Sept. 9, 2024

David Dobrin: seizing AI opportunities in an information rich business

David Dobrin: seizing AI opportunities in an information rich business
David Dobrin: seizing AI opportunities in an information rich business
AI Pathfinder for Private Equity Podcast
David Dobrin: seizing AI opportunities in an information rich business

In this conversation, Steve Budd speaks with David Dobrin, Product Director, AI Platform at YouGov, about the potential of AI in the market research industry. They discuss the importance of taking action now and starting to build AI capabilities, even if the technology is not perfect. They also explore the challenges of managing expectations and keeping teams engaged with AI.

AI Pathfinder Private Equity AI Strategy Meet-ups:

https://www.aipathfinder.co/private-equity-ai-events

Have a question or would like to suggest improvements, please contact steve@aipathfinder.co

SPEAKER_01

Well, hello and welcome to the AI Pathfinder for Private Equity podcast. Um I'm on a mission to lead private equity from AI bewilderment to AI empowerment. And I'm doing that by building a dedicated network that will piece by piece create a playbook for successful AI adoption across PE operations and value creation. I'm meeting some fascinating people along the way that have a lot of on-the-ground experience important for anyone to seek out amongst the noise and hype of AI. Over the coming months, I'm looking forward to sharing these conversations with private equity and those on the AI value chain. And today I'm really looking forward to speaking with David Obryn, Product Director AI Platform at Ugov. Thank you very much for joining us. And well, look, why don't we start by uh by briefly giving a bit more context to you, your role in the company.

SPEAKER_00

Yeah, sure, absolutely. It's great to be here. So I can I'll start by just giving a bit of background on Ugov. So, like you said, UGov is a global market research and data analytics company. Uh, we leverage a vast panel of respondents. I think the last count was 26 million, um, who opt in to take surveys and ultimately help us to provide insights into consumer behaviour, opinions, and trends. Ugov has two main business areas. One is research, and the second is data products. So, research is work we conduct on behalf of specific clients who commission work with us, and that can range in complexity from simple surveys all the way up to the most complex custom studies and trackers for some of the world's biggest companies. Um, and then we also have data products. Uh, so we have a brand tracking tool called Brand Index and a media profiling and audience insight tool called Profiles. And these data products are interesting because they're powered by Ugov's proprietary data. And so that's data collected and owned by Ugov, and they're sold on a subscription basis. I'm currently product director in the AI platform team. Um, that's a new team that we set up at the back end of last year. And essentially, I'm responsible for defining, developing, and implementing Ugov's AI vision strategy and product roadmap. And so my role involves overseeing the development of that AI platform and then working with teams across the business to understand their ideas and their requirements and to help them implement and incorporate AI to improve efficiency and productivity as well as product innovation. Prior to the role that I'm in now, uh set up a startup called Incombo, which was a chat-based polling platform that initially launched on Facebook Messenger and then we built our own application. That startup was acquired by Ugov. So that's how I found my way into UGOV. And I ran that startup for two or three years and then integrated the product and the technology into the main Ugov business. That was at the same time as ChatGPT and this new AI wave uh came to the fore and uh just gradually transitioned into the role that I'm in.

SPEAKER_01

So, what we're saying essentially is that you came first with chat, then it was chat GPT.

SPEAKER_00

Yeah, exactly, exactly. So it's interesting the the use case because the chat product that we built was all manually scripted uh by a team of content editors, but we were trying to mimic what it would be like to have a real conversation, and that that was in 2018. And so I often think now how different that product would be if we were starting now and we wouldn't be mimicking conversations or scripting conversations, it would be powered by AI. Um, and I think that's something that you govov is is looking into, and because the potential to scale and personalize conversations using AI is is incredible, and we were always quite limited by the fact that we had to script conversations and say things like hmm, that's interesting, in response to people to try and like create that you know that feeling of having a real conversation. So, yeah, we did come first.

SPEAKER_01

Yeah, absolutely. Well, look, let's um perhaps get into a little bit more context of of of um well the advancement that we've seen with AI, and and look, I get told many times that it's not new, it's been around for for some time, but obviously the technology has has really uh taken off and become well far more accessible, hasn't it, to to everyone now? What opportunities does that present to you know you as a company and to your customers, and how how generally is that helping to realise some of these opportunities?

SPEAKER_00

Yeah, so obviously you're right in saying that AI as a field is is not new, but I think this new wave of generative AI using uh in particular the the large language models that we now have available presents a whole range of new opportunities. I really think AI or this wave of generative AI has the potential to transform all businesses uh across industries and across sectors, and I don't think that's hyperbole. I really think this AI wave is transformative and is akin to, if not bigger, than some of the biggest historical historic historical technology waves we've seen. I think AI has the potential to transform the industry that I'm working in, the market research industry, um, and youGov potentially more than most, because we're a data company and the type of data that we work with is language-based. So we're asking questions, the survey options that people have available to them are also text-based, so the answers, and we're also collecting a huge amount of text or free-form data, asking people to express their opinions in their own words, and so this language-based data is obviously highly suited to the large language models which we now have have available, and which is incredibly exciting for us. It presents numerous opportunities for UGov, our customers and our members. So we have like a few different stakeholders that we're thinking about. We have obviously our internal colleagues, we have our customers or our clients who are paying for our data or our services, and then we have members, people of the members of the public who join Ugov and take surveys. Um, and across all three of those, I guess let's call them stakeholders, we think there's opportunities for AI to be transformative. So internally, we think it has the potential for us to automate internal processes and workflows to increase our speed and efficiency of delivery. For our members, so the people taking our surveys, we have an opportunity to improve the member experience, including diversifying and improving the ways in which we ask questions and how members can share data and interact with us. And then for clients, we have an opportunity to, I guess, improve the availability, the accessibility, and the actionability of our data to offer an improved experience for them and deeper and even richer insights. So the phrase that we're using at the moment is moving towards conversational analytics. So can we create a more dynamic, more engaging, and more interactive way for clients to understand and interrogate their data to get even more insights from it? And we're working across all three of those areas, and we think the opportunities are there to drive significant improvements in all of them.

SPEAKER_01

There's a sense that there's infinite magic tree sort of options with with AI, which which makes it difficult to maybe identify where to start. And it's not, you know, it's still evolving, isn't it? Not everything works perfectly. You know, if you're looking for perfection, that's going to be quite difficult even now. When you're looking across that broad spectrum of use cases, how are you how are you prioritizing right now?

SPEAKER_00

Yeah, so we we decided to so that the way we got started was that we did a a wide stakeholder stakeholder consultation process. And so we met with so Ugov has around 3,000 employees now, so it's not super massive, but it's also not a small business anymore. Um and there's many, many, many different teams from like product and engineering through to research, operations, marketing, finance, HR. And so we did a big consultation process in January to sit with all as many teams as we could to understand their respective business areas and their day-to-day workflows, and to hear more than anything about the challenges that they face in their day-to-day work and what ideas they had for how they thought AI could help and improve either the internal processes they're working on or the products and services that they're offering to the end client. And we we came up out with a really long list of ideas, uh, way more than we could possibly work on. And then we had to go through a prioritization process to try and work through the ones that we wanted to work on first. As we were getting started, we had, I guess, a few different criteria to consider. So the first was do we work on internal projects or do we work on external or public-facing projects? Initially, we decided to focus on internal projects because this is a new area, it's a new technology, it's new to us. And so by working on things internally, there's like less reputational risk. And if things go wrong, you're only exposing something internally, and you're you're not on the front page of a tech website about how some how something's gone wrong. And so the first use case we did decide was an internal one, but very quickly we felt like we should be more aggressive and ambitious, and so our second use case was an external facing one. And then in terms of how we decided what ones, we just looked at what the problem was that we were trying to solve, what the potential value or return on investment might be if we were to work on this project and it was to succeed. We looked at the technical feasibility, so we did wanted projects with medium to high feasibility, especially for the first projects that we worked on, because we didn't work want to work on impossibly hard projects that had no chance of succeeding. And so there was a careful balance there of making sure we delivered return on investment, but it also had a high chance of succeeding. And then we looked at the I guess the technical complexity in terms of delivery time um and how long it would take for us to get to a phase one and a phase two and so on. And when we mapped all of those out, it was possible for us to identify two use cases to get started on. Um, but it's difficult and it's kind of an imperfect science because there's so many ideas, and ideally, we would be working on five or six, but at least to begin with, we chose two, and that's kind of the process that we went through. And then moving forward, we're gonna establish a more robust operational framework for us to work through use cases, and it might be that some teams are there like lighter use cases, and teams can just work and build on them themselves using the models and the infrastructure that we make available, and then there might be more complex or heavier use cases that do require specific machine learning engineering and support, in which case we'll have we'll go through a prioritization process to work out what ones that we want to lend out dedicated engineering supportive.

SPEAKER_01

It's a challenge for everyone right now to try and prioritize, and I and often, well, it can be that you end up just being paralyzed by that and and not really doing anything. If you've seen the opportunity to act sooner rather than later, and it'd be good just to get a view really from an urgency perspective, even though you just selected two use cases. What are your thoughts around act now versus just wait and see?

SPEAKER_00

Yeah, so I think my my personal view is that it's important to act now and to start building, and there's degrees to which you can decide how much you want to invest and how fast you want to invest. But fundamentally, I think I think it's important to not just wait and see, and I and I think there's a few reasons why. So the first is that it is possible to achieve and deliver real value in ROI from the technology that is currently available, and that should be the immediate goal and the main reason for acting now. And obviously, you don't have a chance of achieving that if you don't do anything, and you're just waiting to see what happens with the technology and other services that are out there. The second thing is I guess like what you said, it's it's also possible that companies will have to go through a number of phases on a given AI project before significant value or ROI is achieved. And I think it's important to act and start building because it will be impossible to realize truly significant value over a longer term in say stage three or four of a project if you haven't laid the foundation in stages one or two. And I think not the the chance of not realizing immediate value is not a reason not to act. If anything, I actually think it's the opposite. So I feel like you have to be on the court, as it were. Another reason, like just more from an operational standpoint, is that there's a really steep learning curve to all of this. There's a learning curve, there's an experimentation curve, there's a user adoption, a user openness curve to like a new technology like this. And I think the best, maybe even the only way to navigate all these various curves, is to start using the technology and to get your hands on it and start building and experimenting. And obviously, you can't do that if you're just waiting to see or if you're not doing anything. Um, and then I think the final reason is that if everyone else is doing and building with AI, which does mostly seem to be the case, whether they're achieving immediate value or whether value will accrue to them in the over a longer time horizon, is really easy to fall behind. In fact, it's highly probable that you'll fall behind. And it might be that starting small and building, and you have to do that just to keep up, rather than to actually get ahead. And I think that's also fine, but obviously you can only keep up, and you can definitely only get ahead if you take action, not if you wait and see. I think mostly all of this comes down to, and this might be a thing that emerges throughout this conversation because it's one that we talk about internally, is thinking about this, I guess, over a longer, a longer time horizon. So say over a three or five year time period. And if you ask the question, if we start now and continue building and learning, where will we be in five years' time? Versus if we don't start now and we don't do anything for two or three years, where will we be? And I personally think it's a much bigger risk to wait and see. Um, and all of this is to say that you have to go all in, which I think is where the nuance comes that I mentioned at the beginning. But I'm pretty certain that doing nothing is the wrong approach, given the state of the technology now and I guess the arc of progress in terms of both capability and cost. As to whether how you decide whether to build things yourself or to use off-the-shelf solutions, we're taking the approach that we're not going to be dogmatic or ideological about that. And so we'll take a given use case, and if it makes sense for us to use an off-the-shelf solution, then we'll absolutely do that, and we'll work with that product and we'll roll it out to users, and that's the route that we'll go down. If, however, we think that there's an opportunity for us to build something ourselves because we're using proprietary data, or it's a proprietary end solution that we want to offer, then we'll build that and we'll go for it. But I think in both instances the same philosophy applies that it's probably best not to wait and see and do nothing. I think it's important to act even if you start small.

SPEAKER_01

Just reflecting on what you're saying there, with the audience, private equity audience. Obviously, they've got their own operational side of things, and then they've got the value creation with their portfolio. And I suppose that you know definitely there should be some, you know, that that that absolutely from an operational side that should be in discovery mode. I think really important, regardless of really your ambitions with AI, is that a lot of the foundational elements for AI are hygiene factors that that should be in place anyway. And but when you look at the portfolio, they're on a cycle. And so it it can be really hard to know know and how to invest. And you know, at any given time they've got a portfolio that at the start, middle, end of the cycle. So where potentially the horizon for you government might be five years, that's a tough one for for private equity to sort of buy into unless they're at the very early early start of the journey with a company. So it's it's uh something for them to to to grapple with to try and think why that that that needs to happen, or at least a pathway uh that shows evidence that more investment would would result in something of value. And it certainly the very least is that there's there's knowledge and understanding of the technology to to to have a uh you know uh it could be a really brief outline of an AI strategy rather than nothing at all, I suppose.

SPEAKER_00

Yeah, totally, and I g I guess just to clarify it, we're we're absolutely aiming to deliver immediate value and ROI. I just and but I think you can hold both those thoughts in your head at the same time, which is we want to deliver immediate value ROI, um immediate value and ROI, or what we're doing now is a building block upon which we can iterate in the future to deliver more and more value. And so I think both of those things are important, is that you you can absolutely go for it in the short term, and it definitely is possible to build valuable things over a shorter time horizon, and also understand that I think that value will accrue over a longer time period, but it's only if you're developing a strategy and experimenting and exploring that you'll you'll realize that value over a longer time period, and not least because as advanced as the technology and the models that are available are, it's still early days and it's still less deterministic than traditional software engineering. And so there is just an element of uncertainty that exists in these projects that while it hasn't matured as much as other technology that might be available.

SPEAKER_01

Do you think do you think that companies should be thinking about a north star for their AI deployment, you know, their AI strategy? Or do you think it's more about well, the north star should be the business strategy and it's then how AI can support it?

SPEAKER_00

Yeah, so I I I think one one mistake that companies can make is to do AI for the sake of AI, or to have this like fear of missing out that you feel like everyone's doing something, so we need to be doing something. I absolutely think it needs to be rooted in the business strategy. What are the business objectives and how and where can we use and deploy AI to further those business objectives? So I definitely think that should be the right approach rather than just doing it for the sake of it or because you think other people are doing it. That said, because it is such a transformational technology, it's possible that those business objectives might change or or evolve because something is now possible where it wasn't before, or because you have competitors doing things, offering new things, or offering the same things at much lower prices that that wasn't possible previously. So I think in either case, the strategy should be to have a set of business objectives and to have an overall company strategy and then to work through how and where AI can support that. But it's also possible that it might be the other way around where the fact that we have this new technology can and should inform and maybe even change what those business objectives and what the strategy is.

SPEAKER_01

Yep, yep. Good thoughts there. Let's move on to trying to sort of unpick what you see as uh an interesting case study or use case of AI. What's what's piquing your interest at the moment of how you know others are doing things that that might be interesting and useful for for you govern yourself?

SPEAKER_00

Yeah, so I I'd probably think about this on sort of like a macro and a micro level. So, like on a macro level, I think there's a few things that are happening in the ecosystem that I think are really interesting and have quite significant impacts on those that are using and utilizing the technology. So, like just to take one example, I think the whole debate around the open and closed models is a really interesting one, especially for enterprises who might be concerned about how and where their data is being sent and how it's being used. So I think what Meta are doing with open sourcing Lama is nothing short of incredible. Like I think they're doing it for their own self-interest, which is obviously perfectly okay. But I think the wider impact that that's having on the ecosystem and the interplay between the various closed and open approaches and how that will play out over time, I think is really interesting. Most of all because I think ultimately it will benefit end users, both in terms of like cost and capabilities. So I think that's just like one really interesting area. The other interesting area, I think, is this debate around who's going to benefit most from this AI technology wave. Will it be incumbents with established distribution or startups building from scratch? And on this, I'm not sure if there's a clear-cut answer, but it comes down to whether established companies can achieve AI innovation before the startups get to distribution. And at the moment, I'm seeing examples of both incumbents successfully deploying and they have established distribution. But we were talking about Canva before. I think that's a really great example of a they're not a legacy company, but they could probably be considered an incumbent given their size and their user base now. Adding AI to improve the user experience and the services that are able to offer. Notion is another really good example of a company that's successfully adding. On the other hand, you've got these like fantastic startups that are building AI first, um, and they're gonna put pressure on existing companies and it and it and um and incumbents both in terms of margin and competitiveness. That's something that we're seeing, and I think I guess from a private equity perspective, I think a lot of portfolio companies are gonna start to see competitors, these AI first competitors that are just changing the game, that they're they might be offering products that are less complex, but they'll be at a much lower cost and they'll and they'll be far more competitive. And so I think that that whole dynamic between incumbents and startups and who's gonna benefit most is gonna be a really interesting one to watch. In terms of like specific use cases, so like bringing it down a level. I mean, Klarner came out quite boldly a few months ago to say how they're using AI to automate huge amounts of their customer support. I think they said AI was doing something ridiculous, like the equivalent of 700, the work of 700 full-time agents or something crazy like that. Customer support is an obvious use case, it's relevant to pretty much every business. Um, and I think it's really interesting to see the success that they've had. In the world of market research, we're seeing we're we're seeing the dynamics that I described before incumbents like like you gov and others that are trying to add AI to their ex their existing products and to try and augment and improve those. But we're also seeing lots of really cool, very exciting startups that are coming on the scene that are offering things that are really innovative and are challenging the market. And uh to quote the I think the Microsoft CEO, like they're making our stance, and and that's that's what they they they should be doing. Um and we're seeing interesting things around synthetic data, like survey and data analysis, and so yeah, there's look there's lots of interesting ways in which it's being deployed, and I think it's really fascinating to see how it impacts companies and the wider ecosystem in general.

SPEAKER_01

Yeah, there's lots of tools out there, isn't there? And it's hard to know on the face of it how good they are. Um, they've all got sort of a great story potentially, but actually, when you get into it, there's still a lot of work to be done to make it really useful and must-have.

SPEAKER_00

Um totally, and like I can give you an example actually. So we we want to roll out and so we've got customized use cases that we want to where we want to incorporate AI and integrate it into workflows or products and services, but we also want everyone at Ugov to be using AI tools just in my day-to-day work. And so we were looking at different options, and one option that we decided to pilot with was my Microsoft Copilot. So it's an off-the-shelf solution, you pay on a per seat basis, it integrates into the work applications that we're using every day, like Word and Excel and Outlook and PowerPoint and so on. And it's powered by OpenAI. So you're you sort of when we went into it, we thought this is an off-the-shelf solution that provides the best of all worlds, it provides the security and privacy that we need. Um, it gives us access to the best language models because it's powered by OpenAI and it's integrated into our workflow. When we started rolling out rolling it out, we I think we trialed it with 30 colleagues initially. We just quickly found that the results just weren't good enough. People would would ask the AI to summarize the latest emails, and it just all the emails just came out in the wrong order, but it hallucinated on quite a lot of company data, and just overall the performance of the tool just seemed wholly inadequate compared to things like ChatGPT and Claude. And so, with having gone through that process, we've now made the decision that we're just gonna build our own AI assistant ourselves. We're gonna integrate with some of the best language models. We'll probably use bedrock on AWS, which makes Claude available, Lama available, Mistral, a few of the models. And we're gonna build it in a UGUV-managed environment so that we don't have to worry about data privacy and security. And by building it ourselves, we'll have the ability to customize it for our workflows. It'll be easier for us to integrate our data into the tool for people to work with that. Um, and so that was just like straight away a good example of it. Felt like an off-the-shelf solution was gonna be an easy and quick win. But actually, once you get into the details, you find some of these things don't perform as well as as you'd hope. And that actually, this is a good example of this is something that we think is better to build ourselves. Now that that's not always going to be the case, like there will definitely be off-the-shelf solutions that work perfectly well and and and are more than suitable. But there's definitely these things aren't always as straightforward as they seem, especially with the off-the-shelf solutions. It's not just with the proprietary builds.

SPEAKER_01

Yeah, and I think there's it breeds the the issue that I mean change management is a big thing here to to make or make uh uh enable adoption, whether that's uh internally or externally. And when you've got Gemini and you've got Copilot are now in line with your workflow there to help you, but is making mistakes, isn't able to help you in the way you think it should. For those less, I suppose, positive about AI, will immediately have then a reason why it's not working for them and probably will not come back to it for some time because it just didn't work. And so it's it's tricky, isn't it? You know, that we're all trying to sort of push the envelope and Microsoft and Google, biggest sort of pockets if you like, can't quite get it right. Then how the hell can well you gov and other businesses get it right? So there's a lack of I think of um you know belief and trust perhaps that's gonna take time to to see through. And I think you know it goes back to just selecting two use cases, um, getting that right, earning that trust, whether that's internal stakeholders or um or your customers, and build build from there.

SPEAKER_00

Exactly. It comes down to, like you said, it comes down to use cases, and I think it also comes down to managing expectations that you can believe and it is true that this is a game-changing technology that allows us to do things that weren't previously possible, and it's also not perfect 100% of the time, but it doesn't necessarily need to be. The question is, is for a particular use case, is it a net benefit? Does it save you time? Does it increase the quality of the output? And if it does, then it's useful, even if there are mistakes that you have to correct manually. If the end result of a net benefit is is positive, then fantastic, and then just over time things will get better. And so there is this, I guess, juxtaposition where you need to sort of manage expectations on the one side, but also like push, like keep pushing the envelope on the other and understand that realistically it's not going to be perfect all the time, but not but nor does it need to be, especially if you find use cases and ways to use the technology that gives you a net benefit.

SPEAKER_01

How do you think um you can keep that level of curiosity and interest in this technology across a team and organization?

SPEAKER_00

Yeah, I think that's the one of the big challenges, I think, because on the one hand, you people use Chat GPT and Claude, and it just feels like magic. And so, and it and it's so easy to use, um, and the outputs are so good, and so people's expectations rise almost exponentially, and then you might come to build a proprietary use case, and people might expect it to work. Chat GPT just works like this, and I just type it and it comes out with an output, and suddenly you're working with proprietary data with certain guardrails or things that you need to you need to account for, and it's and it's not always as easy as that. And so I think that is one of the main challenges is how you can well, one manage those expectations, but two, keep users or customers or members or stakeholders, whoever it is, you know, open and excited to the potential of this. And I think it might it might just come down to like you know, if they can see that by doing this, it allows them to do X, and that saves them time or increases the quality of their work, or they can see pathway if we do X, even though I might not be delivering value now, we have to do X in order to get to Y. And when we get to Y, I can definitely see how this is gonna start to deliver value. So I think it's both having sort of a realistic expectation of these tools and their capabilities, and also demonstrating to people the the value that they can get now andor the value that they can get in the future. But a lot of these, like you've mentioned it before, a lot of these new technologies go through ways where everyone's like, ah, this is amazing, and then it it falls down, it's like, oh my god, this is terrible, and then it just goes back up and then it just becomes normal. And I think if you just keep if you keep that in at the forefront of your mind that these things do go through cycles, especially when things are so new, then I I think that helps. I mean, the other thing is that things are moving so so so quickly. I mean, every day, every week, a new model comes out, new capabilities, bigger context window, more compute, more capability, and so the the rate of progress is truly exponential, and the costs are almost coming down equally exponentially, and so there could be that the capability just continues to improve, and people just every day or every week that goes by, people can see the more and more possibilities that is overwhelming at times as well, but that that could also be a factor that helps just to keep people active and engaged with it.

SPEAKER_01

Final final question for you is there a tool that you can't be without?

SPEAKER_00

Yes, but I don't think it's uh fancy or unsurprising. Literally use ChatGPT and Claude every single day. I I pay for both, and genuinely couldn't imagine working um without them now, which is absolutely crazy because like we've been saying, it's yeah, it's so new and it's uh you know we're talking 18 months or two years that they they didn't exist, but yeah, I use it every day for ideation email summarisation.

SPEAKER_01

Well look, we we have come to time and we could talk for a long time, and we are gonna hear from you again, David. I'm really pleased to say because you're joining us at um my very first breakfast briefing on the 17th of September this year. It's the launch of a quarterly breakfast briefing alongside the the network and that I'm uh starting for private equity. And under Chatham House rules, you're able to sort of go a little bit deeper into your experience and the initiatives you've got running at Ugov. So that's going to be really exciting to hear hear more of. Well, look, um look forward to seeing you in in a few weeks' time. Um and really enjoyed the the conversation today, David. So thank you very, very much um for your time, for sharing your your thoughts and experience with with the audience.

SPEAKER_00

Thank you. Good to be with you. Cheers. Bye bye.