How to use AI to change the game

Ryan Barry: Hi everybody and welcome to this episode of Inside Insights, a podcast powered by Zappi. My name is Ryan and I'm joined by Steve Phillips, the founder and CEO of Zappi. Steve, what's up, man? 

Steve Phillips: Hey Ryan, how are you doing? Great to be chatting to you. 

Ryan: Great to be chatting to you. I mean, it's not like  we don’t chat every day anyway, but here, let's do a podcast.

Ryan: You know what, we should say it's great for all of you to enjoy us chatting to each other. We're going to take you on a glimpse of our crazy. So Steve and I are just off the back of a, uh, monthly town hall that we do with our staff. And, uh, the theme of today's discussion is going to be to give you some tangible inputs of what you need to do.

Ryan: And what specific problems you can solve using AI. And I thought it was really cool at the beginning of our, uh, town hall, we had an AI generated Frank Sinatra song. And that was pretty cool. And it made me, uh, it made me want to ask you, are you excited to see Leonardo DiCaprio play Frank Sinatra in the upcoming film that they're doing?

Steve: As it happens, yes, I am. I'm both a Frank Sinatra and a Leonardo DiCaprio fan. 

Ryan: Same big Leo guy, uh, very big Leo guy. Do you know, I share a birthday with Frank Sinatra. Not as handsome or as successful, but you know, all right. Uh, but yeah, Frank Sinatra, myself and Sanjit Mitra, who's our partner from Sumeru all have December 12th birthdays, but the most important 12/12 birthday is actually Kelsey, who produces this show. Team 12/12! So we're doing a lot of stuff and what we want to do on our podcast always is bring you folks things you can bring to work, right? That's what was the vision when we started the podcast.

Ryan: And that's still what we're going to do. So we talked about AI on this podcast about a year ago as chat GPT was starting to really blow up. We've been quiet on the subject. Well. If you've been following us marketing wise, we haven't actually been quiet, but we've been busy innovating and we wanted to take a pause today and just have a discussion with you of what we're seeing and what we see some of the problems occurring are so we can work through them together.

Ryan: So Steve, let's dive into topic number one. We have some wonderful advisors that help shepherd our business and they produce a quarterly report. Port That basically gives us a glimpse into trends in our industry across the buy side, the sell side, the investor side, the corporate side. And the top two trends for the second quarter in a row are data quality and artificial intelligence.

Ryan: Why do you think those two things are A, the most important, and B, intrinsically linked? Steve? 

Steve: Well, everyone, we are in the consumer insights industry and every insight, all of our insights start with data and so data is the underpinning of everything we do, and it's that old adage of garbage in, garbage out.

Steve: And if we can't get the right data, then we can't make the right insights. We can't make the right conclusions. Frankly, we can start utilizing AI on top of bad data, but we'll get bad outcomes. So you have to get the data right.And the data, right it's not just making sure that the right, the respondent is who they say they are, and they say something that they mean about the product or service that we're talking about.

Steve: It's making sure there's richness to the data, but it's also stored in the right way. It's accessible when someone wants to make a conclusion out of it or make a decision based on it, that it's available to them in a, in a way that's accessible.

Steve: So that underpinnings of the entire industry is the core quality associated with the data and the availability, accessibility of that data. 

Ryan: I agree with you, and I think, you know, with data quality, it's never been a bigger problem, but there's also, I, in my opinion, never been a bigger opportunity to solve this.

Ryan: I was talking to Mike McCreary, our partner at Pure Spectrum. For those who don't know, Zappi uses Pure Spectrum's panel to pipe in all of our respondents. And that brings me to topic two. We were discussing synthetic data, and I hadn't really made an assertion on this topic until last night, and my thought is: The more we bring in what we know about you, the less we have to ask you, the more we can infer from clean data sets, the more we can have real people answering questions we actually need the answer to, and letting the lookalike models fill in the blanks. 

Ryan: But if you think about that, that's a great leverage point for AI, but it's also a sort of mission critical when online data collection became a thing, there weren't nearly as many ways for people to monetize their attention as there are today. And now like, I mean, we're talking ADD central on these platforms.

Ryan: Um, so, so just zoom right into the synthetic data topic. Where are your views on where it's going to show up? Cause I know that this is even more so than AI in our space. I feel like synthetic data is a very polarizing topic amongst the industry. What are your views on the topic? 

Steve: Yeah, I think synthetic data is, I think, one of the most interesting spaces for our industry.

Steve: We've been playing around with effectively what is synthetic data for years, years and years, decades. Um, because if you do any form of prediction modeling, really you're creating some form of synthetic data. I think what's changed is this idea of having a synthetic respondent or, or individual synthetic responses and doing them at scale.

Steve: So my, my view is that, that you can think about macro synthetic data, which would be I'm looking at an ad, the system looks at an ad and says, that will do well without even asking anyone. So, that's a macro predictions piece of synthetic data. You also have micro synthetic data, which would be we're creating a respondent to evaluate something, something we've done, um, but an entirely fictional one.

Steve: I think what to me, what's more interesting is the hybrid interconnection of those two things, where at the moment, if you think we've got to evaluate an advertising campaign, that means we're going to ask 400 people 27 questions about that ad. Well, it may well be that you only really need 300 and then you can do prediction modeling for synthetic people above that, or it may be you don't need 27 questions. You can have 20 questions and then predict the results to the other seven. But it also may be, you can have an interconnection of that. So you can have, you know, 300 respondents ask, and then you randomly ask them of, you know, a proportion of those 27 questions, um, or you may have 400 people, but only ask some 100 of them three questions, and you could predict.

Steve: So it's that confluence of basically taking the least hassle for an individual respondent and the most data and insight for the client, and that's Frankly, it's not new maths. I mean, that's effectively what a conjoint model is. Um, so we've been doing this type of thing. It's really the potential to do it at scale.

Steve: That's interesting. So we, if we talk about marketing data, um, you look at marketing data and lots of it is very big data, clickstream data, sales data, social media data. Uh, and the problem with survey data is that it's small data. It's not big data. It's small data, but the good thing about it. It's incredibly rich and incredibly insightful and much more insightful and much richer than most of the other data streams.

Steve: And I think what the really exciting opportunity is, is to take synthetic approaches to the richness of the data to expand it so you can have. Rich big data rather than just rich small data, I think those are some of the areas we're experimenting with. 

Ryan: Yeah, it makes sense, right? So if you have rich and robust abilities to pull what, you know, then the things that you're learning are things you don't and you're feeding that asset and it's constantly getting richer and richer.

Ryan: I think that that's a smart approach and by the way, it's intrinsically linked to quality because the only way we actually solve the problem in quality is to stop doing stupid shit with respondents that are human beings and expecting them to be engaged. 

Steve: Yeah, I couldn't agree more and it takes exactly back to that previous discussion because the thing about synthetic data is it is trained off real data.

Steve: So that means you need really good quality real data, really up to date real data. You can't be training synthetic data of stuff you collected a year ago or two years ago. It's got to be recent. And so I think any of you project forward to. The insight industry in two years' time, you can imagine a world where the client is doing survey research, not necessarily because they need the answers from those individuals, but they need the right training set of data to ensure that all of the modeling that we're doing with the data is completely up to date.

Steve: So you have a sort of balance of synthetic and real, and you're keeping the real as up to date as possible, managing, covering as many subjects and topics as possible in order to understand what's happening in the lives of real consumers now, and then extending that data across the entire organization for all of their business needs.

Ryan: So this brings me to my next topic. So when we, uh, we were starting our party, um, I joined you in 2014. I think you had started what 2000, late 2012 is when Zappi started. The original promise was faster, cheaper, better. And we then quickly learned that we were selling change, not technology. Um, and you know, I spent, I think I said this on this podcast, I spent a lot of time this year engaging with people on different sides of our ecosystem.

Ryan: And yes, there's a lot of technology, but a lot of the same stuff that was there in 2014 is still there today. And I think a lot of insights departments are still going from thing to thing. And a lot of what Steve is describing is. Moving from projects to your projects, creating a data asset, which gives your company an advantage.

Ryan: So AI is obviously a massive enabler of all the things we're going to discuss today, and it is going to be a commodity. These are open source models, but for businesses listening, Steve, to get value out of AI stuff, whether it's the previous topic or the stuff we'll talk about next in order to get value out of the capabilities AI enables, you must dot, dot, dot fill in your top three answers for me.

Steve: Well, I, I'm not even sure it's three. I think it's one. I think you need to become an expert in data asset management. Now, I suppose there is. Three words in that phrase. Um, so it's thinking about the data and the data quality. It's then thinking, not just about the data, but the data as an asset. So it's not answering a question now.

Steve: It is, it is almost on your balance sheet. It's what you collect. It's, it's what your, your organization. Is focused on and then it's the management aspect of that asset. And the management aspect of that asset is the stuff, you know, we're trying to work on making sure you've got a platform which is robust, which allows you to look across all of your data.

Steve: So you've collected the data, your then your asset management piece is making sure that that data is robust and actionable and available and democratized and, uh, can be used across the piece. We talk about meta analytics all the time, but the truth is, if you are going to be successful in a world of AI, the AI is helping you do specific tasks because it's looking at specific data and it's that data that has to be robust. It has to be well managed, et cetera, et cetera. So my view, I think, of where the industry needs to go is very much on the client side, thinking about the data asset management piece. And then on the supplier side, thinking about useful, interesting, exciting ways of using AI on that data to enable significantly better creation of product services, campaigns, all the rest of it.

Ryan: Data asset management. So it's why we think of all of the business moving from agile insights to connected data because it's where the world is going yet. I think I interviewed 15 chief data officers at best. The only consumer data they're bringing in today is Qualtrics NPS data. There isn't a chief data officer on the planet that's not trying to leverage its company's data asset and harness what they know to drive productivity, to drive scale, to put it in their own data warehouses so that it's safe and compliant, yet the lack of rich attitudinal data isn't even featured.

Ryan: And it's because we don't think of data asset management, and I have empathy for it, right? So the COEs are, it wasn't eight years ago, the COEs of the world were vet vendors. That was their job. Vet vendors in the high horse of New York City, London, Paris, wherever you live. And it still is today's paradigm that the insights manager on the ground is essentially an internal service bureau for the local brand management team.

Ryan: And I don't mean that to say we're all screwed. I mean that to say every day we've got to take steps towards this. So we have some customers that are enriching their data asset with all the attitudinal data they buy. And it is creating a competitive advantage for them. But I think my advice and, and, you know, if, in order to get it and you subscribe to what Steve says, you need to be focused on data asset management.

Ryan: Well, you need to make sure you're starting to get one. And, uh, one of our co-founders, Dave Birch used to always say this phrase: ‘If you want to get to the North pole, stop driving South’. And Dave has many great phrases, but that is a really important phrase here because. Each time you procure data, if you're intentional, it can become a competitive advantage.

Ryan: You don't have to go from, ‘Oh my God, I'm disconnected’ to ‘I'm a data asset manager.’ So, but everybody has a role to play in this. And it requires, in my opinion, quite simply Steve is intentionality. It's not, it's not rocket science. Um, but the intentionality needs to start today, or you're going to be left behind.

Ryan: Your organization will not have rich consumer data in its models. 

Steve: Yeah I was just going to agree. I think, I think if you start thinking of where you're going and then start plotting the direction. So in two years, in three years time, the head of an insight team will be very good mates with the head of data analytics and the head of IT or database management within that organization.

Steve: And that's probably not true now. But as you say that you just need to start moving in that direction, you need to know the vocab that these people use, you need to understand the systems that they're trying to engage with. And the truth is there are, there is lots of data around any form of organization.

Steve: But also a lot of that data has quality problems and a lot of that data is not very interesting. Uh, and I think actually our data stream, uh, is really, really interesting to people, and it's one of the primary problems with it. It's just too small, you know, so if you, if you're looking at customer database, um, for a, for a typical large enterprise, you know, there's millions and millions of people on it, and we have, hundreds and maybe often hundreds of records and thousands of records, not millions of records.

Steve: So, that's the bridge to cross. But it's now with these new capabilities, we can start crossing it. We just need to plan and take one step forward. 

Ryan: Plan, take one step forward. The previous discussion is synthetic data allows us to innovate on the way we buy data. So we can perhaps buy more of it or be more efficient with it.

Ryan: Um, but I think the other thing is like the intentionality of what you're buying aside, recognize that we let everybody decide what they do means you're getting 60 different data maps. 60 different sources of truth. Your data is not going to be in harmony. I mean, the average MR firm, every single project has a different data map.

Ryan: And that makes it very hard to run it through it, enrich it and to do all the exciting things that you can do. Um, so I, I think we need to think about repeatability in what we're doing and what are the things we use to answer certain questions and how do they speak and how do we leverage that UI?

Ryan: How do we bring that data into our own UI? And we need to start to upskill people within, within our client organizations to think like that, um, or to just have the enablement skills. Okay. So now that we've talked about what you have to do, let's talk about the problems that we think we can solve.

Ryan: So, I mean, some of the gimmicky shit you saw last year was AI to write surveys quickly. And that's, I mean, one of the interesting things is that guy Elliot, who's on our team, he's amazing, super smart guy. Every time he does demos, he does this thing is the magic AI or is the magic, the data. And I love that because, uh, like any great technological innovation, like this phone, it becomes, And then it's, that's actually the enabler, not the innovation.

Ryan: So from your perspective, Steve, what are the most exciting use cases and applications of artificial intelligence in insights and marketing? And what specific business problems do they solve?

Steve: Yeah. So I think the way we think about AI exactly, so we divide it into two spaces. One is internal. operations and efficiency, and then one is product improvement. And I think when we think about marketing, I mean, there's so many applications that are just making every aspect of your role within the marketing department and an insight department better.

Steve: But then there are also going to be ones that are transformational. And I think the ones that are transformational are really interesting. And, and, some of the things we are, we're looking at in terms of building new products or services or new ad campaigns using AI based on consumer insight data.

Steve: And if you start by thinking, can an AI look at my data asset? Which knows. So let's say you're kind of trying to make a new beer brand for young men in Texas. Well, your data assets should know about their brands, and it should know about young men in Texas and how they react to their brands and what they like about them and what they don't like about them and what they Think of the areas that they could be improved in.

Steve: And so using a I looking at that data, which is what we're beginning to release the capability of actually creating using that insight with AI to create a new beer brand for young men in Texas. Well, if you can do that in 15-20 minutes, I think what, what becomes really interesting, and I did this talk about potentially Consumer Insight doing a reverse takeover of marketing.

Steve: If we can, as insight people, use consumer insight data to start creating instead of just testing, um, then it can upend the process of innovation within organizations. So if you imagine now, if you create a new product or a new advertising campaign, it's a lot of effort and resources internally to create just one thing.

Steve: Whereas potentially we can. I was having this conversation with the client the other day. We could potentially create 50 ideas of a new product or service in one day, and then we could test them overnight. And now we've also released the capability for not just being getting a market research report back on your concept test, but actually taking that research report and aligning it with the original concept and saying to the AI, given the test results of that project that you've just done, can you rewrite and redevelop that concept? So you can create and test and optimize 50 different ideas within two days, get rid of the worst performing 25 of them and then move into your innovation workshop on day three with 25 ideas that are perfect.

Steve: Based on consumer insight and also been already tested and also been already revised, um, and being optimized. And I think that that to me is the most exciting area, but it's not just product. We could do the same in advertising cafes. We could do the same with packs. We could do the same with marketing strategies.

Steve: So we can genuinely use our data to be at the heart of how an organization, you know, creates and sells their products and services. 

Ryan: It's yeah, it really is important. I mean, I've been seeing some of the stuff we're doing with agents. I mean, the ability to put in your nuance, your brand guidelines, your political guidelines, your environmental guidelines, query that using agents to sort of deploy that querying everything you know, and then embedding You know, in our, in our advertising products, we've basically embedded the smartest guy in the room on advertising into every single report through a series of simple prompts, and then you can then cross reference that against all the data overlay, organizational priorities, and actually scale intelligence, um, which is something that in and of itself is cool.

Ryan: But then to take that into the creation space is, um, I mean, it's fascinating. I mean, we were doing this with a customer the other day, and I think in 15 minutes, we optimized their product five times. And it's again, the backboard NBA fan here, the backboard is your understanding of the consumer.

Ryan: So it's funny. I mean, Steve's already, uh, getting people calling, oh, I want to buy that. And he's like, well, you don't have a data asset. So like, what are you buying? And it's the same thing. Like when we first started doing Zappi, it's like, oh, we want to play whack a mole testing faster. Thanks for doing that.

Ryan: And it's never really the thing. Um, but again, going back to Steve's point on data asset management, um, you gotta have one in order to really unlock the benefits. Otherwise, you can have fun making surveys easier to write. Um, you know, and that, that's cool, I guess. Um, but, but now talk about the infusion or the line blurring between the richness of the data. So we used to talk about qual and quant. Um, I think we're both of the view that those lines are blurred. Um, so talk to me a little bit about that and how you see, uh, other forms of data playing in a more quantitative space, given that the AI enablement that we have in front of us.

Steve: Yeah, I think it's really. It's a really interesting area, and I keep on trying to have a thought experiment of it, knowing what we know about AI, how would you invent consumer insight? How would you reinvent surveys, given what we know about AI? And so I was talking to our data science team this morning. They've just done some work where we took the likes and dislikes from, from a product test and tried to predict purchase likelihood.

Steve: And it was really strong. We could predict it really well. So that's taking an open ended question and predicting the results of a closed ended question. So you then start thinking, okay, now the truth is, from a lot of, if you've got five closed ended questions, you can probably predict the sixth, the answer to the sixth question.

Steve: Closed ended question. But potentially with one open ended question, you can predict the results of six closed ended questions. So where does that lead you? It leads you in a way of thinking more open, more conversational surveys. It gives you greater insight, particularly in a large language model world, LLMs, there's clues in the name.

Ryan: Yeah, there's a word called language in it, but that's, that's helpful. 

Steve: Yeah. They really like language and they find language really inspiring when they're creating things. Now that's not to say that the close-ended things aren't important because the other aspect that you're feeding into the model is the stimulus itself.

Steve: So let's imagine a world where you've tested 100 100 products. Um, what you can say is if you tag the assets correctly, if you tag the stimulus correctly, you can say, okay, which, which are the top performing ones and which are the lowest performing ones and say to the LLM that the, the best performing ones look like this, the worst performing ones look like this.

Steve: Create a new product with that, learning, that insight, even without survey data. Um, and then you say, and why did the top performing ones perform really well. Uh, what did people say about them? What was the visceral reaction? What's the emotions it created? What are the unmet needs that they're meeting?

Steve: And what is it that they're objecting to in the bottom 10? And so you're taking the closed ended and the open ended. Against the stimulus itself. And all of that is just, I mean, incredibly rich data for the, for the AI system. Now, what is great about AI is you, in some sense, you could have done that type of project five years ago, but you would have required five really, really strong researchers, really smart data people, and they would have spent a month combing through a lot of data.

Steve: We can now do it in about two and a half seconds. And the reason it takes two and a half seconds is because, uh, the LLMs are getting hassled by so many people, uh, with, with requests. So it's an invidious shareholder. So you start thinking about what that enables you to do, and it just changes the game. Which goes back to the point you made earlier about behavior change is in this new world.

Steve: It is not just things a bit faster. We can. We can make it more efficient. Yes, we can make things more efficient, but we can also change the game and in order to change the game, the people who change the game the best, will be the ones who succeed in the next two, three years. They just, they just will.

Steve: We know that. In order to change the game, you have to do things that we've been talking about, like data asset management, but much more importantly, you need to have a behavior change process internally you need to have be surrounded by a group of people who go hey if we adopt this then we can do this in a different way we can we can approach the market in a different way we can approach innovation and approach communications in a different way and to be honest The technical problems, certainly the ones that we're looking at, we've either solved or are being solved for us by, Google or Microsoft or OpenAI or whatever, and we can use any of those.

Steve: So the technical problems are gone, increasingly going, because that's not entirely true on some, some of the, some of the space in terms of video, um, and, and I think, GPT 5 is going to be very significant. The rumors are it will be very significantly better than GPT 4. But those leaps and bounds are happening.

Steve: You then need to do leaps and bounds in, in the, in the way that you run your company, the way you run your department. You have to think of this as a seminal moment to rethink the world that you're in and the way you approach things. Otherwise you will get rapidly and very rapidly nowadays left behind.

Ryan: Yeah, like pretty quick. And I don't think Steve's word shouldn't be taken as overwhelming because it's not that hard to get started. Audit what you're doing. Be intentional about the data you're collecting. There's a macro nugget of gold in what you say about text. Text can predict eKPIs. We've been analyzing, even in our own advertising product, we probably have six metrics that say the same thing.

Ryan: But people, people are used to their old metrics. So, be intentional about your stack and how it's collecting, but also be bold enough to say, the answer I used yesterday was good. I'm willing to correlate it and move to something that is more modern. So I can leverage what we know about panels, so I can infuse more text, so I can ask shorter surveys, and so I can spend more time on the data asset moment that's there.

Ryan: You can expect more from us on this topic. We don't want any of you to get left behind. We want to help you, whether you decide to do business with us or not. 

Ryan: Steve, myself, Steph Gans, Kate Schardt have written a book called the Consumer Insights Revolution. It drops in September. We're going to give you the playbook on how you get from whack a mole to data asset management to effective ads, effective growth strategies. And my next guest's name also starts with an S. It's Stephan.

Ryan: And we're going to talk about how he did it. So stay tuned. It's been a great season so far. Steve, it's good to have you back on again. Thank you for joining on a Friday evening.

Steve: Great fun. Thank you, Ryan. 

Ryan: Thank you everybody for listening.