Unitary AI agents deliver ROI day one: Sasha Haco, CEO & Co-founder

0:00Hello, welcome back to the Scaling Europe show. I'm Seb Johnson. I'm here with Sasha, co-founder and CEO of Unitary. Can you touch a bit about what you've built and why?

0:09Sasha Haco:Hi, yeah, thanks Seb. So we've built Unitary, which is the way to automate back office, operating work in the lowest risk, lowest effort way. and so what virtual agents are are they're a sort of mix of AI and software that can integrate with our customers tools and legacy systems even and follow our customers processes to basically automate back office operations and we've been doing that across a range of industries so we started out looking at user-generated content doing lots of work in the content moderation space we then bought it to marketplaces looking at back office operations there and sort of sort of customer support operations um all in the back office and then most recently we've launched an insurance and so we're sort of one by one we're tackling different sectors to basically automate these sort of complex manual workloads um why why was content moderation the place that you started where did the inspiration for the kind of like initial idea for the business come from yeah so it's actually been quite a pivot and quite a journey so when we initially started the business, the sort of mission was to help to take down harmful content on the internet.

1:24Sasha Haco:So we started out building AI models and we initially thought we could actually use an incredibly large data set we were collecting across different customers to build the best models for detecting potentially harmful content, particularly in video. And what was interesting is that although we had some success with that in the early days, when GPT came out, that totally changed the game and made things like our impressive data set seem not so impressive after all. And so we realised that that wasn't where the market was heading. But instead, what we found was that a lot of our customers still had huge amounts of manual work associated with content moderation.

2:05Sasha Haco:But rather than detecting content, it was like investigating things that had been reported. and so that's kind of where the inspiration for virtual agents came from initially in this in the sort of content moderation world was looking at how we could actually support our customers to automate that these like complex investigative workloads and often are sort of all sit under the operations teams there's not a lot of tech and tech resource available to them lots of like heavy laborious process and we started building virtual agents for them and then realized there was actually much broader applicability and where where did i guess you know going back to you know my life is content I used to work in startup now in content I'm curious because you know your background's in theoretical physics uh black holes working with Stephen Hawking and now and you know going into that initial idea of content moderation where did that come from where did that sort of like passion point come from inside you yeah so I I um was working in physics um I was working on black holes um and I was sort of finishing up my PhD and I actually spent my whole PhD um trying to solve one equation in in black holes which is basically trying to figure out like what happens to something when it falls into a black hole like does it really disappear this is the kind of question and it was one tiny equation that I worked on for years and years well four years of my life and it felt like on the one hand I really enjoyed it on the other hand it was so agonizing that I was like it felt like I was going two steps forward one step back not really making any progress um I worked in like a small group but it didn't feel like a high-paced environment and I didn't get the sense that anything I was doing was going to have great impact on anyone and you know in in the black hole world you have to sort of apply sometimes for grants and people write things like you know they were in grants they want to know how it's going to impact the wider society and people say things like you know in maybe two centuries there's going to be this world belted develop this thing and uh for me that was like very unsatisfying and I wanted to do something which had like tangible impact um on people in the world um in my lifetime or at least like my children's lifetime um and not have to like gamble for like you know several generations to come and so I kind of started thinking I want to work on something fast-paced where you know they couldn't really make a difference and I got excited about AI and the potential to do that um and so I ended up joining Entrepreneur First and I thought a startup is a great way to have a potentially really outsized impact so I joined DF um back in 2019 and I met James my co-founder there there's actually sort of him that brought the the knowledge about content moderation because he was like by day like a ai coder and by night like a reddit moderator and so i kind of learned about this this world and thought it was just absolutely bizarre and fascinating and like the sort of underbelly of the internet that people don't really um don't really know about and so the more i learned about it the more i was kind of horrified and thought surely there's got to be a better way and so that was kind of the initial thinking was actually surely we can build better AI models to move the burden on people.

5:17Sasha Haco:And interestingly, we're still removing the burden on people, but in a totally different way. And did you ever solve the equation? Sort of. It's a very unsatisfactory answer. We ended up proving that there could be a solution, or that there was a solution, but we just didn't quite have it. You didn't have it, but then it'll come eventually. Interesting. And let's talk about the new insurance agent just then, right? So you're trying to automate and solve back office pain in admin. Can you touch on kind of like what does that look like in real terms? Like, you know, some real use cases, I guess.

5:52Sasha Haco:Yeah. So like in the insurance world, there are a lot of like really, really admin heavy processes where it's really hard to automate because our customers are kind of stuck in legacy tools. so they might have to like pick up an email download an attachment and then like log into some other system which is a desktop app and then you know do something there and then log into sap and then use some other other like policy system and they've got five different systems and so the approach is like okay we have to overhaul all these systems and create a new like cloud-based thing or you know whole new system and then we can automate and so that's going to take like many years often like they're on this sort of perpetual like transformation journey and so there aren't really good alternatives since that's what sort of virtual agents do they like work in us in the system and so the kind of processes that we're looking at things like um in um in the world of insurance is something called border row where um you basically get reports about what the uh what the firm has done essentially what kind of risk they've underwritten for example um and they come in loads of formats on loads of different people and someone's job to like you know collect each one put it on the right row of the spreadsheet like very laborious and you can it's so easy to make mistakes because there's just so many things that it's all kind of saying the same thing the output needs to look like you know one in one place but it's just spread across those different tools and systems or like another one is um there's a type of insurance company called an mga and um they get emails from brokers who like sending in um like these sort of pdfs where they awesome pbs where they've got to download and then take out information and submit them into their different systems and so there's basically loads of work involved like looking at an email figuring out what it's about and then uh you know downloading attachments and then finding the relevant information and copy and paste it into a different system that kind of thing loads of like really uh really tedious manual work that could just be completely removed and where does this come from i guess like you know do you have experience in insurance Is this a problem that you'd heard from other people?

8:03Are you kind of sitting and working with people who are operating in insurance to try and build these products? What's the story of why insurance is kind of the next vertical that you're tackling?

8:12Sasha Haco:Great question. So before this, I actually didn't know anything about insurance, really nothing. But what was interesting is we basically moved from content moderation to back office operations in marketplaces more generally. And what we were finding was that the customers that really were using our product and loved it the most were the ones which had sort of legacy tooling and where the alternative was to undergo this big transformation. And so they had lots of different systems that kind of prevented them from automating. and often what they were doing is to kind of try and reduce costs is offshoring this work to like BPOs and offshore teams where the cost of labour was lower because it was like hard to automate and the processes, they were sort of basic but they involved a mix of lots of clicking around, access in different tools and also using a bit of human reasoning and so these were the customers that kind of felt it the most and so we kind of looked around for like which other industries have processes in lots of different tools, often legacy tools, and sort of do lots of offshoring as a response to that.

9:22Sasha Haco:And insurance was like a major one in that. That's what led us to insurance. It's not the only one. There's like several others that we're sort of exploring as well. But insurance was a really, really good candidate and has been great for us so far. And can you talk a bit about, you know, why it's been great? What traction have you seen? What are the results been for people using the product? yeah it's been great because we've been able to help people get started really fast i think in insurance people feel especially if you've got these legacy systems it feels like the road to automation is like really long um and there's people excited by ai but often don't really know where to start and it's hard to cut through like the hype of sort of what is an ai agent like that everyone's talking about them um and so it's it's it's not much actual like roi in insurance at the moment so that's why it was really nice to be able to come in and say what we do is we take just take the process and people get the roi on on day one so what we've seen is like companies hit like 99 automation within about five weeks amazing so one of our flagship uh insurance companies in the us um but we've just had really exciting results and lots of like customer expansion and where are you expanding at the moment are you focused more on europe more on the us what's the deduction being like across drug fees um uk and and us there's like a really intense insurance market in london specifically yeah um so that's definitely a focus but also there are sort of larger players in the us and almost every state is like its own ecosystem um and so there are these large players in the us who are sort of excited to try stuff and so we're seeing both um uk like london market really and um us or east coast of the us is the place where we're focusing and outside of insurance you know well i guess like you know assuming insurance goes well what are you thinking about your you could have product roadmap or vertical roadmap but where else are you looking at this is like really similar use cases yeah so we're already exploring health care actually interesting where there's lots of sort of av mini stuff um some of that we touch on insurance like we're doing some work with health claims or like at the moment we're sort of exploring how like how we can help look at sort of patient discharge reports and that sort of thing and so there's lots of manual work there and like in gp practices is actually where we're where we're looking um where there's people doing referrals and and all sorts which kind of relies on using old systems that are sort of hard to rip out so they've stuck with them and need to figure out a way to like alter it on top of that and so healthcare seems like a really interesting space I'm also quite interested by back office in law.

12:05Sasha Haco:I think that there have been a lot of interesting AI companies trying to actually target legal work, like analysing contracts and this sort of thing. But there's a lot of just sort of getting someone on boarded. All this stuff is quite laborious. So I'm interested in the legal space as well. But lots of interesting potential. Yeah, yeah. And the situation you mentioned, there are loads of there's not loads but there's a few big legal ai companies there's a few doing things in health tech it seems like you're obviously going cross vertical but focusing exclusively on kind of back office function did you have you ever thought about just going really deep on a back office for a huge industry maybe if like insurance takes off you know would you ever think okay maybe the market is big enough for us just to service the top 100 200 players in this market rather than trying to build for lots of different i guess yes and no so i think our proposition is like perfect for mid-market um customers so they're the people we're really servicing today and so maybe we would go really deep and sort of start we're already sort of working with some larger but we maybe might that might be a focus for us but what's interesting is i feel like our proposition is only insurance specific but from a go-to-market point of view because that's how like that's because of the language people use in reality like there are lots of different processes lots of similarities but everyone's tools are slightly different and our product has to be able to learn and adapt really fast these different systems and so i think it's actually very very transferable to other to other industries and what i think is exciting is that like we've seen this in customer support like companies like sierra intercom they haven't just targeted a single industry they've gone like deep like sierra's gone deep but on like what seven eight industries and i think that's how i'm seeing back office as well like it's it's sort of very analogous to front office how can you really really fast target as many industries as possible and not because one industry is not big enough but because there's probably going to be a winner that actually is able to cover all of them or a significant number and so like the time is really right now to scale super hard and and try and tackle as many as we can so i think it's a deliberate choice yeah that's really interesting i guess it's quite a hard it's i imagine it's hard to frame it for gtm because you're trying to solve back office problems but i guess like yeah and specific industries want to hear that you're solving it for them uh which is like puts you in an interesting place because i think a lot of front office stuff is is often applicable for lots of companies like a crm system you know most people across a large group of industries will know that they'll be able to use a crm because it's quite versatile whereas back office stuff i guess has never really been done like this and unless you're using like a shared service center.

14:52Sasha Haco:Exactly. But I think that's a really interesting opportunity. I think even front office, like people do want to know that things have been tuned for their industry. Like how do people like to communicate in the insurance sector versus another sector? Like how does, what kind of customer responses people want? How much is done via chat versus email via phone? Or, you know, there are definitely nuances, even if things feel like, oh, it could be across any sector. And so I think it's like, we have to have a different go-to-market proposition for every vertical that we go after. but i do think it's very similar to fund office in that in that respect yeah that's really interesting and and how do you think um like that that that go-to-market approach will change you know i guess like you know right now it seems like you're quite heavily focused on insurance right that's kind of the vertical that you really want to like grow in and succeed in is it going to get to a point where you're kind of having to run three or four verticals all simultaneously pushing equally as hard or like you know at some point yeah like yeah i mean it's interesting because there's a lot of noise about insurance right now but then i guess maybe if that matures then you're going to be talking about health care you know it's 100 percent it's 100 and it's like each one we need to make sure we can deliver the product and it's like really well suited for that use case and for that or the use cases in a specific industry but then once we've like nailed that there might be a different like we might launch virtual agents somewhere else but i think rather than just sort of try and go after everyone all at once i think what we've like deliberately gone after insurance and make sure we like really understand what are the use cases here and how we're going to tackle it and then we'll like do something similar in healthcare when we like really know what is the right what is the right product what are the right use cases and like what's the right go to market motion for a different industry they're more like sort of deliberately go after that have you seen any differences in adoption across either vertical industry or uk versus us the us just moved faster always uh so which is uh frustrating i guess um but it's sort of for us it's been in general easier to find early adopters um when we've entered a new space in the u.s and then sort of use them to really solidify exactly what we're doing crystallize our approach and then the uk's great great market or europe's been a great market interesting okay yeah okay look and as you as you look forward now to rolling out insurance you know what are you looking to kind of achieve in the next 12 to 18 months to make sure that this is a success and that the business sort of you know gets to the next stage so i think it's it's um obviously revenue growth is like one major thing but i think like it's it's revenue retention and expansion because we want to show that our customers really love this product and are expanding because all our customers have use cases all over their businesses and that's kind of partly what makes gives me a lot of confidence that this is a very cross industry issue because when we started out marketplaces we would start out doing something in content moderation they'd be like actually we've got this other back office task over here our finance team has got back office tasks and we move around all the different sort of business units within a within a company and i want to see that happening in insurance so we're going in through you know the most on fire use case because they've got a whole team of people having to sort through these policy documents or whatever it is or claims but actually they they discover this can be widely applicable to all sorts of things and so that's the kind of thing i really want to see happening in insurance is just this like moving around the whole company yeah yeah i guess yeah experience those sort of internal network effects uh well look sasha thank you so much for joining me it's been great chatting it's super interesting what you're building and i've spoken to a lot of people building agents for different companies but you know i think the approach that you're taking is is actually completely unique and i haven't heard of anyone building something like this but i think it's always super exciting it's really cool to hear when people are taking like a novel approach to like a i guess a popular space which is like ai agents um so thanks no i really appreciate you taking the time and uh best of luck with it thanks so much thanks for having me Thank you.