YT YouTube2Text

YouTube2Text → Transcripts → AI in Insurance: What Leaders Get Wrong About Automation with Sasha Haco, Unitary

AI in Insurance: What Leaders Get Wrong About Automation with Sasha Haco, Unitary — Transcript

by MGAA · 4,465 words · 633 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:05Today's conversations episode with Sasha
0:08Ho from Unitary AI is so exciting. It
0:11covers her sort of sliding doors moment
0:14when she decided to create her company
0:17as a as a co-founder. Uh she gives
0:20fantastic tips on people where they are
0:22in the journey in terms of implementing
0:24AI. She also has a great advice for
0:27young women as they sort of enter the
0:30career and and what they need to do to
0:32be successful. So take a listen. It's a
0:35great episode. Sasha, good afternoon.
0:38How are you?
0:40>> Good afternoon. I'm well, thanks. Thanks
0:41for having me.
0:42>> Great. So really looking forward to our
0:44discussion and I'm sure our listeners
0:45are. So let's get straight into it. So
0:47first of all, I think we should start.
0:49So can you give our listeners a sort of
0:52introduction into Unity AI? obviously
0:54which you co-founded you know what you
0:57do what your specialtity is what your
0:58products are and we can start from
1:00there.
1:02>> Yeah. Brilliant. So um I'm as you say
1:04one of the co-founders and the CEO of a
1:07business called Unitary. Um I started
1:09the company about seven years ago now
1:11and we're an automation company. So we
1:14essentially figure out how to turn
1:17really manual processes into ones with
1:19maximum automation as fast as possible.
1:22And we have a product which we call
1:24virtual agents. So virtual agents
1:26essentially mimic human workloads. So
1:30they log into systems just like a human
1:32would. There's no like tech integration
1:33at all. They type in username and
1:34passwords and two factor authenticate or
1:37whatever it is. And they actually mimic
1:39the whole process from end to end um
1:41using a combination of software and AI
1:43to basically get the whole job done um
1:46with maximum sort of robustness but also
1:48with like the power that you can achieve
1:50with AI. So they they take on human
1:52workloads and they um of automate them
1:56with um actually guaranteed human level
1:59accuracy. Um and they do that super
2:01fast. So within sort of a few weeks
2:04they're up and running automating a
2:06submissions process or claims admin or
2:09whatever it might be.
2:11>> And and and just the genesis of the
2:13idea. So seven years ago, were you
2:16sitting on a train? Were you were you
2:18were you sort of lying on a beach? you
2:21know just you know what what what sort
2:22of what was the thoughtprovoking sliding
2:25doors moment or whatever.
2:28>> Yeah. Uh well you know what we've had
2:29such a journey so it's been like very uh
2:33lots of twists and turns I guess. So
2:35before this I was doing a PhD um in
2:39studying black holes. So I totally
2:42different was working on theoretical
2:45physics trying to understand like the
2:47universe I guess and it was really
2:49interesting really cool in some ways but
2:53also like painfully unimpactful I found
2:57like I could never see like tangible
2:59results from what I was doing and I
3:02wanted to work on something where I
3:04could see um see impact I guess on like
3:06a daily basis and so I felt like
3:09starting a startup was a way a really
3:11good way to do that. And I was really
3:13excited about AI and like the potential
3:15that AI was going to have. And this was
3:17seven years ago, eight years ago. So AI
3:19meant something totally different. But
3:21even then AI was like an increasingly
3:23powerful tool being used in like lots of
3:25different industries. And I was really
3:26excited about that. And so I started
3:29thinking about AI. I met my co-founder
3:31James who had a background in um AI. And
3:35actually the first use case we became
3:36interested in was looking at how we can
3:39sort of automate the manual work
3:41involved in content moderation for
3:43online platforms. So, if you can imagine
3:45like on your dating app or whatever, um,
3:48someone reports another user, we're not
3:51using dating apps, but you know what I
3:53mean, there could be whatever it is,
3:54game, PlayStation, um, somebody reports
3:57another user for like harassment or
3:59whatever, whatever, and then that report
4:01goes to a queue that's reviewed by a
4:03person somewhere. So, um, all of these
4:07social apps and marketplaces have teams
4:10and teams of people whose job it is to
4:12manually review and investigate reports.
4:16And so, we thought, you know, keeping
4:18internet safe, that's a very very uh
4:20kind of important cause. Let's try and
4:24first build automation to do this. And
4:27that kind of over a few years led to
4:29this product which is virtual agent. So
4:32basically being able to without any tech
4:34integration um mimic this human work and
4:37we we have a human in the loop. So it
4:38means they can actually do the work as
4:40well as a human but automate it as much
4:42as possible. And so we started off we
4:44were doing it in in this content
4:46moderation world. We started working
4:48with platforms across like dating apps,
4:50social social apps, marketplaces,
4:52gaming. It was going really well. And
4:54then what we recognized was actually
4:56this product that we built had much
4:58broader applicability. We were
5:00essentially um fig taking on a manual
5:02process and which looks different in
5:05every company, figuring out the
5:06different tools that they use and
5:07building a product that can
5:08automatically uh you know navigate the
5:11different systems even if they're like
5:12really legacy and horrible um and and
5:14mimic the human human workloads. We
5:16thought actually this has so much
5:17broader applicability. And the customers
5:20that kind of loved this this the most
5:22were ones which had sort of their
5:25workloads were spread across lots of
5:26tools and systems which meant that it
5:27was really hard to automate. Often they
5:29had legacy systems and they just had
5:32like really like a lot of pressure on
5:34manual repetitive tasks. So we kind of
5:37looked around for other industries which
5:38had these these characteristics and one
5:42of them has been insurance. We've also
5:44been working with we expanded across
5:46other parts of marketplaces and and
5:48healthcare as well. But um insurance
5:51through that became a real focus because
5:53in the insurance industry like as you
5:54know there's so much manual work there's
5:57so much automation to be to be done but
5:59at the same time you I guess there's two
6:02interesting things about insurance like
6:04one is you have to get it right and like
6:07the cost of an error is really high and
6:09so that's why not just having like an AI
6:12agent that like can do any old thing
6:13felt really important like we need to be
6:15able to have a robust um like really
6:18robust process behind ind it. So that
6:21sort of fitted our model really well,
6:23but also um there's a lot of legacy and
6:25that's often stops people able to
6:28automate. So the fact that we could
6:29actually just start working in these
6:31like legacy systems and tools meant that
6:33we could actually add a lot of value
6:34really fast. So insurance sort of quick
6:37became our biggest like growth area.
6:40>> Interesting. What a what a what a great
6:42story board in terms of from start to to
6:44finish. That's fantastic. Yeah.
6:46>> No sliding doors moment there really.
6:49So you know moving into AI effectively
6:54uh and obviously you know you you was
6:56fully aware of it sort of seven eight
6:57years ago uh clearly now AI is the talk
7:01of the buzzword of insurance and and
7:03other industries to be fair.
7:05>> Yeah. So, so through your own experience
7:08and obviously you deal with a number of
7:10sort of partners and and clients in the
7:13insurance industry and everybody in the
7:15sector probably is at different points
7:17of their evolution in AI. You know, what
7:20would be your advice to you know
7:24customers or our listeners in particular
7:26who may be at the start of their journey
7:28or maybe been partway through their
7:31journey. you know what what are the key
7:33what are the key things to to do and the
7:35key things to avoid.
7:37>> Great question. So I guess if you're at
7:40the start of the journey, my main advice
7:42is to start small, pick off a piece and
7:44do it well and then expand from there.
7:46But in terms of like you know what how
7:48should you grapple with AI? I think what
7:50I've learned is that AI sounds great and
7:54so excited about AI and there's every
7:56time I go to a conference it's like you
7:57know everyone's talking about AI but I
8:00do think AI is not the right the job in
8:02every situation and so I think people
8:04need to be careful about confusing
8:06automation with AI because actually AI
8:10and AI agents and agentic stuff
8:12everyone's talking about these models
8:14are inherently probabilistic. They're
8:15basically making a guess and a good
8:18guess usually um at like every step like
8:20what should it do next? What's the next
8:22word in this sentence or you know and
8:25that's super powerful and often is like
8:2895% accurate. But the problem in lots of
8:30these like insurance um workflows
8:34that of they often have like 20 steps,
8:36you know, like open up this attachment,
8:38find where the right part of the
8:40attachment is this other thing, put this
8:42into some policy admin system or
8:43whatever it is. and they're like 20
8:46steps and so if it's the AI is 95%
8:48accurate every step then these kind of
8:50even if just the error is 5% they
8:53compound so then after you've done a
8:5420st step process the errors like I
8:57don't know less than 40% sorry the
8:59accuracy is less than 40%. And so in in
9:02a most most businesses wouldn't tolerate
9:0440% accurate uh like process. Um which
9:09means that AI isn't the right solution
9:10the whole way. So you need to find a way
9:12to leverage the power of AI without like
9:15falling into this trap. And actually
9:16we're hearing so much in the news about
9:18how most AI pilots are failing. Like
9:20most AI pilots never make it into
9:22production. Um even though they're all
9:24so excited about AI and why is this? And
9:25I think it's because of this problem
9:26that you can test something on one step
9:28and it works great. But when you try and
9:30put it into this whole chain, whole
9:32workflow chain, suddenly it doesn't work
9:33anymore. And so I think the the the
9:36trick is to actually only use AI when
9:40you need it. And make sure that for all
9:42other steps you might want to automate
9:44but with software. So you have sort of
9:45traditional software or so traditional
9:47but like you you need software that can
9:50do the deterministic step. So anything
9:52which is not a human reasoning based
9:54decision is bit done by software and
9:56then you have um AI for the bits that
9:59require like human judgment and then I
10:01think the third important piece what we
10:03have in our product is having a human in
10:04the loop so that you have a combination
10:07of software AI and humans which allow
10:09you to maximize the power of AI while
10:13also maintaining the sort of robustness
10:14reliability
10:16um and like repeatability that you need
10:18from software
10:20>> and and are you a supporter of the you
10:22speaking to a number of our members
10:23around at different parts of their
10:24journey that you know before you you
10:28know they should cut any real checks
10:30that there are some free sort of tools
10:33which you can sort of experiment with
10:35and you know you know you obviously sort
10:37of your first advice is that you know
10:38start small on that piece. So, so don't
10:40necessarily feel that you have to invest
10:42heavily at the start, you know, look at
10:44what's currently available maybe off the
10:46shelf just to get a get a sort of feel,
10:48you know, is that something you
10:49subscribe to and you would you would
10:50support.
10:51>> Definitely. Definitely. I think I'm a
10:54big believer in no big bang projects,
10:56right?
10:56>> Like
10:57>> nothing that takes a long time to do.
10:59You should see results really fast. And
11:01if you can't see results in a month,
11:03then you know, especially if you're
11:05paying, you should you should get you
11:06should only pay for ROI. you should make
11:08sure that you get some some real return
11:10within the first month I think or two.
11:13Um it shouldn't be you know a year year
11:16out and even companies we speak to who
11:18are on these big transformation journeys
11:19where they're like putting everything
11:21into a whole new system can still get
11:23quick wins. You can still actually
11:25implement something today within a few
11:27weeks that allows you to start
11:28automating some parts of the process
11:30which then you might migrate onto the
11:32new system when it's ready. So I think
11:34definitely start small and try out
11:35things. Um there are like tools out
11:37there that are you know specific model
11:39AI models for specific specific things
11:41calculating risk about something um that
11:43people can play around with but I think
11:45it's possible to start automating like
11:46simple workflows um just really quickly
11:50and in a way that demonstrates value and
11:51then you can expand from there.
11:53So Sasha picking up on sort of you know
11:56the journey sort of seven years and
11:58identifying insurance as a real sector
12:00which needs sort of transformation and
12:02and automation and and I'm smiling
12:04because I'm mentioning Bardro but we no
12:06doubt we'll come on to that. So can I
12:08ask you do do you do you see any
12:10particular processes either being in
12:13claims operations or underwriting which
12:16are are particularly ripe for sort of AI
12:19and AI agents or or do you feel that
12:21they're all on the same same sort of
12:23level you know what what's your sense
12:26>> I think there's so much opportunity I
12:29think there really is so many things to
12:30do um underwriting has a lot of
12:33operational uh admin so like
12:35underwriting assistants spend spend so
12:38much of their time taking information
12:39out of emails and putting it into like
12:42internal systems and that's a very
12:45laborious process that is basically just
12:47rekeying information some thinking
12:49looking something up on the internet
12:50maybe but it's it's not high leverage
12:53work so for me anywhere in the business
12:55which is full of people doing like lowle
12:58leverage activities where they should be
12:59spending their time on more important
13:01things that's the place to automate so
13:03underwriting definitely claims have a
13:06lot of admin definitely
13:08You mentioned Bordo, that's like,
13:11you know, the big horrible one that
13:13everyone talks about, but there's so
13:14much you can do there. Um, but but to be
13:17honest, everywhere in the business has
13:19we're seeing we often start doing a PC
13:22in one underwriting part of underwriting
13:25operations and then that expands then we
13:27end up working with different teams in
13:29the business. Um, all sorts of like
13:31customer care stuff, back office. Um, so
13:34there tends to be just like manual
13:37process absolutely everywhere. Um, so I
13:39think it's I think it's an exciting time
13:40to be honest to be an insurance company
13:42if you're like open to adopting like
13:46latest technology and and getting on you
13:49know getting on it because
13:52it can you can start accelerating in a
13:54and doing much more with less. I think
13:56that's what AI enables people to do or
13:58just automation. And it means you can
14:00suddenly scale um in a in a way without
14:03going headcount. Like what we're seeing
14:05is we're we're for a business we're
14:07working with in the the US and MGA. We
14:10um we've taken on parts of their of
14:12their loss fund processes, but actually
14:14what what we saw is not only can we turn
14:18things around much faster and it's much
14:20cheaper and everything, but they've also
14:22had a big boost in their like broker
14:24satisfaction. Um so they sort of
14:27reme-measuring it and it went a lot
14:28better. So that's also almost revenue
14:29generating as well. So you can you have
14:32all these there are the obvious benefits
14:33like cost savings and faster faster sort
14:35of speeds getting things done faster,
14:37but you also have all these extra
14:38benefits as well. And I think that's
14:40going to be really compounding and the
14:41businesses that really embrace
14:43automation and and change and AI are
14:46going to have just be able to scale in a
14:47way they've never seen before and
14:49without having to scale their headcount
14:51which is really exciting.
14:52>> Yeah. Well, there were certain as you
14:54say sort of revenue generation and
14:56broker satisfaction and that just shows
14:58in terms of the how how powerful sort of
15:01you know those tools can be in terms of
15:02your evolvement and sort of
15:05profitability of all businesses. So it's
15:07that's great metrics actually to be able
15:09to sort of to sort of play back. So it's
15:11good.
15:13>> I want to I want to sort of just shift
15:14the topic a little bit if I may. Uh now
15:18obviously sort of you're a successful
15:20woman. you've got co-founding with your
15:22with your sort of co-founder in terms of
15:24uh unity uh obviously on you know great
15:27momentum I'd really like to sort of you
15:30know share with our listeners in terms
15:33of you know your view about you know how
15:35women can be successful in the insurance
15:38industry but obviously in the industry
15:40where you know you've come up in that
15:42tech sort of space from that bit and you
15:45know can you sort of give some of your
15:47own experiences and and sort of uh any
15:50sort advice to our sort of younger
15:52listeners who are no doubt admiring what
15:54you're what you're talking about today
15:55in terms of what they could replicate in
15:57terms of it carving out a very
15:59successful career for themselves in what
16:02is still you know to be honest and
16:04recognizing still can be quite
16:05challenging for for females certainly in
16:08insurance sector.
16:10>> Yeah, it's interesting. I think I mean I
16:11think I spent my life in like
16:12male-dominated industries and like being
16:15a tech person is you know they're still
16:18predominantly men but actually the most
16:20male-dominated industry I've been in is
16:22black holes. Um, so when I was working
16:26as like, you know, doing physics stuff,
16:28that was when there were there was I was
16:30often the only woman in the room. And
16:32so, um, I actually notice it less now I
16:35did before. Um, which is funny. But and
16:37so, so I really I don't notice it as
16:39much. Um, I think though it's just about
16:42having uh confidence and backing
16:44yourself. Like um, people always say
16:47that when they put job ads out, women
16:50make sure they like tick every single
16:52criteria before they apply. Whereas like
16:54men, this is obviously massive
16:56generalization, just just like
16:57stereotyping, but men are like, "Oh, I
16:59just tick one of the boxes. I'm going to
17:01apply for this job." And so I think it's
17:03it's that mindset. It's like, well, it
17:05could be good enough, so let's go for
17:06it. Um, and just having more like good
17:09enough mentality. Um, I think is what
17:12women need. Just put themselves out
17:14there a bit more. Um so I think that's
17:16my advice is just
17:18>> and and I just just before we move on
17:20are you encouraged in the direction of
17:22travel or in terms of sort of you know
17:25you know women given opportunities to be
17:27successful in the industries in terms of
17:30which you engage with. Are you
17:31encouraged that it is going in the right
17:32direction or do you feel we've we've
17:34reached a sort of plateau and and
17:35perhaps there's more work to be done?
17:38>> There's always more work to be done for
17:40sure. I mean, it's not just women, but
17:41there's so many like minority groups
17:43that just, you know, that are being left
17:45behind. So, there's always more work to
17:47be done, but I definitely am encouraged.
17:48There's like more and more and I go to
17:51events or meetings, I'm seeing more and
17:53more women. So, the direction of travel
17:55is definitely positive.
17:56>> That's good. That's good to hear.
18:00>> I just want to touch on very very
18:01broadly uh you know, regulation actually
18:05and and at a very high sort of level.
18:07So, so and and really what prompted this
18:09sort of question and and it's more to
18:12sort of get your view not around the
18:14regulatory environment but you know
18:17where you may see challenges with the
18:20use of AI and you know from effectively
18:23customer detriment whatever. So
18:25obviously the the the FCA announced a
18:27couple of weeks ago they were going to
18:28do this AI sort of review which is quite
18:30right because you know AI can be used
18:32for pricing for risk appetite and
18:34whatever. So there there's obviously
18:36challenges. So, so if I put to one side
18:39the rigory and whatever the FCA do and
18:42actually just ask for your, you know, to
18:44share your expertise in terms of, you
18:47know, AI tools and how that works, you
18:49know, does it give you any concern that,
18:52you know, they could be used in the
18:54wrong in the wrong way to disadvantage
18:57customers and whatever and and, you
18:59know, any sort of particular examples or
19:01or where you feel the guard rails need
19:03to be to ensure that, you know, we don't
19:05become a a wild west show with uh with
19:08with AI tools. You know what's you know
19:10be interested in your view.
19:12>> Yeah, I think I think we need to use AI
19:15with caution. Um so we need to use AI in
19:18the right places and there's some tasks
19:20which is just a no-brainer for AI but
19:22there are other tasks where AI can be
19:23super powerful but where it's the
19:26results are so influenced by like what
19:28they've been trained on. So like risk
19:30and that sort of thing, modeling
19:32specific like these sort of point
19:34solutions um are likely to pick up
19:37biases from you know everything they've
19:40learned. So I think there's there's
19:42definitely um we definitely need to be
19:44cautious and I think there needs to be a
19:47a large element of human in the loop and
19:50human oversight before we let these
19:53things just do their do their thing. Um,
19:56so that's that's my advice, I guess,
19:57that I'm I'm you know, we're an AI
20:00company and use AI every single day, but
20:03um I think AI is not the right tool
20:06every always for the job. And so it's
20:08like when when should you use AI? When
20:09should you really have software or a
20:11person doing something? And there's
20:14definitely a place for people. Um we're
20:16not going to replace it. And I think
20:18that's really important to remember.
20:20>> Yeah, that I think that's a very good
20:21point. So in a way if if if I played
20:24back from a regulatory perspective is
20:26that you know AI can enhance the
20:29customer experience and outcomes but it
20:32has a has a role to play and not
20:34basically replaces everything in terms
20:36of that in terms of that sort of process
20:38whatever because of the you just said
20:40because it it can then become all all
20:42consuming and then there could be
20:44detriment.
20:46>> Exactly right.
20:48>> Okay. So, so look, you know, fantastic
20:50to sort of have this conversation, a
20:53great journey from sort of seven years
20:54ago. Uh, I suppose hopefully we're going
20:57to have this conversation again in 12
20:59months time. So, you know, if and when
21:02we do have that conversation, sort of
21:04where are we now? February 2026. So, 12
21:07months, which unfortunately comes around
21:10very very quickly now in time. you know
21:12what would you hope to be sharing sort
21:14of with me and our listeners in terms of
21:16you know let's say take let's take say
21:18say two things one AI in general uh and
21:22just as importantly in terms of unity in
21:24terms of what you'd hope to achieve in
21:26the next 12 months
21:28>> so in terms of AI in general I think the
21:31capabilities are getting like more and
21:32more powerful every day and that means
21:34that like magic seeming applications are
21:38becoming possible so I think it's going
21:40to that uh I guess this is the same as
21:43for Unitarian for the world. It's going
21:45to make things that are really hard
21:47today easy. Um so working with like you
21:53know really hard systems or automating
21:57things that just feel impossible are
21:58going to become possible. Um already
22:00when I sometimes do demos to people
22:01they're like this is this this surely
22:03isn't real. And I think we're going to
22:05get that in a whole new way in the next
22:07year. Um, so I think that's really
22:10exciting. I think also like from a
22:12unitary point of view, um, we use AI in
22:15our business every day. Like people use
22:17now you can use AI to code for you. And
22:19so suddenly we're able to just do so
22:21much more. Our best engineers are people
22:23who are who aren't coding all day, but
22:25they're using AI to write their code.
22:27Um, and just getting so much more done
22:30and we're going to see that like
22:31productivity gain happen again. Um, but
22:34then also for us as a business, I think
22:36we've got a really exciting year ahead.
22:38Like we're now firmly in the sort of UK
22:41London market, but and we're just
22:43starting out in the US and that I'm
22:45hoping that's going to grow a lot in the
22:46next year, too.
22:47>> Okay, that's fantastic. Okay. Well,
22:49look, really thank you very much for
22:50your time. Extremely insightful. Uh,
22:54debunked some of the myths. uh really
22:57really pleased about how consistent you
22:59were that in terms of AI is not
23:02replacing humans, it's complimentary to
23:04what humans do. Uh which is fantastic
23:07and uh again thanks very much for your
23:09time and and for sharing all things AI
23:11and particularly unity. Thank you very
23:13much.
23:14>> Thanks Mike.
23:18>> Thank you very much for listening. The
23:20MGIO welcomes all feedback on all our
23:22episodes. So please get in touch. Please
23:24get in touch with any suggestions on
23:26topics or new guests. And don't forget
23:29to subscribe on your chosen platform so
23:31you don't miss out on any of our future
23:33exciting episodes.

About this transcript

This page contains the full transcript of AI in Insurance: What Leaders Get Wrong About Automation with Sasha Haco, Unitary by MGAA, generated from the public captions YouTube serves with the video. The transcript has 4,465 words across 633 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 translate the transcript into 100+ languages.

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, translation to 100+ languages, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.