56. Interview with Sean Williams: How I built a new AI category

0:00Welcome to The Difference Engine, the show for tech founders, investors, and innovators.

0:13Our guest today on The Difference Engine, in association with BoardWave, is BoardWave member, Sean Williams.

0:20Sean Williams:Sean is the founder and CEO of Autogen AI, a UK-based scale-up that helps organizations bid for public sector and corporate contracts by using AI natural language processing to write compelling proposals, ultimately saving time, reducing costs and boosting success rates. Clearly, Sean isn't alone thinking this is a good idea. We assume anyone who has to write any sort of proposal would like some help, to the extent that Autogen AI has secured over$60 million in funding with top-tier investors such as Blossom Capital, Spark Capital and Salesforce Ventures. Autogen AI isn't Sean's first entrepreneurial rodeo.

1:00Sean Williams:He previously founded and served as CEO of Cornel Limited, which helps HR, learning and development teams across many industries to find ways to invest their apprenticeship levy to add real lasting value to their organisations. He scaled the business from the ground up to a team of 350. In November 2020, he successfully sold Cornel to THI Holdings for$60 million. A philosophy graduate from Cambridge University and, as you're about to find out, a bit of a renaissance man. He still finds time to be a board member of the Sussex Community Development Association. Sean Williams, welcome to The Different Engine.

1:41So your current day job is being founder and CEO of Autogen AI. can you tell us you know why you're doing it and and how did you get there yeah so um great to be here um i sold my previous business uh in september 2020 and then i left that business in september 2021 and actually retired i retired for three weeks it's most boring three weeks of my life but i did have an opportunity to sort of explore some technologies while i was um while i was So I looked a bit into blockchain. And so I thought I'd find out what all the hype was about. And so I've learned about hash functions and cryptography.

2:25And I thought this was intellectually fascinating, but I couldn't really see any real world applications that I believed in. I didn't think there were any real problems that were being solved by this technology. Very interesting, though, it was. And then I started looking at Internet of Things, and I bought myself a little Arduino kit and started turning LED lights on and off from the Internet. I thought I was sort of pretty hot at this electronic engineering. You were clearly very, very bored, Sean. I was very, very bored, but I was having great fun. And I could see tons of applications for Internet of Things-type technologies, but mainly in sort of agriculture and manufacturing.

3:04And I don't know anything about agriculture and manufacturing, and who am I kidding? I also probably wasn't the best engineer in three weeks of having sort of self-taught myself. And then I was just really fortunate. One of my best friends, Andrew Cowie, who works for Google's DeepMind on the protein folding team, phoned me up and said, hey, Sean, I know you're bored and you're sat around. Why don't you look at these things called large language models? Like they're really interesting. Like we've built one and this company no one's ever heard of called OpenAI have built one. And, you know, they're doing interesting stuff with language.

3:39They can sort of read and write. And I said, don't be silly, Andy. Computers can't read and write. I sort of argued for 20 years that computers would never be able to do language. And Andy's a mathematician. He doesn't understand language, right? Don't be silly. You know, I'm a philosopher. Let me tell you about language. It's very hard, far too hard for computers. So, but as I said, I was bored. And so I thought I'd take a look. And I was utterly blown away. So, you know, I never thought I'd see in my lifetime computers able to read and write and suddenly had this machine that could. So I started playing with it.

4:10You hear a lot of people who work with large language models use that metaphor, play. Like, you know, it really still does feel like playing to me with the technology. So you talk to the large language model. One of the first things I did was use multi-shot prompting, which was kind of like a new technique at the time. I'm obviously completely passe now. But there were so few people doing this. It was all Frontier. And I got it to write new PG Woodhouse one-liners. And so, again, I discovered very early on, you can get the AI to write PG Woodhouse one-liners, get them to write 10. Nine of them will be rubbish, but one of them will be really good.

4:47So you need a human to pick the one which will be really good. There is a little bit of humor in there.

4:51Sean Williams:I don't know where it's come from, but you could probably tell us. But yeah, you can make it do jokes. Yeah, you often, I mean, I think it's accidental. That's really what I think. I think it's accidental. So it will come up with 10 lines it thinks P.G. Woodhouse would come up with, and nine of them P.G. Woodhouse definitely wouldn't have come up with. But one of them it accidentally gets right. And so, again, you start to notice things like human in the loop and things which would become kind of like commonplace, but we were discovering for the first time. And then I got to do things like translate Paradise Lost into modern English and kind of say all of these things which were kind of quite interesting in playing with the machine.

5:25What was it that made you start to turn your attention to something a little bit more serious? And may we say a little bit dull? Yes. So my very, very first career was actually as a proposal writer. So I was writing bids at my first job to the Department of Work and Pensions in the UK to run unemployment services and then to wider kind of social programs and then other kind of wider government services. so I knew both how much money and how much time was spent putting together these bids tenders and proposals for government contracts but also business to business contracts and um so so I knew the problem case and I saw this new technology and I kept thinking to myself for all of the strengths and limitations actually of this technology like proposals is such an obvious and good use case for this new tech and nobody really understood proposals as well as I did There were lots of people who understood it as well as I did, but no one really understood it better because I'd come from that world right in my very first career.

6:26So I understood the problem case backwards and kind of no one else was really doing this with large language models. So it seemed like an obvious intersection of ideas with a very clear commercial application. And I thought it would be too boring for people like Google or OpenAI. You know, OpenAI, we're talking about AGI at the time. You know, Google, when Andy showed me the technology, he says, yeah, we're not really doing anything with it. You know, like building stuff with it is a bit downstream for us. It's not going to win them the Nobel Prize. Not that interested in it. So really, I could see an opportunity to be downstream, to do the commercial boring stuff, but to create a brilliant product that would actually solve a very real problem for real businesses and real people.

7:07Right. So there wasn't any particular light bulb moment. It was just simply your experience of being involved in the relentless, soul-destroying task of turning out huge, complex proposals. There has to be a better way of doing it. Absolutely. And, you know, loads of people have tried to sell me software when I was working in that space. And I said, well, you know, there's like bid management software and there's tender finding software. But, you know, the real interesting thing is the writing. And prior to large language models, computers were absolutely no good at that. They added no value to the writing process.

7:46Whereas post-large language models, computers could add an enormous amount of value to that writing process when driven by a human. What is it you've had to do to go from large language model to a full commercial product that actually solves that problem that you've had to live through to identify? I think you start from what actually is it proposal writers are doing. So I've actually still got like the first picture I drew and I said, well, what are the tasks that proposal writers find, you know, irritating or kind of like, well, ideation, how do you actually start to come up from, how do you go from a blank page to ideas to a structure?

8:25and so the very first version of the tool we bought it was literally just trying to help people get ideas which they could then use a nice user interface to then put those ideas into a logical structure and then could click a button and get the large language model to take that structure take those ideas and write the first draft of a proposal outline and that that process like to write a couple of thousand words might have taken a decent writer half a day to do now you could do that first draft in you know a couple of minutes using our software and so then and then it was thinking about how along the all along the way could you then so that that would be the first thing you do and they go well we need some case studies well how do you find case studies and insert case studies how do you put in statistics how do you and that took us into a whole lot of things around retrieval augmented generation and linguistic engineering and what's now being called

9:13Sean Williams:context engineering and a whole host of other things i've got recent very very painful experience of form filling for government tenders. It seems to me that the ideation piece relies on, at least in our manual process, a lot of ingestion of the context of what the F are they writing about. Some of it's nice and it goes into a spreadsheet, but other times it's like, what does the English mean and what is it they're looking for? So presumably before the ideation, you guys were already sucking in a lot of, ingesting a lot of context. Yeah, absolutely. So really starting to understand that whole workflow.

9:49And the start of proposal workflow is kind of pre-opportunity when you're shaping the invitation that's about to come out. You're talking to the officials to try and make sure that what they put out is something that you can respond to and is something which is sensible. right through then to taking the documentation, reading the documentation, finding out what needs to be answered, taking everything from a minister's speech to a PowerPoint presentation that was given at the kickoff event to all of these pieces of information that have got nuggets that you need to either must include or should include if you want to score a winning score.

10:29And really this thing about proposals and particularly public sector proposals, but also business to business, private sector, it's scored evidenced pros they literally tell you how they're going to score it you just have to make sure that you're scoring it you know according to what they want to see and that again this then there's both explicit scoring criteria but also implicit scoring criteria that you get so you know we a customer using us uh software for a very big contract and they particularly focused on laser weaponry over other sorts of weaponry um because they knew that one of the admirals who was going to be marking it was particularly into his laser weaponry right

11:06Sean Williams:no that's that's kind of an implicit you would have thought little things like that would be more human than ai but i guess things have moved on quite a lot yeah well again i'm gonna come back to but it's about helping you know the ai doesn't know that a human knows that but the human can tell the ai and the ai can then use that as uh as context so it strikes us that you've you know you've really stole the march and created a very particular category using AI. I mean, it's clearly defined. And how big is the total addressable market for responding to requests to tender? The great thing about the public sector market is it's quite well understood because, you know, modern democracies have to be transparent.

11:50So they actually publish, you know, how much money they're spending and who they're spending it with. so if you add up everything that the u.s government spends with external suppliers and everything the uk government spends with external suppliers and everything the european union spends with uh external suppliers if you add all of that up um globally um that comes to about five trillion dollars five trillion which is a nice number in anybody's language exactly and then your portion

12:18Sean Williams:of that is the is the bidding cost portion well if you yeah if you said what's the bidding cost And we estimate about 10 % of that would be spent on bidding. So you've got a half a trillion dollar global market. And if you can make that 50 % more efficient, that's a quarter of a trillion dollars than you can spend. So look, the total addressable market is probably a little less than that. But it gives you some sense of the scale of how much money is spent actually on the cost of competition. in all of the companies that are using your technology. Listeners, this is not about amazing applications of AI at some point in the future.

12:56This is AI actually being used right now to do some of the most basic tasks that have to be done in large business and large government. So what is it that actually happens within the companies that you sell when they start using their technology? is it just they're reaping a great advantage in efficiency they're doing things much much more quickly with less people or are they going oh we can apply for a lot more contracts now what what's actually the balance or is the one yeah it's uh it's a bit of both sort of depends for different customers so some customers um you know maybe they've got a small proposal team you know five or six writers and they would love to then maybe they're putting in 150 proposals a year and they'd love to put in 300 they simply don't have the capacity so taking our software on allows them to you know put in twice as many proposals win twice as much work grow twice as fast and so that's a clear use case other like with very large businesses often they're already going for all of the business that they can go for and that will be you know a typical proposal process might take them nine months to a year to go through and how they're generally doing that is they're spending six, seven months getting to a first draft and then they're spending the last couple of weeks refining it.

14:15And actually, as we know, in all professional writing and most writing, it's the redrafting that is the important bit. You know, all first drafts are rubbish, but it's the second draft, third draft. So what they're doing is just moving everything to the left using our software. So instead of spending like six months getting to a first draft, they're spending six days getting to a first draft And then what they're doing is they're using that additional time that they've got to iterate, to get better, to really understand the customer better. And so what they're seeing is their win rates increase.

14:48We actually got – so there was a study done by MH &A Consulting of businesses who use our software versus businesses who don't use our software. And they found that on average – and they compared the revenue growth between 23 and 24 for businesses using our software. and businesses using our software grew 60 % of their revenue more than businesses not using our software. So we've just got really, really good external evidence points now as to the value of using also Gen AI. So, of course, it's a time-saving, but mostly what businesses have done with that time-saving is use it to grow faster rather than use it to cost-com.

15:27Can you talk to us a little bit about your strategy in getting to those businesses? We are a European-based podcast, and we like talking to European companies and we like to understand how companies that are founded in Europe become very successful around the world. Can you talk about your strategy in expanding the business and serving companies that are outside of just the UK? Do you know, it's another wonderful thing about the market that we're in. If you do business with the US government, the US government has to be transparent. It's a big democracy. So the US government publishes a list of everybody it spends money with.

16:10The UK government publishes a list of everybody it spends money with. The European Union publishes a list of everybody it spends money with. So we actually have the best data in the world about that. We just go, right, who does the UK government has 40 strategic suppliers? They're people they do more than 100 million pounds worth of business with a year. All of those 40 strategic suppliers should be our customers. Indeed, three quarters of them now are our customers. So you've got this incredibly good data set about the people who are competing for government contracts. In Australia, the data is even better.

16:40They don't just publish who won the contract. They publish who lost the contract. So that's a fantastic source of being able to go to anyone who's lost a contract and said, you should have been using our software.

16:49Sean Williams:That doesn't get away from the problem that you're a small supplier and some of these guys are very, very big. So I don't know if that's where you're going, Jonathan, but how does one start to break into those big-ass contracts? I see it's a classic, you know, account based marketing, right? So while other people are going to AI conference, I hate AI conferences, AI conferences of AI conferences are full of my competitors. I go things I go to things like construction week in Milton Keynes, where there were a whole load of very, very boring, you know, kind of big construction companies who compete for a lot of work, both in the private sector and the public sector.

17:26So it's, you know, it's going where our customers hang out. You know, we're members of, you know, lobbying groups in London, in Westminster, in Washington, because that's where those big businesses do, that's where they do business with government. So we talk at events, we attend those events, we publish a lot of kind of thought leadership in this space about how to win contracts. And so that's really, and then it is, you know, it's, I mentioned how long it takes to sell to the public sector, you know, selling to big businesses, you know, these are nine month to 12 month sales cycles for big enterprise businesses.

18:04So, you know, you've got to get in, you've got to get in at multiple levels, you want to be talking to a lot of different people. typically we'll be talking to actually proposal writers as well as proposal directors, as well as corporate development directors, as long as finance directors, CTOs, CIOs, chief executives, all of those people who are in different levels of the organisation are responsible for that organisation's growth. So who have a stake in buying our software so you can grow fast. And that's certainly a lesson for anybody listening to this who's in the relatively early stages of building their first firm, is that just because you've got something absolutely brilliant, which has amazing advantages, it saves loads of people, loads of money, don't think you're going to have a very, very complex procurement process to actually get to people who have the money and do want to buy it and have a reason to buy it and all those classic things that you have to have in a sales cycle.

18:56100%. Do you know, I talk to a lot, I invest in a lot of startup businesses and I talk to a lot of founders. and one thing I say to every founder because founders all like they like we like building beautiful things right we like building brilliant products and what I say to to pretty much everyone on this is look a nine out of ten sales team with a three out of ten product always beats a three out of ten sales team with a nine out of ten product always great marketing always beats great technology yeah so so you know you want to now look I'm not interested in selling rubbish life's too short right you know kind of so we we have a nine and a half out of ten products and a nine and a half out of ten uh sales team but you need both you have to invest in both so talk and talking about such things you know where do you see autogenai evolving over the next three to five years both in terms of its technology and the types of problem that you will be tackling yeah so look in the same way that nobody writes a financial model in the corporate world without using excel no one will write competitive prose without using Autogen AI.

19:59So it would just be, and already people are using us as a verb. We had one woman was talking to us about how, because government does this all the time, right, and big business, that a proposal had dropped on the Friday and it needed to be in on the Monday. And, you know, she was supposed to be, you know, going out with her husband and, you know, kind of like doing nice things over the weekend. And she said, look, I got it. Usually I'd have spent all weekend doing it. I've managed to get it completely finished on the Friday because I auto gen AI'd the shit out of it that's you know that's what you know that's that's what we like to that's what we like to do so in terms of the product advancing it's really we'd be the everything proposal engine so everything from finding opportunities to down selecting those opportunities to then coming up with that first draft to then incrementally improving that first draft to then sending it out for review to getting those reviews back to using AI to action those reviewed comments to second pass reviews to submission to then implementation plans to learning from where you've won learning from where you've lost really everything sitting in our engine from flash to bang that that written proposal cycle and you know we're already kind of like a considerable way there but we're always thinking in that process how do we concertina it how do we make it more efficient where's the right place to keep a human where's the right place to use ai where's the right place to use other technologies because just because you've got a hammer not everything's a nail large language models are remarkable but we don't we don't have to use large language models for everything what we're trying to do is help people get to winning proposals faster more efficiently more

21:36Sean Williams:effectively do i detect from that there's a bit of post proposal as well like where you may take this like okay so i've won it how do i live up to the criteria that that i've won it with in my submission. Exactly, that's a key thing people are already using our software for. So how do I take this now into an implementation plan? How do I take out all the key commercials that we identified? How do we take out, you know, what are our deliverables that we promised? All of those things that you can now just, you know, pull out in seconds using our software. So exactly. So how do you go from the finding of the opportunity to the winning of the opportunity to the implementing the opportunity?

22:12And of course, then, you know, how do you win it on the rebid in five years time? Great. It sounds like you're already on the road to becoming the generic, and we can all auto-gen AI the hell out of every proposal we do in the future.

22:34Sean Williams:Are you still developing separate marketing plans for each of your different channels to market? Can you really sanction messages reaching the same customers from unconnected targets? Listen, it's now the AI age. Things need to change. Positive is a different sort of AI-powered agency. From our very beginning, we've integrated marketing communications from PR to events, social to paid ads. Working smarter means working integrated. Let us show you what a truly joined-up strategy could do for your tech business. Contact Positive today with an email to hello at positivemarketing.com.

23:14You've got to learn to earn. So, Sean Williams, the first question we wanted to ask you, what are some of the most unexpected lessons you've learned since launching and scaling Autogen AI? And have they been different from your previous journey of founding, scaling, and exiting Corndale? Yeah, I think one of the big lessons at Autogen AI is just how dynamic you need to be because of how fast we're growing, but of course, because of how fast artificial intelligence is moving. So I talk a lot about when you're exponentially growing as we are, whatever you did three months ago is now half as good as it needs to be.

23:59So, you know, your systems and processes which were built for a million dollar company don't work when you're a four million dollar company, don't work when you're a 10 million dollar company, don't work when you're speed of adaptation, as soon as you've identified a problem and solved it, the solution you've come up with probably isn't going to work in six months' time. So that, I think, that pace of change, that dynamism, that understanding that these are complex iterative systems, I think, I did have that to an extent at Cornel, but, you know, the Education Skills Funding Agency don't move as fast as AI does.

24:41So one of the things you have to commit to as a founder and leader is constant and rapid relearning, knowing that change is going to keep getting faster inexorably. And if you don't understand that, you're pretty soon going to fall off the entrepreneurial horse. I think there's a thing as humans in general, that we're always, once I've finished this or once I've done this or once I've solved this problem then all the problems will be solved everything will be done and I'll be happy right you know I think that's part of the human condition and of course the you know the secret is to enjoy the journey that it's never going to be like that as soon as you solve one problem another 10 will have appeared somewhere else and you need to go and solve those and so you have to you have to take that on board you have to enjoy it you have to embrace the chaos.

25:32I think there's a certain sort of individual that survives and thrives in very, very fast scaling businesses. You know, if you come in and say, where's the rule book? If you come in and say, have we got a process for this? You know, kind of, you're probably in the wrong place. I say to, you know, everyone who works for us, you know, the great thing about being here is you're a big cog in a small machine. What you do matters. You're in a big corporate. What you do doesn't matter. No, kind of.