:38listings, dips. Thermal leggings without
10:41fleece: 16 orders versus 21.5 yesterday
10:43.
10:44Women's oversized shirt: one order after
10:46one return. Turtleneck: four orders
10:48versus five and eight. Thermal leggings
10:51with fleece are marked as a dip only in
10:53the cart. Meaning the cart is showing a
10:55decline. This can also be investigated.
10:56Growth. Which SKUs showed growth? Such
10:59and such bodysuit, other thermal
11:01leggings, such and such swimsuit.
11:02Reasons for the other stages. Stock
11:04levels are normal for all eight SKUs;
11:06there is new negative feedback only for
11:07the thermal leggings without fleece. In
11:09short, it sends the entire report. What
11:11to do today? Organize an FB delivery for
11:13the thermal leggings. We already did
11:14that, by the way. Reply to one-star
11:16reviews, because I handle those, and I
11:19provide approval for you to reply to
11:22them.
11:23Check the positions and ad spend for the
11:25thermal leggings. Check the bodysuit
11:27listing and see if the shirt has dropped
11:28out of search results. Impressions are
11:31down 49%for some reason. Stop-list.
11:34FBO stock is less than 10 days for a
11:36couple of my SKUs.
11:38And the average order price changed by
11:40more than 10%compared to the 7-day
11:43average. Also need to check the
11:45stop-list, which is empty.
11:47And a dashboard is provided immediately
11:49. This is the dashboard, the report is
11:50sent this way. You can look at it like
11:52this, or you can expand it and take a
11:54closer look.
11:55And, for instance, the morning report.
11:57Everything from yesterday, 7-day
11:58averages, absolute values.
12:00Next is the top list of dips, which
12:02products have seen a decline.
12:04Top growth items, which products are
12:06starting to perform better and what
12:09happened.
12:11It includes the brand, the title, and
12:12the revenue. There’s nothing
12:14supernatural here, but it covers orders
12:16and sales, pulled daily and sent to you
12:19.
12:20Separately by brand, separately which
12:21products are going out of stock, and
12:23separately the funnel and charts for
12:24them.
12:25This is the report you wake up to every
12:27morning while drinking coffee or tea
12:29with lemon; you actually see and read it
12:31.
12:31What else is there?
12:34For example, a roast of the top 3
12:36product cards. This is also sent out
12:37once a month, once a week, or once a
12:39month. Today is the day when the ad
12:42campaign report arrived, and that’s what
12:44we see here.
12:46I decided to roast myself and look at
12:49three products from different categories
12:51.
12:52My long johns, some down jackets, and a
12:54bodysuit.
12:55Long johns, search position, best median
12:57, best median.
12:59So, my position is 59 versus 7.76, which
13:03is the average median for the comparable
13:06products my Claude decided to pick.
13:11In terms of price range, we are in the
13:12top 5.
13:13The same goes for down jackets and the
13:15bodysuit. And let's look right away, so
13:17we don't spend 3 hours reading what to
13:18consider, blah-blah-blah, let’s read the
13:20analytics immediately. You can also
13:22expand it. Here is the full analytics,
13:24and here, for example, let's look at the
13:25long johns.
13:27Opening them up. What do we see here?
13:29It’s compared directly via MPSTAT.
13:32Where I have deep analytics, we pull
13:33data from product card comparisons; it's
13:35even deeper and better there.
13:37There is this free one, well,
13:39conditionally free, because you still
13:41need access to MPSTAT, regardless of the
13:43tier, and you can analyze everything
13:46through it. So we see that I have 10
13:48photos in the carousel, but others have
13:5015, 11—so it turns out they have more
13:52than me. I have a video, I have rich
13:54content. Wow, it turns out I don't have
13:57rich content, although many others
14:01didn't have it either. Description
14:03characters, that is, the number of
14:05characters in the description.
14:06Just in case, the characteristics in the
14:09public card—it's missing some features.
14:12By the way, I have a skill for checking
14:13characteristics, I’ll show it later.
14:15Next, 4.8 rating. The others are
14:17basically the same or even a bit worse,
14:19mine is highlighted in yellow. Reviews—
14:21well, guys, I have 3,300, although why
14:233,300? I could have merged them and it
14:25would be six. This also needs to be
14:27looked at. Share of negative reviews is
14:2920%, which is okay. Average search
14:30position is fifty-ninth. FBO and FBS
14:33stock levels, revenue for October. Next
14:35is price, sort of an analysis of what to
14:37fix. It describes that there is not
14:39enough stock, others have more.
14:41We only have four sizes in stock, while
14:44the top five have 14.
14:47Therefore, we can add sizes here and
14:50grow sales volume.
14:52We don't have enough specs, because it
14:54turns out they need to be expanded.
14:55Reviews, obviously, photos in the
14:57carousel, description, and everything
14:58else.
14:59Customer tags and reviews by IP. Here
15:02you can also see what we have, what
15:04mentions there were, and what needs to
15:07be fixed.
15:08And here separately, after we have fixed
15:11this, is what still needs to be refined
15:13. The cover, background, and everything
15:15else. Interesting? Yes, interesting.
15:17This is the kind of analysis done; it
15:19can be done more or less often, as you
15:20wish. For me, it's sent once a month for
15:22the top 3 products, and I take a look.
15:24You can do it once a week.
15:25For example, a weekly ad campaign budget
15:27drain.
15:28We open the ad campaign, or rather, look
15:30at
15:32the week and see what happened here. Top
15:355 problems. What do we have here? What
15:38is our problem?
15:41Check which size has run out.
15:43Here, this one has zero orders, zero,
15:45and not zero stock, because it all went
15:48out of stock.
15:50Here, zero orders, no stock. For some
15:52reason, stock also ran out, need to
15:55restock.
15:56The T-shirt bodysuit has a 35%ACOS,
15:58needs to be lowered and worked on.
16:01And here, it basically shows everything
16:03: why there was a drain and what needs
16:04to be done. The same audit in a file.
16:08You can do it in a spreadsheet, you can
16:10do it separately as an image, you can do
16:11it however you like.
16:13We open recommendations, top campaign
16:15drains, top clusters, negative phrases,
16:18and everything else is there too. It's
16:20very convenient.
16:22By the way, let's run some skill right
16:25now. Let's go into an advertising
16:27campaign. Let's say, well, let's take
16:28long johns, for example, if we're
16:29talking about them. Let's go to long
16:31johns. So we open it, let's say, how you
16:33act every morning, right, or every day.
16:35You open the ad campaign, you look, you
16:37have these long johns. Opened it, looked
16:39at the specs. Everything looks kind of
16:41red here, it's gotten worse.
16:43It was there at the beginning, let's go
16:46back, look at costs of 12,600, 199
16:48orders, but these aren't paid orders.
16:51You understand that, right? ACOS is 6.46
16:53, average position is tenth, and
16:55everything else has passed.
16:58Then we drill down into the campaign and
16:59see what we have here.
17:01And it seems like, well, we saw that
17:03it's red.
17:04Oh, the statistics immediately.
17:05We saw that it's in the red.
17:09But how much worse or better has it
17:10gotten? How much, you know, has actually
17:13changed? And what is working and what
17:14isn't? Because in this case, there's
17:16search, recommendations, and everything
17:17in between. There are many clusters.
17:19There are quite a lot of clusters,
17:20honestly. There are six pages with 20
17:22items each. Plus, some of them are in
17:25the excluded list. How correct or
17:26incorrect is all of this?
17:28And analyzing it deeply for every
17:30product every day, or even once a week,
17:31well, that's hardly possible.
17:33But all of this can be delegated to your
17:38agent or given to Claude.
17:41It is important, however, that the
17:42necessary skills are configured and
17:44developed for him.
17:45Let's assign him a task now and go take
17:47a look at what skills exist.
17:50Let's copy the SKU, add it, and simply
17:53dictate to him what needs to be done.
18:00Analyze the advertising campaign for
18:02this SKU.
18:05And analyze not just the currently
18:07active campaign, but also any campaigns
18:10that were running this week, if there
18:12were any.
18:15Write down which negative keywords can
18:20be restored, because there is
18:26potential growth there, meaning money
18:29can be made from them, and also list the
18:32top 10 keywords generating the most
18:34profit.
18:37And advise whether we should increase
18:39the bid or if we are at our limit, based
18:42on the profit for this SKU.
18:46In short, this is the kind of speech
18:48being drafted right now.
18:50I hope it recorded. Yes, everything
18:51recorded. There will be mistakes and
18:53slips of the tongue here, but it doesn't
18:54matter at all. And let's start, what
18:56else should we add?
18:57Let's add: "Prepare a report on the top
19:0010 SKUs in a nice table."
19:03Let's add that and click "Act." That's
19:06it. From here, Claude acts independently
19:08, doing everything, measuring everything
19:10, and checking everything.
19:12I'll show you what skills we have built
19:16up over time.
19:18This is our Claude, so to speak.
19:21I'll show you this right now.
19:23Here, for example, are the skills we
19:25have active now.
19:27We have general skills, and others for
19:30Yandex, Lamoda, Ozon, and Wildberries.
19:32Well, you could expand "General" and
19:34look, but that won't give you much
19:35insight because it's all skills here.
19:37That's how many general skills there are
19:39. That's how many there are for WB.
19:42Well, let's open the description instead
19:43and see what skills exist and what they
19:45are actually for. Word, and here we are
19:47looking directly at them. You have, well
19:50, we have 63 skills: 28 for WB, 12 for
19:52Ozon, 3 for Yandex Market, 3 for Lamoda
19:55, and 17 general ones. Let's look at the
19:57skills, for example, for Wildberries,
19:58what they are.
20:00Morning briefing, that is, the one I
20:01showed you, where yesterday's sales
20:02growth was asked about compared to the
20:03week and at what stage of the funnel.
20:06Moving on. WB Analysis. Full dashboard
20:08analysis for sales, advertising, and
20:10stock.
20:11WB Analytics.
20:12Funnel by product card, search queries,
20:14stock history, all reports are displayed
20:16as well.
20:17WB Traffic.
20:19This covers commissions, logistics, or
20:21rather, rates and storage, for which a
20:23final weekly report is sent to you.
20:26Moving on. Smooth price increases.
20:29That is, we can set a skill to
20:32automatically raise the price of our
20:34product after a promotion, say, by plus
20:372%.
20:39Or we are preparing for a new promotion
20:40, want our median price to be higher, it
20:42rises by 2%over two weeks, and we are
20:43ready to enter this promotion.
20:46Next is ABC-analysis of the assortment
20:48for 30-60 days with a forecast; I will
20:49show it to you now.
20:51Product seasonality, which month is the
20:52peak, when to purchase, when to ship,
20:54and when to sell is mandatory.
20:56Ad management, bids, pause, launch—we
20:59have a separate skill for this as well.
21:03Then advertising statistics: CPC, CTR,
21:06orders, clusters.
21:08This is also necessary to correctly
21:10understand how much was spent on
21:11advertising and what the profit from all
21:13this will be.
21:16Next is searching for budget drain,
21:17specifically how the budget is leaking
21:19and which key clusters and words are not
21:21working.
21:24After that, product card analysis with
21:25top competitors via MPStats for free.
21:29In some places there is deep analysis,
21:30here is a full SEO description analysis
21:32.
21:33Next is reading reviews and questions,
21:35plus answering them.
21:38Such skills.
21:39Drafting replies, just as I mentioned
21:41earlier.
21:43Copyright complaints, the same thing
21:44exists; you can watch competitors '
21:46videos and then take them to analyze and
21:48possibly create similar videos for
21:50yourself.
21:52And then, growing niches.
21:55You can also separately use the basis
21:58for API requests; we even have skills
22:00that allow you to parse data from WB
22:03faster and more easily if your tokens
22:05are running out.
22:08Promotion calendar, of course.
22:11Unfilled specifications—there is even a
22:13skill that analyzes how poorly your
22:14product's specifications are filled and
22:16what can be fixed.
22:18It can even analyze product groupings (
22:20skleiki), which products need to be
22:21added to a group, and disassembling
22:23groups, what to ungroup, and where it
22:25will be better.
22:27Search rankings, including a skill that
22:29can monitor your product’s search
22:31queries every day and then see exactly
22:33which queries you’ve lost or which
22:35positions you’ve lost over the week.
22:38And all of this will be highlighted in
22:40some kind of top list, not just as a
22:42huge wall of text.
22:43There you go. Shipments, supply plans,
22:45where to ship, what to ship, FBS, FBW.
22:48Next, assembly tasks for FBS, how to
22:50pack, what to pack, picking list. Even
22:52that is included.
22:54WB chat, replying to, uh, responses,
22:55working with competitors, oh god, with
22:57competitors, with customers. There you
22:59go. And there’s a pre-trial claim, even
23:01that’s here. How to write a pre-trial
23:03claim if, uh, WB isn't paying us our
23:06money.
23:07Plus, there are other skills, other
23:10skills for general things, for analysis
23:13, for, well, a million things, a million
23:15things that you can use.
23:19It’s taking a while to think here, it’s
23:21analyzing the advertising campaign. Let
23:23me show you what else there is. These
23:25are the reviews collected over the week
23:27.
23:28The account has improved, 93 reviews, an
23:30average rating of 4.81, it was worse,
23:32now it's like this.
23:33Some products received bad reviews, some
23:36good. And all of this, all of this, all
23:39of this, has been pulled in right here.
23:42And right here it has prepared what we
23:45wanted, our report on long johns is
23:48ready.
23:49This is what it’s telling us.
23:51I’ll analyze ads by SKU for the week,
23:52active running campaigns, negative
23:54keywords, top 10 keys, profits for, well
23:56, such and such.
23:58Two campaigns were running this week,
23:59one and the second, one on CPC, the
24:01other on CPM.
24:02The CPM campaign has been paused since
24:04October 15th and is not included in the
24:06calculation. And this is a third
24:08campaign. Okay. Our bids aren't at the
24:10limit, this is an estimate, not a figure
24:12from the API. I calculated it based on
24:14the sales report with a cost of 270
24:15rubles. With 5%taxes plus 6%VAT. On one
24:19order, before advertising, about 195
24:21rubles remain.
24:23So this is the limit we can allocate for
24:25advertising orders. Right now it’s 118
24:28rubles. So 60 percent of the margin is
24:30being used.
24:32The SKU profit after ads from October
24:3520th to November 4th is about 15,000,
24:3815,000.
24:39We aren’t touching the price. The main
24:40constraint is the redemption rate, well
24:42, a 50%redemption rate is absolutely
24:44normal for clothing. But it’s also
24:46highlighting this for us here. Top 10
24:48keywords by profit, uh.
24:51together they generated 50 orders, with
24:53a cost per order of 82 rubles. , profit
24:55after advertising is about 5,500 rubles
24:57.
24:58And here are those orders, or rather the
25:00rings, oh my god, he showed the keywords
25:02.
25:03He also processed the negative keywords
25:05that are worth returning.
25:06Uh, one, second, third, why return them
25:08, and how to return them. And he made a
25:10special table. Let's open it and take a
25:12look, too. So, everything isn't just
25:14done randomly; it's all done
25:16specifically with data and everything
25:18else. For example, the top 10 keys, what
25:21they were, and why they are like that.
25:25Next, the negative keywords that exist
25:27now and need to be returned. He did it
25:30right away. And I can give the command
25:32now—return those negative keywords you
25:34identified, and he will return them
25:35himself. So, you won't need to log in;
25:37he will return them and do everything
25:38himself. You don't even have to log into
25:40your WB account.
25:42All the week's clusters, unit economics
25:45separately for all of this.
25:47And there's the methodology further down
25:48.
25:50So, there you go, it's that easy and
25:52simple; Claude, or any other agent
25:55capable of making changes via your API,
25:58can do everything for you, set
26:01everything up, and provide you with such
26:04detailed, specific analyses.
26:07What else happened this week, for
26:10example? Right now I have Zoom open here
26:13, and everything else.
26:14Not very convenient, but okay.
26:17Let's do it this way.
26:19For instance, quality by reviews for the
26:21week—we talked about this—ABC analysis.
26:23Well, let's take a quick look at the ABC
26:24as well.
26:25Today he sent an ABC analysis of the
26:28items: category A has only 11 units, B
26:30has 15, and there are 389 in category C.
26:33Those are items that aren't selling or
26:35are just listed for testing via FBS.
26:40Here is a complete breakdown of
26:41everything in the risk zone, how many
26:42items there are, and which ones have no
26:44sales. And he prepares a report just
26:46like this one afterward. Again, this is
26:48all because there are skills that
26:49specifically categorize everything.
26:52Separate ABC for sales, a 30-60 day
26:54forecast, a 90-day outlook, seasonality
26:56by item—it's all shown.
26:59So you even understand now when each
27:01product will see a spike and when a
27:04decline.
27:06There is even a table where it looks at
27:09each item and shows you when your season
27:11starts and ends.
27:13For example, thermal leggings start in
27:15October, November is bigger, October is
27:17even bigger, in January there's a
27:18decline but they still sell, they sell
27:20in February too, but March, April, and
27:22May are already, well, they perform
27:23worse. The same goes for every product
27:25and item; everything is detailed and
27:27shown. Cool, right? Of course, it's cool
27:29.
27:30And AI can easily do all of this for you
27:32if you understand it and know how to do
27:35it.
27:36But that's not all, that's not all yet.
27:39Now I will show you something else
27:40interesting.
27:42Our course, which starts this coming
27:43Monday, will be about AI for sellers,
27:45where we will explain everything to you
27:47.
27:48That is, we will tell you, show you, set
27:50things up, and explain how to configure
27:52a VPN so it all works, and how to
27:54connect both Claude and ChatGPT, and
27:56install them on your computer.
27:58Furthermore, how to pay for it all
28:00correctly so that nothing gets banned.
28:03Then we'll sort it out; let's get
28:05everything ready before Monday. In short
28:07, the AI department course is suitable
28:09for WB, Ozon, and Yandex Market.
28:12It's about employees who don't get sick
28:13, don't quit, and don't lie about the
28:15numbers. I’ll show you more about this
28:16right now.
28:18In 4 weeks, just one month, you will
28:20build your own AI department for your
28:22Wildberries or Ozon account, or both,
28:24and it will handle your daily routine
28:26and help you significantly as an analyst
28:29, manager, and logistician, helping to
28:31fine-tune all your processes.
28:34We set everything up together during
28:37live sessions; there's no need to
28:39program, and no knowledge of HTML or
28:42anything like that is required; the
28:44knowledge you already have about
28:47marketplaces is perfectly enough to
28:49transfer it over to the AI, and you will
28:52succeed.
28:54What will change in your account?
28:55Right now you do everything yourself,
28:58but after the course you will be
28:59receiving ready-made reports, ready-made
29:02actions, and will be working with this
29:04data.
29:06It's 4 weeks: first we start, then we
29:08handle the numbers, we analyze
29:10everything, consolidate all analytics
29:12and finances, including ad spend, and
29:14you will have ready-made skills that we
29:16provide—meaning all the skills I showed
29:18you, over 50 of them, we give you
29:20completely for free as part of this
29:22course.
29:24Plus, I will tell you and show you how
29:26to write prompts, because a prompt is
29:28also an important thing that you need to
29:30learn to phrase correctly, and a part of
29:33your AI's success certainly depends on
29:35it.
29:37In short, we first teach you how to work
29:40with AI, then we give you prompts and
29:43skills, you analyze all your sales, and
29:45then you work on your product card,
29:48maybe even using GPT to create great
29:50photos, cool infographics, and
29:52everything else for it.
29:56After that, we set up a special
29:57department with you, I'll show it to you
29:59now, and then we'll refine this whole
30:00process.
30:01So, in 4 weeks you will get everything I
30:03showed you; that’s the first part, and
30:05now I’ll show you the second part.
30:07A bit later, actually, let me show you
30:09the part that will be on autopilot, how
30:11employees send you everything and do it
30:13all for you.
30:14You set up the department during the
30:15course, not after. This means all live
30:17sessions are designed so that during or
30:19right after them, you can calmly sit
30:20down, set everything up, and get it done
30:22. Everything is very accessible and easy
30:24to understand. I know how to explain
30:27things, we know how to teach, and the
30:29entire methodology is built around this
30:31, so basically, whether you are 15 or 60
30:34years old, don't worry, you can learn
30:36everything. We have an employee who is
30:3957 years old.
30:41Even he understood everything, figured
30:43out the accounting, and, in principle,
30:45everything is absolutely fine. Who needs
30:47the course and who doesn't. Well, I
30:49mentioned last time, it’s suitable for
30:51all Wildberries sellers who have live
30:53sales, who have reports, and who have
30:55actual sales figures. You need it. If
30:58you don't have any sales yet, it might
30:59be a bit too early for you.
31:01First, get on the marketplace, try it
31:03out, get used to it, and then go to the
31:05AI course.
31:07Although, in reality, AI now, I think,
31:09is something that will be a must-have
31:11everywhere.
31:12And humanity won't be able to live any
31:14other way. I’m afraid that in a couple
31:16of years, AI will disrupt everything. So
31:18there. But you absolutely need to engage
31:19with this now. This is exactly what
31:21needs to be done.
31:22What else? Cost. Regarding the cost.
31:26The price is 59,190—it’s inexpensive; it
31:29covers the month, plus skills, plus
31:31prompts, plus full live streams, plus
31:33recorded lessons.
31:35And we provide the entire methodology on
31:37how to properly create an absolutely
31:39autonomous department for your
31:41marketplaces.
31:44There are installment plans, card
31:45payments, and bank account transfers
31:47available.
31:48All three methods are available. Again,
31:50they will become active for you once you
31:52click the "Book a spot" button.
31:54There you go. That makes five live
31:55sessions plus workshops, four program
31:57modules, a chat with an expert, an IT
31:58platform, a personalized certificate,
32:00and everything else. All of this will be
32:02provided to you.
32:03Most importantly, in a month, you will
32:05have a concrete, ready-made
32:07understanding of how to do everything,
32:09and you will have your department set up
32:11, as well as a specific chat in Claude
32:13programmed with our skills that are
32:15already trained, refined, digitized, and
32:17all of this will be yours.
32:20Let’s go, and I will show you our
32:21employees. These are the employees we
32:23created yesterday.
32:25Here I have a startup analyst, warehouse
32:27, logistics, CFO, ad manager, and
32:29customer care.
32:31And let's say, let's take a look at what
32:33they wrote together today.
32:35So, they sent a summary.
32:38Yesterday, 50 orders worth 100,000
32:40rubles. Crucially, the main ring is
32:42running out.
32:44Biker shorts returned, five units after
32:46hitting zero.
32:48What risks exist and which ad campaigns
32:50are running. Next, a decision is needed
32:52. Biker shorts, ad on pause, enable it
32:55or not. ACOS for 7 days is 10%, target
32:58is 15%, so, well, basically it's fine.
33:0125 orders for 30,000 rubles. for 3,000
33:03rubles. in advertising. Stock for 204
33:05days. Meaning, a lot of them have been
33:07shipped. What should I do?
33:09I can click "yes" and it will turn on,
33:11and I'll get a message saying the
33:13campaign has started.
33:15Let's look further.
33:17This was sent to me by the same ad
33:19manager. Stock 005 is at risk because
33:23one item has less than 7 days left. Six
33:26pieces still need to be ordered from the
33:27factory.
33:29The main ring stock has 3 days left.
33:32Risks: sizes will start to drop out from
33:33the 7th. First 4XL, then 3XL, and then
33:36XL. A decision is needed. Confirm
33:38whether the reorder was sent to the
33:39factory.
33:40So even, you see, this is what is
33:42actually happening in reality.
33:44He analyzed it and, well, we will also
33:47give you special prompts and skills,
33:48plus a special instruction on how to
33:50create such agents for yourself so they
33:53can show you things like this. Meaning
33:55he is warning me that my stock is
33:57running out. Did I order from the
33:59factory? And I have to confirm to him
34:00whether I ordered or not, so he doesn't
34:02worry or stays worried.
34:03Then socks, two ad campaigns for 4,000
34:05rubles. zero orders in 7 days. Both are
34:07already on pause, and we will leave them
34:08that way.
34:09So, I had some ad campaigns for New
34:12Year's socks.
34:14He decided to put both on pause.
34:17For the white tank top bodysuit, ACOS is
34:1949%, sleeve bodysuit 6.4%, orders minus
34:21day is high, we can buy more traffic.
34:24Meaning, we can increase the budget, for
34:26example.
34:27And here he is telling me about the
34:29white tank top bodysuit. Need a decision
34:30. They're asking me about an 8.3 ruble
34:33bid. 10 rubles. It can be raised per
34:36click. Current ACOS is 49%, target is 15
34:39%. 26 orders in 7 days. 407 days of
34:42stock.
34:43So, we can raise the cost per click
34:46under the CPC model to 10 rubles.
34:49At the same time, I'll be within my
34:51target ACOS, and the traffic volume will
34:54increase.
34:55I'll click yes, it will go ahead and set
34:57everything up itself.
34:59Next, long-sleeve bodysuit. Same thing
35:01for the long-sleeve bodysuit, seven
35:03negative keywords. I think I confirmed
35:05the decision here that there are seven
35:06negative phrases that don't suit us.
35:09Women's underwear, women's winter
35:10clothing 25, women's items from Turkey,
35:12and others.
35:14I confirmed them, and that I included
35:15the bonus. I think I excluded the trust
35:17.
35:19Then the ad manager did the same for the
35:22long johns; we had some dialogue about
35:25it.
35:26The CFO sent his ad report for all SKUs
35:30.
35:32For some bows, he sent a request to
35:35raise the price because people are
35:38buying them, or because they are at
35:40minus 11 rubles.
35:42At purchase, but if we raise the price,
35:44it will be in the black. I say let's
35:46raise it, let's raise it.
35:48And he raised it, I think, to...it was
35:50somewhere further on.
35:53Raised it to
35:56plus something there. We'll find it now
35:57, alright. Next, the long johns shipment
36:00, checked it, reconfigured the ad again
36:02because we said we were shipping it, or
36:04where is it here?
36:06Ah, I wrote to him that the ad is in the
36:08shipment. He says let's turn on ads for
36:10the long johns.
36:11He saw that the stock is low, 35 pieces
36:14for 2 days, and currently the ads are
36:18good, 6%, but considering it will be
36:21spent, the long johns will be ordered.
36:25And considering the position is good,
36:26they will be ordered anyway.
36:28So he says let's pause the ad campaign
36:30so we don't waste money, since there's
36:31no more stock.
36:34I ask, why didn't you look at the
36:35analytics?
36:37It turned out I hadn't added the
36:39shipment analysis to his access
36:42yesterday.
36:44Now I added it, he checked, and indeed,
36:47saw that the product, or rather the
36:49goods are in shipment, and turned the ad
36:52campaign back on.
36:55What else was there? Long johns without
36:57fleece, I mentioned that. And about the
36:58bow. Changed the price to 1,500.
37:01He says, okay, changed it, switched the
37:02price, and after, it turns out, the
37:04price went up. Price from 1,300 to 1,500
37:07, 50%discount. So, the price for the
37:09buyer has changed from 700 to 750 rubles
37:11. Regarding profit, it was -11 rubles.
37:13Now it has become +5 rubles. This is a
37:15recalculation based on the initial plan
37:18of 1,610.
37:19There you go.
37:20And in a week, when the financial report
37:22is out on Tuesday, it will show me
37:23exactly what happened with this product
37:25, how much better or worse it got, and
37:27will send me the full analysis.
37:30What else? Click approved the biker
37:32shorts, so the ad campaign is now active
37:34.
37:34Let's take a look, there were three ad
37:36campaigns, let's check them all at once
37:37.
37:38Okay, so for the biker shorts, Click
37:40increased the ACOS, or rather, Click
37:42raised the bid.
37:45Next, it approved the white tank top
37:48bodysuit, increased the bid, and also
37:51raised the bid for the turtleneck
37:54bodysuit.
37:56And there they are, well, it must have
37:58been enabled. It enabled it itself.
38:00Let's say we turn off the biker shorts
38:03campaign, then go back to Claude—or
38:06rather, not to Claude, but back to my
38:08assistants.
38:11And, say, find the biker shorts ad
38:13campaign somewhere and tell it to enable
38:17it again.
38:19Okay, let's scroll up and look for where
38:21it asked us.
38:23And this all comes in throughout the day
38:24for the new—these are long johns, we
38:25already mentioned that.
38:27I connected this new account yesterday,
38:29did everything, and now it’s all running
38:31today. Cool, right? Of course it’s cool
38:33. This is a huge help.
38:37I already connected it here, right?
38:39Let's do this.
38:42We'll call up this ad agent and write to
38:45it: enable the campaign for, let's see,
38:49the biker shorts. Let's see if it
38:52understands or not. Now I'll show you
38:53how everything works for us. Ours is
38:55much cooler, of course, but overall for
38:57you, well, this is just what we put
38:59together literally yesterday evening.
39:03That is, you can add other employees,
39:04and you can add a CEO who will monitor
39:06all this, or a project manager, and
39:08everything else.
39:10So, somewhere in there, let's see,
39:11pam-pam-pam, there are two ad campaigns
39:13for the biker shorts.
39:15One is already working, over 7 days,
39:17pam-pam-pam, I suggest enabling the
39:18second one. Well, it apparently doesn't
39:20see that the second one isn't enabled.
39:22I'll tell it now. I could use voice, but
39:24fine, I'll just say it this way. I
39:25disabled the second campaign.
39:28Enable it again.
39:30Well, you can even write it like that,
39:32it understands everything. And it's
39:33suggesting I turn on the second ad
39:34campaign right away, but I don't need
39:36that; I want to enable the campaign that
39:37I turned off.
39:38Now he will go, take a look, and, uh,
39:39decide what's next.
39:41Look, uh, he even has this. Pam-pam-pam
39:46, cycling shorts, click yes. Aha, well
39:48okay, that means they are there. I'm
39:50clicking yes.
39:52And he went and turned on some
39:53advertising campaign there.
39:55Click-click, click-click, done.
39:58Let's refresh.
39:59Let's take a look right away.
40:01Checking active ones one more time. Not
40:03yet, okay, we'll wait, we won't even do
40:05anything. Although, what does he say?
40:07Error. Ah, the budget isn't enough to
40:09start the campaign, I see.
40:10Ah, I see, I see. Then I can say, oh man
40:14, okay, let's just do it.
40:16Ah, alright, I don't really want to top
40:18it up, uh, after all.
40:20Turn on another one.
40:25Ad campaign for cycling shorts.
40:28The budget is insufficient, so he can't
40:30launch it, and he's telling me about it
40:32. I could go and top it up later,
40:34although, in principle, no, I don't
40:35think they can top it up, or maybe they
40:36can. In general, everything can be
40:38configured too. Pause the ad campaign.
40:42Oh my gosh, in English, let's try to
40:44write in, in, in Russian, in English. Ad
40:46campaign for long johns.
40:49I wonder if he'll understand if I write
40:52to him in English. Okay, understood,
40:54pausing the ad campaign for long johns.
40:56So, you can even speak in English, he
40:57understands everything, he knows how to
40:59handle everything.
41:00Oh my gosh, non-brushed, two yeses, yes
41:03, yes, yes, do it.
41:05He asks a lot of questions. Well, that's
41:06right, so as not to make a mistake.
41:09Approved, done, pause, ad campaign is
41:11paused. Opening, checking, refreshing,
41:15and that's it. At least it worked here,
41:17he paused this one.
41:19Let's suggest that he should change it
41:21back.
41:23Change it back, let's see if he guesses
41:25it or not.
41:27Set up the agents, and now product cards
41:28will be added, everything is being
41:29pulled in automatically for all of them
41:31.
41:31Of course, for all of them, naturally,
41:33everything is pulled in automatically,
41:34it's like, well, you won't need to do
41:36anything for sure.
41:38Yes, for the non-brushed long johns,
41:40it's on pause, yes, return it.
41:43Executing, done, refreshing. That's it,
41:46there you go, he returned it. If a
41:47laptop has low memory, will it not work
41:49for installing the program? Ah, it
41:50doesn't really eat up memory at all, I
41:52think it's like 200 MB, something like
41:54that.
41:55Well, it shouldn't be an issue, but just
41:57in case, check it, I don't know for sure
41:59. Will you show how to make Telegram
42:00bots in the course? Of course, yes.
42:02These are the kinds of bots we will
42:04create, and you will be able to reach
42:06out to each one. Take, for example, a
42:08logistics manager.
42:09Let's ask him, the logistics manager. I
42:11think you don't even need to tag him
42:13here.
42:15How many fleece-lined long johns do we
42:19need until 12/31?
42:23Of 2026, so we have enough at the
42:30current sales rate, and...
42:38Let's see if he can handle it or not.
42:41No need to address him here either, he
42:42understands on his own because he is the
42:44only one here. Let's ask the other one
42:45for now, okay. Which product is the best
42:50in terms of profit?
42:53Let's ask this one for now. Right.
42:55Fleece-lined long johns until 12/31.
42:58Currently selling 3.3 per day. These are
43:00the fleece-lined ones. 48 orders, 69 in
43:02stock. This will last until the 31st, 56
43:05days remaining. At the current rate, we
43:06need 185. Look, isn't that just perfect
43:08? He showed you, calculated it, and did
43:10everything. You won't have to think
43:12about it either, and everything will be
43:14ready for you. Let's see what the
43:16analytics say. The best for profit is
43:18long johns without fleece.
43:20102 pieces, pam-pam-pam. Next come the
43:22bodysuits, followed by bodysuits.
43:24The long johns without fleece give the
43:27most per unit, 537 rubles.
43:29For one single piece.
43:31Uh, for both down jackets,
43:32param-param-pam. And all this, just
43:34imagine, let's bring every question to
43:36the bot.
43:37Mm.
43:40Which product had the best reviews the
43:46day before yesterday?
43:51So, all the questions you have in your
43:53head, you can ask this employee of yours
43:55, and he does it all. It's not that
43:57Indians are sitting there chatting with
43:59me, no, or my staff, no, this is all
44:01really, uh, like, well, doing it, and
44:03very fast, and it does it all, day
44:05before yesterday 11/03. Well, even any
44:07of my employees wouldn't have been able
44:09to write that so quickly. Don't you
44:10agree? So, there was one review with the
44:13text "bows on a headband," five stars.
44:15They write it's very fluffy and bright,
44:17bought it for a morning party, they
44:19praise it and param-param-param. That's
44:21it, for each of your questions, there
44:23can be an answer like this. Cool? Of
44:26course it's cool. It is very cool, and
44:28we will tell you how to do all this. Is
44:30this a competitive advantage?
44:32It is a super wild advantage that will
44:35give you, uh, well, kind of a huge
44:37breakthrough, distance from your
44:40competitors, because naturally some
44:42already possess this, but I think that
44:45right now, active and correct use of AI
44:47...Out of the entire flow of sellers,
44:50the volume of sellers, I think maybe 0.1
44:52%...you will be among those who enter
44:55the 1%of those who know how to use it.
44:58And if you have this kind of
45:00understanding, this knowledge, of course
45:02, it will be much easier for you to
45:04manage your sales and work with them. I
45:06think you understand everything here.
45:09One SKU, one chat. Actually, one chat
45:11for everything—I will talk about that
45:12later in the course.
45:14RNP isn't needed, RNP is needed, it is
45:17needed too. Let me show you quickly.
45:20Let’s go into another one of our
45:22entities.
45:23It's called our dashboard.
45:26Here we have the Mercatus dashboard
45:28where everything you need is gathered in
45:31one place. This little person is Claude
45:34.
45:35Let's go straight here into Dinel. We
45:37have marketplaces here, as well as our
45:39own websites, agents, and disputes.
45:41Let's look at the marketplaces.
45:43There's WB, Ozon. And here we have a
45:46summary for today, an RNP for today
45:48that's already ready, and everything is
45:50here for you.
45:51So this is our dashboard, which we made
45:53by weeks, a separate report.
45:56It immediately shows the dynamics, how
45:58much, what, where, and what for.
46:00Everything is shown here according to
46:01the dynamics.
46:03Shipments that are needed, unit
46:04economics, which products, product
46:06reviews, including compensations—yes, we
46:08track them and monitor them too.
46:12We opened compensations; there are
46:13moments where we can get reimbursed and
46:15how much money can be claimed.
46:18Everything is separate for advertising.
46:20There’s a bidder, a repricer, and
46:22everything else already.
46:23This is basically the bomb here. Whoever
46:25has this will be far ahead of everyone
46:27else.
46:28There is a separate analysis for
46:30competitors, ABC analysis, PnL analysis
46:32, and everything else—it's all already
46:33there. Plus, in this dashboard, we have
46:35information for wholesale and for your
46:37own website. Plus, there are disputes;
46:39we write pre-trial claims there as well
46:41. Please, you can do all of this right
46:43here.
46:44And there are also various agents who
46:47write there—analysts, financial experts
46:49, and everyone else who reports every
46:52day, sets tasks, and does all this.
46:55This is very cool and very awesome.
46:57This can also be done with us, but it’s
46:59done on a different plan. For those who
47:01already have a big, working business, we
47:04have just such a dashboard where you can
47:08create your own department. A department
47:10and a dashboard that will generate
47:12everything for you in one place. It’s
47:15like an analytics service, a bidder, a
47:17repricer, all in one. And you just open
47:19it all in one place, and you will have
47:20everything. It's cool, but it kind of
47:23requires more expertise. Your team will
47:25need a separate person who will learn
47:27everything there, who will do all this
47:29and put it together. But, in principle,
47:31it can also be done.
47:33There is a tariff called VIP, where you
47:35join with one person from your team,
47:38give them everything to learn, and after
47:40that, you will have such a dashboard.
47:43You connect your WB, Ozon, 1C, Yandex
47:45Market, MoySklad, whatever you want, you
47:47connect it, assemble it, and that's your
47:50department.
47:51And, then there will be your own
47:52dashboard, just like I showed you. Plus
47:54, you can add to it, remove from it, or
47:56do anything you want with it.
47:58And, we will constantly provide you with
48:01new skills, our new developments, and
48:03new no-code agents. You can simply add
48:06them yourself and manage them. They
48:07actually work even better than what I
48:10showed you.
48:11And, there is monthly support and
48:13updates. This tariff costs 159, but also
48:15keep in mind that 50,000 rubles. you
48:17will pay per month just for maintenance
48:19. Not for maintenance, but for the
48:21database.
48:22So the subscription will be 50,000
48:24rubles. for the fact that this platform
48:26will run on a database, on a server, and
48:28function.
48:29So this tariff also exists, this kind of
48:32option also exists. Whoever is
48:34interested, please, you can take it.
48:37When calculating unit economics, where
48:38does it get the purchase prices from?
48:40You prepare the purchase prices
48:42separately, where I showed you in Claude
48:44. In our account, for example, there is
48:46a separate instruction and context.
48:49In the context, I have also uploaded my
48:52own, uh, purchase prices, including
48:54packaging and labeling.
48:57It can help you compile them, it can
48:59help you make them, but it takes my data
49:00from here.
49:02This data, again, resides only with you
49:03. If you are afraid that Anthropic will
49:05steal it from you, well, that's super
49:06silly, because who the hell needs your
49:08purchase prices? But it won't be with us
49:10, or Mercatus, or WB, or Amazon. It will
49:12stay inside.
49:14It's possible to make it pull them
49:15directly from your account, but what
49:17difference does it make if it pulls them
49:19from your desktop or from the context
49:21here, there is no difference.
49:23But overall, only it will see them. Data
49:25cannot be obtained anywhere else. Either
49:27API, or scan the client's Amazon, or
49:28look for something in MPStats. There is
49:30no other data, only who to collect from
49:31. It is unclear what Roman is talking
49:32about. If we talk about Ozon, we have
49:34the exact same thing for Ozon.
49:36The same reports, the same everything
49:37else, we do exactly the same.
49:40Regarding replying to reviews, there is
49:42a bit of a problem because they only do
49:45it via API through some paid plan, but
49:47we bypass all that using an intermediary
49:49.
49:51One email manager by specialty, one
49:52project chat. Well, basically, yes, yes
49:54. You can add any others, I don't know,
49:57like a search engine or whatever else.
49:59We can—I will teach you in the course
50:01how to make parsers for yourself, and
50:02from the parsers, look at Chinese sites
50:04, American sites, and search for new
50:06products there. There is also a lot of
50:08cool stuff. This is where we have the
50:10other skills, which I showed you, or
50:12rather the WB ones, but we also have
50:14general skills.
50:15Let's take a look through this summary.
50:18You will be able to analyze different
50:20categories, different niches, different
50:22marketplaces, and lots and lots of
50:24really cool skills that we’ve been
50:26developing over these six months.
50:28How do I get API access to a marketplace
50:30? Go into the marketplace, go to API,
50:33and get a token there. That's all there
50:34is to it. Do your bots work from a PC or
50:37from servers? From servers. If we are
50:39going to connect bots for ourselves, we
50:40will only need to buy a server, it costs
50:42, I think, about 1,000 rubles. per month
50:43, and deploy all of this on the server.
50:45Does it know how to select and analyze
50:47niches? Yes, it does. I showed you
50:48yesterday. Or we can do it right now,
50:50yes. Find on the marketplace amazon.com
50:54and analyze for me new categories, or
50:58rather new products in
51:01baby products.
51:03Find those products that appeared 60
51:05days ago
51:07hmm, which have, uh, no, uh, similar
51:10products, where there is good growth in
51:13orders and good ratings.
51:16And send me the top 10 of such products
51:18. You set a task like this for it. It
51:20already has the skill for how to do all
51:21this correctly, so there is no need to
51:23explain in detail, you don't need to
51:24tell it to apply this skill, you don't
51:26need to do any of that. It already
51:28launches the search itself, the search
51:30for new products.
51:32This is already launching our skill,
51:33which has already done everything here.
51:35It has a skill there through a special,
51:37uh, parser, through a special tool, and
51:39all of this is done. Currently, they
51:41provide info for one marketplace, for
51:43one. But I showed you, there is the WB
51:44account, and the Ozon account, and
51:46Yandex Market, and Lamoda.
51:48For four of them. Which Claude
51:50subscription plan should I buy?
51:52Uh, I said last time that you should buy
51:54the $ 100 plan, although Katya really
51:56assured me that the $ 20 plan would also
51:58work just fine.
52:00In principle, you can start with $ 20,
52:01and if you hit the limits, you can
52:03always upgrade to the $ 100 plan later.
52:05But even $ 100, uh, 8,000 rubles.
52:09per month, for what these analytics KBs
52:11do, I think it's something none of you
52:13will regret paying for.
52:15Will general skills be covered in the
52:16course too? Of course, yes. All the
52:18skills I showed you, both general ones
52:20and those for WB and Ozon, will
52:21naturally be in the course. I will hand
52:23the entire folder over to you. Save it
52:25on your desktop just in case, because if
52:27your Claude account suddenly gets
52:29blocked, you'll still have those skills
52:30; you can just create a new Claude or
52:32use GPT and continue working with them.
52:34Uh, someone asked, I missed it, does it
52:37set up ads on WB and Ozon by itself? It
52:39can be done automatically.
52:41Yes, it can do it itself, you can
52:42configure it and tell it what to do. The
52:44only thing is you’ll have to top up the
52:46advertising budget, but everything else
52:48, well, you can set it up to do it
52:49itself, though it still needs to be
52:51checked. So, let’s have no illusions
52:53here that they will do everything for
52:54you.
52:55They will do everything for you, but you
52:57need oversight. Does Claude or Chat GPT
52:59make mistakes? Of course, they make
53:01mistakes.
53:01Uh, is it possible to add an extra
53:03verification step? It is possible to set
53:04it up, but do you need to control it?
53:06You always need to control everything,
53:08including how the marketplaces work.
53:10Oh, my god, how the marketplaces work
53:12and how the AIs work as well. Will this
53:14course be held on a regular basis now?
53:15Starting Monday, I have a different
53:17course. When can I sign up for the AI
53:18course next time? Mikhail, I don't know
53:20.
53:21I don't know. Uh, the idea for it
53:23actually came up by accident, because,
53:24well, we've been working there for a
53:26long time. I met with some sellers, and
53:28they were, well, there was a decent
53:30seller there, doing like 20-30 million.
53:33And we had a mastermind session there.
53:35I said, well, we already do this, this,
53:37and this automatically in the panel, it
53:38files complaints and claims by itself.
53:41On Ozon, Chinese sellers are blocked, uh
53:43, well, automatically. They say, how can
53:46that be? I ask, how are you doing, well
53:48, there were some big guys there with
53:49large turnovers, and they say, we do
53:51everything by hand. I ask, why? They say
53:53, we don't know anything about this at
53:55all. And I thought that everyone already
53:57knew about this, it turns out nobody
53:59knows, so, well, that format was
54:00unexpected for us, because, well, Katya
54:02was saying that everyone surely knows
54:04about this already, it makes no sense.
54:06It turns out that, well, few people know
54:07, and it was necessary to launch a
54:09second cohort. I don't know, maybe there
54:11will be, maybe there won't.
54:12But actually, well, I don't know, you
54:14can combine them or not, see for
54:16yourselves.
54:17I don't know yet, it's not in our plans
54:19. I think that everyone who wanted to
54:21probably joined the first cohort, and we
54:23won't be launching any more.
54:25During the training process, will new
54:26product cards be added, and will the
54:28agent pull them up itself or does it
54:29need to be done again? Of course they
54:31will, they will pull everything up. It's
54:33like a real person, only better, you
54:34understand?
54:36You add a new product, the analyst will
54:38ask, what is the product, what is its
54:40cost price, you upload it, it will show
54:42you, well, I don't know, what is the
54:43plan for it, you tell it, what is the
54:45action plan, what you want, for example
54:47, to sell 300 units in February, what is
54:49needed for this, it will tell you.
54:52It is such a job.
54:53That is, well, everything is the same,
54:55just better than with regular people.
54:57And will there be an agent that works
54:59only with content, meaning photos and
55:01videos? Have you developed such an agent
55:03in your courses?
55:04Well, there is no such agent, because,
55:06well, we don't need it, but in principle
55:09you can make it yourself, please, it
55:11will not work with Claude, it will work
55:13Why?
55:15Because Claude doesn't generate normal
55:17photos, it doesn't have that capacity,
55:18it is an analyst, it analyzes, collects
55:20, does everything really cool, but in
55:22terms of content it is a bit weak.
55:25Therefore, you connect ChatGPT here,
55:27there the plan even for the simple one
55:29is, I think, 20 dollars, and, in
55:31principle, it is quite enough to
55:32generate cool pictures. Both pictures,
55:35and reviews, and these, and offers, and
55:37you can even make rich content with it.
55:40Super cool.
55:44I think I have, yes, I have skills
55:46regarding how to create the right task
55:48descriptions even for Claude and pass
55:50them on.
55:51There are such skills as well, and there
55:53are separate skills for photographers.
55:54So those skills exist too. If not, we
55:56will create them; it's not that
55:57difficult in principle. But, in
55:58principle, we will teach you in this
56:00course how to create skills for yourself
56:01, answer all those questions, and do
56:03everything independently.
56:04The main thing is that, well, you
56:06actually do all of this. Is GPT a good
56:08analyst?
56:09Well, they are all good, but to me,
56:11Claude seems much more powerful, much
56:13faster, and much smarter.
56:16Although, I think there will be those
56:17who disagree with this.
56:19But as of today, well, all tests show
56:21this and everything else.
56:23It really works better, gathers
56:24information better, and answers better.
56:26The most interesting thing is that the
56:28answers can even be a bit different.
56:31And depending on the context, the prompt
56:33, and the skill, everything will differ
56:35significantly.
56:37Can agents collect keyword sets for SEO
56:38? Of course, yes.
56:40We have now created both an agent and a
56:43skill for it, where on our Mercatus
56:45platform we will add all the search
56:47queries, and powerful SEO will be
56:49collected from those queries that are
56:52currently trending, so that you
56:54automatically get SEO that is relevant
56:56right now.
56:58That will also be a total blast. We are
57:00finishing it up now and will implement
57:01it very soon. Tell us more about
57:03connecting GPT.
57:04I don't even know. There it is, my guy,
57:06this GPT thing is right there.
57:09Well, I think it connects the same way.
57:12You download it to your computer,
57:14install it, register, connect via, well
57:17, set up a VPN, and pay with a foreign
57:20card or through services.
57:22But we will provide all the services as
57:23well.
57:24Can agents also be connected to GPT?
57:27Agents, I think, well, I don't actually
57:29remember. I think they already rolled
57:32out that their agent scheme works too,
57:35but I don't know for sure.
57:39Because we do everything through Claude
57:41.
57:42So, I don't know exactly about that part
57:44.
57:45But believe me, Claude is...well, if you
57:48want to get GPT to do images, videos,
57:51and analytics all at once, and to
57:53optimize things a little, perhaps yes,
57:56but Claude currently works much better
57:58in terms of data service.
58:01What agent options are currently
58:03available for GPT and Claude, and how
58:05can they be used for marketplaces? Well
58:07, think about it and take a look. Well,
58:09that's a bit slow too. Claude's speed is
58:11much faster.
58:12It's just parsing right now through our
58:15special skill, through the key, it logs
58:17in, opens, plus there's special
58:19marketplace analytics, that's why it's
58:22slow.
58:23But it's faster that way.
58:24So, yes, it can be used for Wildberries
58:26and Ozon work too. I checked the
58:27official OpenAI documentation.
58:29There are options for an agent within
58:31the current task, a permanent assistant
58:33with a specific specialization, and a
58:35scheduled agent.
58:37Even here, there is a scheduled agent,
58:39so everything I showed you can be done:
58:41sales, inventory console, card manager,
58:43working with Ozon.
58:45Super, right? It turns out that it's
58:47possible too.
58:48Everything is possible and can be done
58:50easily. For Claude and GPT, do you pay
58:52for a server in Europe, and at what
58:54price?
58:55You don't need to pay for anything in
58:57Europe for Claude and GPT. You'll just
58:59need to connect a VPN and pay for that,
59:01and that's basically it.
59:03If you are going to create employees and
59:05bots for yourself, then you will need to
59:07get a server, I think it's 1,000 rubles
59:09.
59:10it costs.
59:114 weeks, how many days a week? How many
59:13hours a day?
59:15You'll have one meeting per week, or
59:18maybe two, well, one or two.
59:21I think it's one, I think with, you know
59:23, workshop practices, plus there will be
59:25recorded lessons. So it's up to you to
59:27decide how much time you'll spend on all
59:29of this. The service costs from 1,000 to
59:313,000 rubles. Well, there is your answer
59:33.
59:34I didn't see how to install the Node.js
59:36program myself. I only access it without
59:39a browser.
59:40Well, we'll teach you all of that. And
59:42what should I do about the big bell?
59:44We'll cover that in the course as well.
59:46I'll buy the 150k plan. How much more
59:48time can one waste on routine?
59:51Andrey, well, please, if you don't want
59:53to waste time on routine and are ready
59:55to pay 50,000 rubles. a month, then
59:57please, go ahead and buy it, it'll be
59:58great. Anyway, it's taking a long time
1:00:00to think; it has already selected the
1:00:02items and is now checking whether the
1:00:04products it found on Amazon exist on our
1:00:07Russian marketplaces, looking at which
1:00:09ones have a better trend.
1:00:11So, it doesn't just look for new
1:00:13products in Skel; as I told you, it
1:00:15finds items with increased sales volume
1:00:17and good reviews, but it also compares
1:00:19them to see if such products exist on
1:00:21Russian marketplaces, which categories
1:00:23here are growing, and what the average
1:00:25check is, in order to fully tailor the
1:00:27entire top list that it will eventually
1:00:30provide to your request.
1:00:32That's how it works. That is all for
1:00:34today.
1:00:35I think everything has been answered and
1:00:36covered. Bye-bye.

About this transcript

This page contains the full transcript of Я сделал х4 в прибыли с помощью Claude на Wildberries: Рассказываю как! by На Маркетплейсах с Лео Шевченко, generated from the public captions YouTube serves with the video. The transcript has 10,954 words across 1,774 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.