Pete Bishop: hey Ray, how's it going? ⁓
Ray: I'm good, Pete. It's great to talk to you today.
Pete Bishop: Yeah, it's nice to catch up again. I've I've I keep saying this over and over, but it's been an excuse for me to kind of make friends Sabah introduced us, And he was on epis ⁓ season at some point. We were at Inventures I was speaking him and ⁓
Ray: Yeah.
Pete Bishop: we had this like little closet where we were doing the podcast. I was so embarrassed 'cause the lighting and the sound was terrible. So I have to get ⁓ back at some point.
Ray: been making the rounds. ⁓ travels in very different classes than I do. And so he just back from Cannes where he was rubbing shoulders with all the, know, all the hoi polloi in the marketing world. And I can only dream of that kind of thing.
Pete Bishop: yeah, I know. But he works with you, right? At Arima. Okay.
Ray: That's correct. Yeah, ⁓ he came and joined us earlier this year on a fractional basis, but we love his enthusiasm and his imagination and his ability connect dots. It's wonderful.
Pete Bishop: ⁓ okay. enough about him. How about you? So how long have you been ⁓ Arima now?
Ray: I joined ARIMA three years ago ⁓ after a life in the corporate marketing world. so the corporate marketing world, both in research and data and CRM at banks and insurance companies, I spent a good deal of time at Ipsos. And probably, you follow on questions. Well, how did you come to be at ARIMA? So.
Pete Bishop: Okay. Mm-hmm.
Ray: About three years ago, I was having lunch with a really good friend of mine, a long time friend of mine. It was one of our first lunches after COVID. So I'm guessing it's actually longer than three years ago. Now, maybe it was four. And it was a beautiful sunny day. We were sitting outside on a patio because that's what we did right after COVID. didn't eat indoors yet. We were sitting outside on a bright sunny day. And friend...
Pete Bishop: Right.
Ray: Chris, he says, I met this really interesting guy yesterday who I think you'll be very interested in. he said, I met this guy, Winston. He has modeled a synthetic version of the Canadian population for marketing purposes. And I ⁓ stopped with fork halfway to my mouth and I said, come again? And he said, Winston has modeled and created a synthetic population of Canada for marketing purposes. And I said, wait, that means you could, and Chris said, yup. And I said, and you could also, and Chris said, yup. And I said, that changes everything. And Chris goes, yup. fast forward to six after that, I, you know, I learned a little bit more. about what Arema does, little bit more about Winston and his background. I met Winston, we talked and we chatted and you know, six months later it was like Winston, I need to join you to build the future and to build this thing out because you know, I closer to than many maybe of your listeners are and I thought myself with the remaining useful years of my life, I want to build something for the future. and be part of that something future oriented. And so I dropped it all, left my cushy corporate life and joined the startup world.
Pete Bishop: my goodness. Yeah, the chaos that is startup. ⁓ ⁓ had a similar I I can't remember when I was introduced to this concept, but that notion of synthetic data and just I I couldn't believe it. And it w it seemed to happen early in at least at least in the public eyes. but do you mind just a quick brief explanation of what that is? 'Cause I think the concept's really interesting.
Ray: Hahaha! Sure, thanks for that. Yeah, let me back up half a step from that. who works with data knows that there are increasing in problems with data. And so the marketing world talks a lot about data-driven decision-making, and yet the quality of the data on we are basing our so-called data-driven decisions ⁓ is it's deteriorating for a number of reasons. One is the pure research world, so the primary research world, survey research or polling world, there a lot of respondent fatigue. And so there are ⁓ increasing demographics, incitements of the population who are not responding to surveys and are very, difficult to get, or
Pete Bishop: Right.
Ray: there is a preponderance of bots in marketplace where you send out a survey and you can actually subscribe to a bot to answer that survey. And that's just random responses or what in the research community is called straightlining. And so there's a quality deterioration in survey research which many decisions are made. And then even first party data, data that is owned by loyalty programs or by the banks or by, know, Tim Hortons or whomever. There are limitations to that as well, because that data is limited to people who own the app or people who choose to participate in the loyalty program. therefore by definition, your customers only. And so there's a narrowness there.
Pete Bishop: Right. remember I remember like we used to do a lot of UX research back in the day and we got introduced to the concept professional responders, like people who kind of did lot of surveys and a lot of that the point where they become invalid because they are so good at answering questions where you're trying to kind of unexpected answers, you're getting people who are almost like pros at answering. Is that part of this as well?
Ray: That's part of it as well. And those professional responders sometimes will also be really good at sussing out what it is that they might want you to hear. And that's always been the case. And really good research firms have ways to mitigate these. But they are getting more more so. And as the weight of decisions increases and as the
Pete Bishop: Yeah. Yeah.
Ray: you know, the scrutiny on ROI and as the scrutiny on data-driven decision-making increases, you know, the quality of the data that we have to make our decisions and the connectedness of the data that we have to make our decisions is falling a little bit short. And so this is where synthetic data comes in. In its purest form, synthetic data is basically modeled data. where real-world data doesn't exist. ⁓ a simple example might be, Pete, you you are, I'll say you're a 40-year-old male who lives in who is married, ⁓ and ⁓ let's you have two kids, I don't know how many kids you have, but let's say you have two kids, and we don't know what kind of car you drive, but we can model based on information that we know that you might own a sedan or a minivan or a SUV or something like that. in purest sense, what synthetic data is, saying, you know what? Pete's household is likely have an SUV. We don't know for a fact that you do. And ⁓ actually the other reason why primary data is ⁓ starting to is people's concerns over privacy. because Pete Bishop may not want us to know that you have an SUV and maybe a Ferrari somewhere in the garage, you know, for privacy reasons. And while we may not know that you have a Ferrari, you know, we can imagine that you might, you're more likely to have an SUV than, than,
Pete Bishop: Right. Right. Right.
Ray: Paul Bishop, who is of a similar age, lives in a similar area, but is a construction worker. And so we would imagine that Paul Bishop has a pickup truck and Pete Bishop has an SUV. We don't know that for sure, but that's what synthetic data is. It is a probabilistic model of only car ownership, but behavior, store ownership on real world data.
Pete Bishop: off, I'm really happy that you picked me out ⁓ as forties. that's flattering. And you're pretty accurate with the ⁓ I could see that really working The amount of data must have to feed into this must be pretty significant for that to feel like a real audience. Is that fair?
Ray: Yes, is fair. it one of the reasons why synthetic data is, I will getting to be, there's a lot noise around it now because just now the computing power and the computing algorithms to make it possible are just coming on stream and online now. Theoretically, it's all, it's been possible for many years. And it used to be called what was called data stitching, where you would take two surveys or 10 surveys, or in the case of a REMA 26, different survey data pieces, and you'd stitch the data together in a probabilistic way, like a really large Sudoku. So, you know, more or less, know, for who does Sudoku, you know that the rows and the columns need to add to 10 in each row and in each column, you can only have one, one, one, one, three, one, four, one, five, whatever. And so,
Pete Bishop: Mm-hmm.
Ray: You know, we know that in a postal code, there's a Pete Bishop and there's a Paul Bishop, both of the same age. One's a construction worker. One's an advertising executive innovator. And we know that from StatsCan that there is one SUV and one pickup truck in that postal code. And so we're going to guess that Paul Bishop owns the pickup truck and Peter Bishop owns the SUV. Now that's a very simple example. But you multiply that by 26 data sources, 60,000 variables. No human can do that. You need to probabilistically do that using an AI algorithm.
Pete Bishop: it's fascinating. So, you know, we talked a lot about the kind of the disadvantages of, you the the way of how we're doing it and then maybe some of the fatigue that's happening. For this, there be a lot of advantages beyond just ⁓ pricing and speed, There must be like I f f me, ⁓ the fatigue you could ask this group at nauseum, I expect, as many questions as you want, as often as you want, right?
Ray: Yeah, absolutely, 100%. And you know, one of the interesting things, we just did a test with a research partner. One of the interesting findings out of this test is that in the real world, you might say, do you think that Canada's inflation rate is going to go up or down, up or down, or I don't know? And a lot of people will answer, I don't know.
Pete Bishop: Right.
Ray: in a synthetic research world, you're not allowed to answer, don't know. model will always assign a yes or a no. And so you actually get much more, you get more completed ⁓ so advantages other than cost and speed, which you have outlined already is I'll call it data So in our hypothetical example, we have in Arima's case, we have 60,000 behavioral, psychographic, demographic ⁓ Pete Bishop or the synthetic Bishop and the Paul Bishop. You might recall that one of the disadvantages to traditional data is that I might have Pete Bishop's attitudes towards the Alberta government. And over here, I have Paul Bishop's attitudes towards pet ownership. But I don't have Pete Bishop's attitudes towards pet ownership or Paul Bishop's attitudes towards the Alberta government. But in a synthetic data situation, you actually have all variables for all synthetic individuals, which allows for really cross category explorations innovation use cases. ⁓
Pete Bishop: Right.
Ray: for new product concept in use cases. And so, you we had a use case where a candle manufacturer came to us and said, we really need to look for males of this gender who use candles. Well, there's no survey talks about candle usage amongst young males, for example, but our data has that. Or interesting example, a real life example is that one of ⁓ clients is a outdoor manufacturer who wants to come out with a line of pet clothing. So, you know, your pampered pet can walk down the street in the middle of winter with a ⁓ down jacket by this clothing manufacturer, but they had no data whatsoever their customer base or even their prospect database on pet ownership. what kinds of pets were owned, where they shopped for pet supplies, how much they spent on pet supplies. we had all of that. We have all of that information because we have 360 degrees worth of data on every synthetic individual in our population.
Pete Bishop: So just to dig into that, how is how is that? Like so you mentioned the candle, right? Like at some point, is just because you can query this with anything you want, or it's it's been queried so much that it actually has that data already in it? Like how does it how would it know about candles of that data? I'm just trying to wrap my head around it.
Ray: It's a little bit of both. So we buy a lot of usage and attitude data to the synthetic population. So whether it comes from, ⁓ you know, Numeris or whether it comes from Vivid Data or whether it comes from a media company, whether it comes from the Canadian government, the University of Calgary, wherever. And there's two parts to what we do. We buy a lot of source data.
Pete Bishop: Mm.
Ray: And that source data covers a lot of ground. using proprietary algorithm, we take, let's say it's a survey of 50,000 individuals, we then probabilistically create our database of 30 million for Canada for those variables. But we don't cover it all. So we've actually also developed an algorithm which allows us to query and ask the synthetic population. about variables that we don't have. So for example, at the height, when the tariffs were starting to, the US was starting to impose tariffs, I guess it was just about a year ago, or maybe a little bit more than a year ago, many of our clients said, we need to understand what our audience's reaction to those tariffs might be based on these parameters. And so using our ⁓ we are able to get a sense of that the population within ⁓ a very short amount of time. Now it's modeled for sure. It's not actually going out and asking people question, but it's saying somebody votes conservative, believes ⁓ in or the family value of, ⁓ family's the thing, then they're more likely to answer yes to tariffs than no to tariffs. I'm making that string up. But you can see that there's a string of logic. It's actually called a multi-step reasoning approach. that says if this, then this, if this, then that, if this, then that, if this, then yes to the question. Much more complicated than that, but that's kind of the idea. And so you can fill in the data gaps as well. And it self-perpetuates because the more you have, the more data points you can triangulate to fill in the missing data, so to speak.
Pete Bishop: Right. Ugh, that's brilliant. So we talked we talked a little bit about like obviously the upside. Is there a downside to this or does this replace like research or it is ⁓ is a balance or how do you see this unfolding as it ⁓ kind of moves ahead?
Ray: I think there's a couple time horizons to your question, and I think your question's a really, really good one, and it's a little bit untestable. ⁓ recently did a test with a partner where what they did was ⁓ actually a live survey, and then they compared their live survey results against the same questions asked of our synthetic population. By and large, there's...
Pete Bishop: Right. Right.
Ray: There were four criteria. was ⁓ a very well-grounded test. You know, the upshot of it is like, well, you know, the company that the test, they were like, wow, we were really impressed with how good the answers were. Now, there were limitations. You know, if you very future-oriented questions, like, do you think you're going to get married in the next three years? Those answers ⁓ versus the real... live answers were not as, you know, not really comparable because the data we have by its nature is hindsight looking less foresight looking. You know, there were language issues, there were, there were various issues, but, and, but we are at what I would call, you know, version two of synthetic data right now. And, you let's say in five years, ⁓ we be at version five.
Pete Bishop: Right.
Ray: And in seven years, we might be at version 17. And so in that, what did I say, seven year span, the use cases are going to get so much better. They're going to, it's going to get so much more accurate. Today there are outages, there are known outages and there's clunkiness, but as you already identified, it is a self-perpetuating model. more it... It's a machine learning model. The more it learns, the better it gets. The more data it has, the better it will learn going forward. Will it replace primary research? I don't believe so. Because primary research is always the seed. You know, the spine of the Arema synthetic population is the census data, which is every four years. We can't build it without census data. We can't build it without some actual starter
Pete Bishop: Yeah, right. And Yeah. Right.
Ray: don't do sourdough, you need a sourdough starter to get it going. Otherwise, you're just it up as you go along. There will always a place for primary research, though will always be a place for certain types of primary research. is an augmentation and is a complement rather than a replacement.
Pete Bishop: Ha ha ha. how do you kind of keep bias out of this? Like I know, I think that's true of any research, right? And anyone you're talking to, but in especially if you're if you're using research that's already been done, is there is it the problem that you would have anywhere else or is it is it different because of this new model?
Ray: know, bias, like non-response bias or biases, which all of humans have, those are present in any type of research. And there are to against that. They do get, they can get magnified when you are taking, let's say a sample of 10,000 and you're making it 30 million, because some would argue that what you're doing is you're trying to...
Pete Bishop: Right. Right.
Ray: that you're actually magnifying the bias. And there is some to that. There's two mitigations though. One mitigation is that we use multiple data sources. And so while there are systemic biases in things like non-response bias, you actually do lessen that by using multiple data sources. Secondly,
Pete Bishop: Right.
Ray: The models that we've created, actually we create some randomness into the to try to ⁓ that. Now, there proof that it does or does not theoretically, yes, but is there ⁓ proof that it does? No. so in the simple example that we have now, we know that Pete Bishop is an advertising executive and so probably owns an SUV and Paul Bishop is a construction worker, so the model would definitely say that he would probably own the pickup truck. But when we introduce randomness in there, it will say, but no, you know what, we're going to give Pete the pickup truck and Paul Bishop the SUV because we want to introduce that randomness into the model to reduce any kind of systemic bias or other systemic issues that might be present in the source data.
Pete Bishop: yeah, that makes sense. Is ⁓ is just for big business? Like I I'm just trying to get a sense of scale. Like, can any kind of use ⁓ synthetic at level or does is this kind of big problems only and big money only kind of thing? ⁓
Ray: I guess it depends on who you ask. And there are many versions of synthetic data. approach has been to, we're a bunch of data geeks who've spent our time in this world for many years. And sometimes we say to each other, we're building the tools that we wish we always had. And so the data and the tools we've built on top of this to access that data, because it's really important to access the data.
Pete Bishop: Yeah.
Ray: are available to anybody and are built for non-data scientists. And we have a platform that sits on top of our data, which is really easy to use and yet really powerful. The analogy that I will draw is that the platform that we have is similar to our do-it-yourself tax platforms these days. You go it or Wealthsimple or whatever to do your taxes, you input your data,
Pete Bishop: ⁓ yeah.
Ray: You know, there's a bot that pops up and says, you know, if you increase your RSP contribution, you could save this much. Or if you gave that much more to charity, you could do this, that, or the other. And there's a very dynamic iterative interface. And we don't need to know tax law to file our taxes anymore. We allow the AI and the platform to guide us. And so we have built a platform on top of our synthetic data that guides and helps the users through the process to answer whatever question they have. whatever questions or whatever marketing problems that they're trying to solve without an analytics background. And so they can model cross-media plans without really needing to know advanced statistics. We have a GIS module which allows for mapping of traffic patterns without really needing to understand the principle of centroids. and of how the postal codes overlap or don't overlap. It just does it for you. It just helps you do it, let's put it that way. So it is accessible to everybody. We've tried to make it accessible to everybody. ⁓ in fact, most of our clients, and I don't wanna shill too much, but most of our clients are small to medium size agencies and small to medium size brands. Literally, I just spoke to a...
Pete Bishop: Yeah.
Ray: a brand new client just before coming onto this call. And she wants to activate 15 to 20 of her, of her, her team, everything from a media buyer to a junior analyst, to a, know, to their head of stats, everybody to use it so that it makes the data and data driven decision-making accessible to all. Whereas you know, it was always a great theory, but in practice it was not. Not so.
Pete Bishop: Yeah, I it it you know, it's funny 'cause when you hear it all explained, it's it feels like such a noble brainer, but I also get why people are so reluctant 'cause it's it's a big mind shift to s you know, for the people who are thinking, you know, the robots are taking over everything and This is a definitely one of those moments where you're like, ⁓ my goodness, we're all getting replaced by robots here. But can really see the advantage of doing this type of thing and like you said, you know, removing some of that fatigue. I remember going doing surveys where like, wow, we can't ask more than fifteen questions because people just get tired. So, you too bad. it's like, Well, you could ask a thousand questions ⁓
Ray: And you know, in post COVID, I'm in attention spans, attention spans, like, I don't know about you, can't sit through a 10 minute YouTube video anymore. Give me 30 second version.
Pete Bishop: Right. No, it just seems fascinating. So do you feel like that's the the the barrier in thinking, just getting your brain rewired for this moment? what would what would be some of the like reasons why people wouldn't do this, I guess, or that you hear of like what are the common reasons they're like, ⁓ about this, what about that?
Ray: Yeah. And there's a lot of what about this and what about that. And it's valid. The main one is like, how valid is this data? it is a mind shift for sure. you know, the first EVs were introduced in 2002. And I remember hearing the first electric vehicles, battery vehicles, and I was like, ⁓ that's awesome. Well, but I have a car already that I just bought three years ago and What about charging infrastructure and what about reliability and what about this and what about that and change takes time, period. Change takes time and this is a big mind shift as you have already identified. And so will take time. know, the early adopters will adopt and they'll create use cases that will, you know, that will that will convince the mass and then the mass will, you know, then convince the late adopters and, you know, so the cycle goes with anything new. My advice to listeners, my advice to everybody though is, is got to jump in and just start to use it. It doesn't have to be in a big way. You don't have to base multi-gazillion dollar decisions on it.
Pete Bishop: Mm-hmm.
Ray: But you need to be familiar. There are so many different sources and types of synthetic data. You need to understand them. You need to understand the use cases that are right for you and your organization or you as an individual because it is moving quickly. And you you use the analogy before when we were talking, Pete, you know, if you drop an iPhone 17 on somebody to begin with, you would be lost because you wouldn't know how to use the functionality. But We started with iPhone twos and then they went to iPhone threes and then we went to iPhone fours and everyone got incrementally better until we're now at, I don't even know what we're at now, 17 or 18. ⁓ And the advanced, you know, early adopters are using those, well, or Discord or using Discord, you know, in ways that us Luddites can't even imagine. at the leading ⁓
Pete Bishop: Yeah, I think so.
Ray: But if you don't get started, then you are going to be behind the curve because the early adopter companies are embracing it in a very big way and are gaining meaningful competitive advantage from it, from synthetic data.
Pete Bishop: Yeah, it's it is fascinating. I don't know if this will make the podcast, but I do need to talk to you, I think, a little bit about travel research, because I'm in the middle of it and I would love your perspective on it. it's just kind of like the new travel behavior using AI ⁓ and how they're booking differently.
Ray: Yeah. Well, and we actually have two cruise companies, destination X, like destination Toronto, destination ⁓ types of, I guess those are tourism boards as clients. You know, one of the reasons why I think I might've, so Arima has a population of Canada, a synthetic population of the U.S., a synthetic population of Germany, and a population of the UK. ⁓ And the reason we built a synthetic population of Germany and the UK to start with, because they may seem like odd choices, was that one of our clients said, we need to target tourists from those countries to come to Canada, and we need the synthetic population to do that. One of the other things that we're doing is building a synthetic population of fans. So sports fans. sports fans are not just sports fans. They consume on media. They consume gaming. They travel. They consume in stadium.
Pete Bishop: ⁓ right.
Ray: you know, they buy merchandise, they are definitely multifaceted and understanding that group is, some thing that one of our very large clients has identified. And so they are funding our development of that database, but travel is definitely one of those pieces as well. When, when we think about travel, gosh, my eyes were completely wide opened wide when I saw the travel. ⁓ behavior around this World Cup. had appreciated before, never appreciated before the extent to which people will travel to follow their team during the World Cup. And, know, and so that's
Pete Bishop: ⁓ I believe it. Yeah. Like fan fanatics.
Ray: But maybe not even when I saw the pictures of the Tartan army in Scotland and the number of people in Scotland or witnessed the number of people of the last game was the Portuguese in Toronto and they closed down the 427 because they were jamming up the 427 waiting for Ronaldo and the team bus to drive by. It's like, what the heck? This is crazy, man. ⁓
Pete Bishop: was I was in Scotland when they won their first game, which they were not expecting to win any of them, and th they gave everybody the day off the next day 'cause they knew that no one was gonna be able to work.
Ray: Well, or just to see the pictures at the Red Sox games of them taking over the stadium. I mean, was so much fun. It's not over yet, but it's so much. It's over for Toronto anyway, but it's so much fun. It's over for Canada too, I guess.
Pete Bishop: Yeah, I imagine that There must be so much data pouring out of those things too, just like realizing how like I thought it was really neat to see how Canada got on board this year. not like they haven't in the past, but this year it feels especially big. it's just like what's driving that and why are we so excited about it this year? Obviously hosting, but you know, there's lots of other things I'm sure at play. It's ⁓ yeah, it's it's it must be a fascinating job just in general, watching trends and seeing what's what's ticking and what's not. There's all sorts of insights I imagine you get.
Ray: know, one of the things that we say to our detriment is that use cases are limited only by imagination. And to our detriment because people are, you know, in this type of world, people are not good at imagining things. You know, we couldn't imagine where the internet was going, you know, when the internet, when we first heard about the internet, but, and this is kind of the same, but imagine, and we do, have mobility data. on top of our synthetic people. And so now you can understand travel patterns, time of day where people, you know, spend their evenings. We have a golf course client who use data to model before and after the round. Where are people going before the round? Where are people going after the round? ⁓ So we can understand. Actually, we an auto dealer who does the same, models the
Pete Bishop: Mm.
Ray: before and after what dealerships are, our audience visiting before they come to our dealership and what audiences, where do they visit afterwards and where are they going? And that way we can better shape our marketing messages and our out of home placements and ⁓ offer, our better because we know they're visiting Toyota before they come to us. And so we can target Toyota. mean, that kind of thing is possible when you have synthetic population data.
Pete Bishop: head's spinning right now, and it's not just because of my hangover. ⁓ the you it it it ⁓ I do it's I I love the thought around the only limit. I've said this a lot. Like the only limit right now is your imagination. I it really feels like that with the tools that we have at our disposal lately. Is I I feel like I feel like I am
Ray: It blows your mind. You
Pete Bishop: constantly not thinking big enough because you can you can do these things and this is such a great example. And I think it's been around for a while. Like I we did talk a little bit around you know this notion of creating a persona and then asking it questions. I've seen people create like their board of directors, like a virtual group of them and feed them all their personality you know ticks and everything like that and then present to them and try and like work out some of their problems ahead of time before presenting to their board. Like things like that. feels like a like that steroids is what's been built here, right?
Ray: You that actually brings up a good analogy, I think, Pete, you you asked about will primary research ever go away. And so let's, let's call synthetic data a flight simulator. And, know, modern flight simulators, you can spend hours and hours and hours learning how to and you can pretend to crash the plane and you can cut out engines. You can do all kinds of scenarios in a very safe environment. but nobody's going to give you a pilot's license having you be in a real plane, flying in real world and having a certain number of hours in a real plane. So synthetic data is kind of the same. You can play all you want, but at the end of the day, I think most people will still want to test, know, test the new packaging in a test market or, you know, try out a new, you know, test communications A versus B or, you know, you know, test in the real world before you give the pilot's license, right? Air Canada is never going to give somebody a pilot's license who's never flown anything but a flight simulator, but the flight simulator is so valuable for training. It cuts the training cost, it cuts the training time, it increases the value of training, but all in a very safe environment. so data and we call population simulators are the same. They're just... place that you can. test and accelerate your your bad ideas out ⁓ all of those kinds of things.
Pete Bishop: Mm-hmm. ⁓ that's a great analogy. Yeah, I love that. It does feel a safe spot to ask stupid questions or or find probably ask questions than you would have, right? Because it is so accessible and and probably a little more efficient to try things out and experiment and maybe get to find those and insights that you would never have otherwise found because of the time and effort it would take to to do that, right?
Ray: it's synthetic populations and they're not allowed to say, don't know, you could find things that you may not have ever occurred to you before, right? ⁓
Pete Bishop: Right. yeah. No exactly. let's on question 'cause I'm I am curious about this. Just if you can fast forward five years from now, where's wh where do think Arima's going? Where where is all of this going, do you figure? Like what is it what does it look like?
Ray: I think that it will be much more wide, things like what you said, like recreating the board of directors ⁓ ARIMA was asked to recreate the house of representatives in the U S I think those are extreme cases. there are data integrity reasons why those small audiences will not get created, but the use of synthetic personas and particularly the use of synthetic populations will become, I think a lot more commonplace than they are now. Earlier this year, May, Gartner came out with a report that basically said that by 2030, 80 % of companies globally will use synthetic data at scale for decision-making, up from 10 % today. it will accelerate very quickly. It will become more commonplace. It will never, never replace primary research because you'll never give somebody a pilot's license without them actually flying a plane. But it will become more commonplace and the of use cases and the quality of the data will just only get better with time.
Pete Bishop: Mm-hmm.
Ray: Now my advice to everybody, as I said before, always just get started now because 2030 is only four years from now. And that's not very long.
Pete Bishop: Yeah. and like you said, there's no pilot these days that doesn't have to spend a certain amount of time in that simulator, right? So it's become commonplace. but all these things just take time and once the once that little You know, the fear of this or the anxiety around it kinda you get over that little hump, then it grows so quickly, right? And th there's like we see this all throughout our careers for the most part. You saw when the internet came in. I remember to pick on Scotland again, but there at one point ⁓ the government decided to put a computer with a a modem into every household in Scotland. And a lot of them just looked at it like the devil's box and it just sat in a corner and no one used it. ⁓ Right? But once you know you get over the hump of like, ⁓ okay, I get this now and it's starting to feel normalized, then you just see it kind of connect and swing up and it feels like this is in that moment, right? Like it's it feels early days, it's not common knowledge and a lot of people are probably loving it and running with it, but it hasn't quite ticked over in a lot of people's minds that this is where it's going and ⁓ you know, we need to be part of it.
Ray: Yeah. Going back to, once again, going back to the flight simulator analogy, you know, in the early days of flight simulators, the scenery was really blocky and the trees didn't look photorealistic and the cities were not properly rendered. But, you know, the latest iteration Microsoft Flight Simulator, it's like, like that's my city. I can fly over my house. And it you know, you can set as many guardrails or non-guardrails as you wish to simulate flying and they just have gotten so much better and synthetic data for marketing purposes is going to do the same. It's just going to get so much better over the next five years and that photorealism we'll call it will just improve vastly.
Pete Bishop: My brother, he had the flight simulator on the Commodore sixty four and he would literally spend 'cause I think it was almost real time, it'd be like, ⁓ if you want to fly over to Europe it's gonna take seven hours he would start it and it would like literally take that long and he would just be staring at these two colors with a line in the middle. I was like, How are you doing this? This is crazy ⁓
Ray: Yeah, exactly. now you can look at the ⁓ know, all kinds of crazy stuff.
Pete Bishop: Yeah. And all the dashboard and everything, you probably dial in whatever model plane you want and all that stuff. Yeah, I can just you can probably nerd out heavily on this. But it goes to show just the how that stuff changes grows and I can just imagine where you're at in the next little while. But it's ⁓ yeah, fascinating. It's like honestly still wrapping my head around it, but I just feel like it's a secret treasure trove that everyone needs to know about. So this has been fascinating.
Ray: Yeah. You can't. It's been, it's exciting. It's an exciting space to be in.
Pete Bishop: Yeah. Well, thanks much, Ray, for spending some of your time. I know you're busy today, but I I appreciate the chat. and hopefully we can connect soon.
Ray: a pleasure Pete, I'm glad you reached out and I really enjoyed our chat, thank you. You know, I think there is one thing that I want to say is that I think we're still at the stage of buyer beware. And so there are synthetic data sources with do which with with what I will call dubious provenance. And so there are some synthetic data sources where they will say, well, you know, you just tell us who your target audience is. Females, 35 to 50 who have body conscious images.
Pete Bishop: Mm.
Ray: a negative body conscious image. And they will then, you know, they will have an interface that allows you to ask questions of those, but you don't know where that data is being drawn from to give you those answers. And so there is very much a buyer beware mentality that I think users synthetic data have to have today still because, you know, ⁓ knows the term ⁓ hallucination and so there is synthetic data hallucination. If you don't know the source data, if you don't know how the persona the population is made, you don't know where your answers are coming from, it is not a good thing to be just blindly trusting them when you have multi-million dollar decisions to make based on the ⁓ outcome.
Pete Bishop: Right. You know, I feel like that's good advice for right now, just understanding your data. Like there's a lot of people especially when you're dealing with even anthropic or open AI, just understanding where were things trained, where is this stuff coming through, right? I had a friend that was I Hopefully not repeating myself, but I had a friend who was every time he was getting a response out of chat, it was it was saying something like, Thanks for tuning in on the end of all this stuff. And he's What is this coming from? Or and he's like, Yeah, it's because it was trained on YouTube data. So most of the time it's talking about episodes, and that was kind of how it would sign stuff off. So when you're like that's where from, right? And you just don't realize, ⁓ that's right. Yeah, it's just it is a product of its data. And if the data is garbage in, garbage out half the time, right? So it's really important to know Know where it's coming from because think these days that's just part of being a literate in AI. So and this is no different by the sounds of it.
Ray: It's no different. if the person providing your synthetic data or the synthetic data source is a little bit vague and cagey about what the data source is, then I would say run away, right? You want people who are open and you need to understand because there's, you in a, you know, particularly for, you know, agencies or brands or whoever, like I said, there's, there's big money decisions at stake here ⁓ and you to
Pete Bishop: Yeah. Yeah, and your reputation's on the line, right? Still is, right? No matter what.
Ray: Mm-hmm.
Ray: I'm good, Pete. It's great to talk to you today.
Pete Bishop: Yeah, it's nice to catch up again. I've I've I keep saying this over and over, but it's been an excuse for me to kind of make friends Sabah introduced us, And he was on epis ⁓ season at some point. We were at Inventures I was speaking him and ⁓
Ray: Yeah.
Pete Bishop: we had this like little closet where we were doing the podcast. I was so embarrassed 'cause the lighting and the sound was terrible. So I have to get ⁓ back at some point.
Ray: been making the rounds. ⁓ travels in very different classes than I do. And so he just back from Cannes where he was rubbing shoulders with all the, know, all the hoi polloi in the marketing world. And I can only dream of that kind of thing.
Pete Bishop: yeah, I know. But he works with you, right? At Arima. Okay.
Ray: That's correct. Yeah, ⁓ he came and joined us earlier this year on a fractional basis, but we love his enthusiasm and his imagination and his ability connect dots. It's wonderful.
Pete Bishop: ⁓ okay. enough about him. How about you? So how long have you been ⁓ Arima now?
Ray: I joined ARIMA three years ago ⁓ after a life in the corporate marketing world. so the corporate marketing world, both in research and data and CRM at banks and insurance companies, I spent a good deal of time at Ipsos. And probably, you follow on questions. Well, how did you come to be at ARIMA? So.
Pete Bishop: Okay. Mm-hmm.
Ray: About three years ago, I was having lunch with a really good friend of mine, a long time friend of mine. It was one of our first lunches after COVID. So I'm guessing it's actually longer than three years ago. Now, maybe it was four. And it was a beautiful sunny day. We were sitting outside on a patio because that's what we did right after COVID. didn't eat indoors yet. We were sitting outside on a bright sunny day. And friend...
Pete Bishop: Right.
Ray: Chris, he says, I met this really interesting guy yesterday who I think you'll be very interested in. he said, I met this guy, Winston. He has modeled a synthetic version of the Canadian population for marketing purposes. And I ⁓ stopped with fork halfway to my mouth and I said, come again? And he said, Winston has modeled and created a synthetic population of Canada for marketing purposes. And I said, wait, that means you could, and Chris said, yup. And I said, and you could also, and Chris said, yup. And I said, that changes everything. And Chris goes, yup. fast forward to six after that, I, you know, I learned a little bit more. about what Arema does, little bit more about Winston and his background. I met Winston, we talked and we chatted and you know, six months later it was like Winston, I need to join you to build the future and to build this thing out because you know, I closer to than many maybe of your listeners are and I thought myself with the remaining useful years of my life, I want to build something for the future. and be part of that something future oriented. And so I dropped it all, left my cushy corporate life and joined the startup world.
Pete Bishop: my goodness. Yeah, the chaos that is startup. ⁓ ⁓ had a similar I I can't remember when I was introduced to this concept, but that notion of synthetic data and just I I couldn't believe it. And it w it seemed to happen early in at least at least in the public eyes. but do you mind just a quick brief explanation of what that is? 'Cause I think the concept's really interesting.
Ray: Hahaha! Sure, thanks for that. Yeah, let me back up half a step from that. who works with data knows that there are increasing in problems with data. And so the marketing world talks a lot about data-driven decision-making, and yet the quality of the data on we are basing our so-called data-driven decisions ⁓ is it's deteriorating for a number of reasons. One is the pure research world, so the primary research world, survey research or polling world, there a lot of respondent fatigue. And so there are ⁓ increasing demographics, incitements of the population who are not responding to surveys and are very, difficult to get, or
Pete Bishop: Right.
Ray: there is a preponderance of bots in marketplace where you send out a survey and you can actually subscribe to a bot to answer that survey. And that's just random responses or what in the research community is called straightlining. And so there's a quality deterioration in survey research which many decisions are made. And then even first party data, data that is owned by loyalty programs or by the banks or by, know, Tim Hortons or whomever. There are limitations to that as well, because that data is limited to people who own the app or people who choose to participate in the loyalty program. therefore by definition, your customers only. And so there's a narrowness there.
Pete Bishop: Right. remember I remember like we used to do a lot of UX research back in the day and we got introduced to the concept professional responders, like people who kind of did lot of surveys and a lot of that the point where they become invalid because they are so good at answering questions where you're trying to kind of unexpected answers, you're getting people who are almost like pros at answering. Is that part of this as well?
Ray: That's part of it as well. And those professional responders sometimes will also be really good at sussing out what it is that they might want you to hear. And that's always been the case. And really good research firms have ways to mitigate these. But they are getting more more so. And as the weight of decisions increases and as the
Pete Bishop: Yeah. Yeah.
Ray: you know, the scrutiny on ROI and as the scrutiny on data-driven decision-making increases, you know, the quality of the data that we have to make our decisions and the connectedness of the data that we have to make our decisions is falling a little bit short. And so this is where synthetic data comes in. In its purest form, synthetic data is basically modeled data. where real-world data doesn't exist. ⁓ a simple example might be, Pete, you you are, I'll say you're a 40-year-old male who lives in who is married, ⁓ and ⁓ let's you have two kids, I don't know how many kids you have, but let's say you have two kids, and we don't know what kind of car you drive, but we can model based on information that we know that you might own a sedan or a minivan or a SUV or something like that. in purest sense, what synthetic data is, saying, you know what? Pete's household is likely have an SUV. We don't know for a fact that you do. And ⁓ actually the other reason why primary data is ⁓ starting to is people's concerns over privacy. because Pete Bishop may not want us to know that you have an SUV and maybe a Ferrari somewhere in the garage, you know, for privacy reasons. And while we may not know that you have a Ferrari, you know, we can imagine that you might, you're more likely to have an SUV than, than,
Pete Bishop: Right. Right. Right.
Ray: Paul Bishop, who is of a similar age, lives in a similar area, but is a construction worker. And so we would imagine that Paul Bishop has a pickup truck and Pete Bishop has an SUV. We don't know that for sure, but that's what synthetic data is. It is a probabilistic model of only car ownership, but behavior, store ownership on real world data.
Pete Bishop: off, I'm really happy that you picked me out ⁓ as forties. that's flattering. And you're pretty accurate with the ⁓ I could see that really working The amount of data must have to feed into this must be pretty significant for that to feel like a real audience. Is that fair?
Ray: Yes, is fair. it one of the reasons why synthetic data is, I will getting to be, there's a lot noise around it now because just now the computing power and the computing algorithms to make it possible are just coming on stream and online now. Theoretically, it's all, it's been possible for many years. And it used to be called what was called data stitching, where you would take two surveys or 10 surveys, or in the case of a REMA 26, different survey data pieces, and you'd stitch the data together in a probabilistic way, like a really large Sudoku. So, you know, more or less, know, for who does Sudoku, you know that the rows and the columns need to add to 10 in each row and in each column, you can only have one, one, one, one, three, one, four, one, five, whatever. And so,
Pete Bishop: Mm-hmm.
Ray: You know, we know that in a postal code, there's a Pete Bishop and there's a Paul Bishop, both of the same age. One's a construction worker. One's an advertising executive innovator. And we know that from StatsCan that there is one SUV and one pickup truck in that postal code. And so we're going to guess that Paul Bishop owns the pickup truck and Peter Bishop owns the SUV. Now that's a very simple example. But you multiply that by 26 data sources, 60,000 variables. No human can do that. You need to probabilistically do that using an AI algorithm.
Pete Bishop: it's fascinating. So, you know, we talked a lot about the kind of the disadvantages of, you the the way of how we're doing it and then maybe some of the fatigue that's happening. For this, there be a lot of advantages beyond just ⁓ pricing and speed, There must be like I f f me, ⁓ the fatigue you could ask this group at nauseum, I expect, as many questions as you want, as often as you want, right?
Ray: Yeah, absolutely, 100%. And you know, one of the interesting things, we just did a test with a research partner. One of the interesting findings out of this test is that in the real world, you might say, do you think that Canada's inflation rate is going to go up or down, up or down, or I don't know? And a lot of people will answer, I don't know.
Pete Bishop: Right.
Ray: in a synthetic research world, you're not allowed to answer, don't know. model will always assign a yes or a no. And so you actually get much more, you get more completed ⁓ so advantages other than cost and speed, which you have outlined already is I'll call it data So in our hypothetical example, we have in Arima's case, we have 60,000 behavioral, psychographic, demographic ⁓ Pete Bishop or the synthetic Bishop and the Paul Bishop. You might recall that one of the disadvantages to traditional data is that I might have Pete Bishop's attitudes towards the Alberta government. And over here, I have Paul Bishop's attitudes towards pet ownership. But I don't have Pete Bishop's attitudes towards pet ownership or Paul Bishop's attitudes towards the Alberta government. But in a synthetic data situation, you actually have all variables for all synthetic individuals, which allows for really cross category explorations innovation use cases. ⁓
Pete Bishop: Right.
Ray: for new product concept in use cases. And so, you we had a use case where a candle manufacturer came to us and said, we really need to look for males of this gender who use candles. Well, there's no survey talks about candle usage amongst young males, for example, but our data has that. Or interesting example, a real life example is that one of ⁓ clients is a outdoor manufacturer who wants to come out with a line of pet clothing. So, you know, your pampered pet can walk down the street in the middle of winter with a ⁓ down jacket by this clothing manufacturer, but they had no data whatsoever their customer base or even their prospect database on pet ownership. what kinds of pets were owned, where they shopped for pet supplies, how much they spent on pet supplies. we had all of that. We have all of that information because we have 360 degrees worth of data on every synthetic individual in our population.
Pete Bishop: So just to dig into that, how is how is that? Like so you mentioned the candle, right? Like at some point, is just because you can query this with anything you want, or it's it's been queried so much that it actually has that data already in it? Like how does it how would it know about candles of that data? I'm just trying to wrap my head around it.
Ray: It's a little bit of both. So we buy a lot of usage and attitude data to the synthetic population. So whether it comes from, ⁓ you know, Numeris or whether it comes from Vivid Data or whether it comes from a media company, whether it comes from the Canadian government, the University of Calgary, wherever. And there's two parts to what we do. We buy a lot of source data.
Pete Bishop: Mm.
Ray: And that source data covers a lot of ground. using proprietary algorithm, we take, let's say it's a survey of 50,000 individuals, we then probabilistically create our database of 30 million for Canada for those variables. But we don't cover it all. So we've actually also developed an algorithm which allows us to query and ask the synthetic population. about variables that we don't have. So for example, at the height, when the tariffs were starting to, the US was starting to impose tariffs, I guess it was just about a year ago, or maybe a little bit more than a year ago, many of our clients said, we need to understand what our audience's reaction to those tariffs might be based on these parameters. And so using our ⁓ we are able to get a sense of that the population within ⁓ a very short amount of time. Now it's modeled for sure. It's not actually going out and asking people question, but it's saying somebody votes conservative, believes ⁓ in or the family value of, ⁓ family's the thing, then they're more likely to answer yes to tariffs than no to tariffs. I'm making that string up. But you can see that there's a string of logic. It's actually called a multi-step reasoning approach. that says if this, then this, if this, then that, if this, then that, if this, then yes to the question. Much more complicated than that, but that's kind of the idea. And so you can fill in the data gaps as well. And it self-perpetuates because the more you have, the more data points you can triangulate to fill in the missing data, so to speak.
Pete Bishop: Right. Ugh, that's brilliant. So we talked we talked a little bit about like obviously the upside. Is there a downside to this or does this replace like research or it is ⁓ is a balance or how do you see this unfolding as it ⁓ kind of moves ahead?
Ray: I think there's a couple time horizons to your question, and I think your question's a really, really good one, and it's a little bit untestable. ⁓ recently did a test with a partner where what they did was ⁓ actually a live survey, and then they compared their live survey results against the same questions asked of our synthetic population. By and large, there's...
Pete Bishop: Right. Right.
Ray: There were four criteria. was ⁓ a very well-grounded test. You know, the upshot of it is like, well, you know, the company that the test, they were like, wow, we were really impressed with how good the answers were. Now, there were limitations. You know, if you very future-oriented questions, like, do you think you're going to get married in the next three years? Those answers ⁓ versus the real... live answers were not as, you know, not really comparable because the data we have by its nature is hindsight looking less foresight looking. You know, there were language issues, there were, there were various issues, but, and, but we are at what I would call, you know, version two of synthetic data right now. And, you let's say in five years, ⁓ we be at version five.
Pete Bishop: Right.
Ray: And in seven years, we might be at version 17. And so in that, what did I say, seven year span, the use cases are going to get so much better. They're going to, it's going to get so much more accurate. Today there are outages, there are known outages and there's clunkiness, but as you already identified, it is a self-perpetuating model. more it... It's a machine learning model. The more it learns, the better it gets. The more data it has, the better it will learn going forward. Will it replace primary research? I don't believe so. Because primary research is always the seed. You know, the spine of the Arema synthetic population is the census data, which is every four years. We can't build it without census data. We can't build it without some actual starter
Pete Bishop: Yeah, right. And Yeah. Right.
Ray: don't do sourdough, you need a sourdough starter to get it going. Otherwise, you're just it up as you go along. There will always a place for primary research, though will always be a place for certain types of primary research. is an augmentation and is a complement rather than a replacement.
Pete Bishop: Ha ha ha. how do you kind of keep bias out of this? Like I know, I think that's true of any research, right? And anyone you're talking to, but in especially if you're if you're using research that's already been done, is there is it the problem that you would have anywhere else or is it is it different because of this new model?
Ray: know, bias, like non-response bias or biases, which all of humans have, those are present in any type of research. And there are to against that. They do get, they can get magnified when you are taking, let's say a sample of 10,000 and you're making it 30 million, because some would argue that what you're doing is you're trying to...
Pete Bishop: Right. Right.
Ray: that you're actually magnifying the bias. And there is some to that. There's two mitigations though. One mitigation is that we use multiple data sources. And so while there are systemic biases in things like non-response bias, you actually do lessen that by using multiple data sources. Secondly,
Pete Bishop: Right.
Ray: The models that we've created, actually we create some randomness into the to try to ⁓ that. Now, there proof that it does or does not theoretically, yes, but is there ⁓ proof that it does? No. so in the simple example that we have now, we know that Pete Bishop is an advertising executive and so probably owns an SUV and Paul Bishop is a construction worker, so the model would definitely say that he would probably own the pickup truck. But when we introduce randomness in there, it will say, but no, you know what, we're going to give Pete the pickup truck and Paul Bishop the SUV because we want to introduce that randomness into the model to reduce any kind of systemic bias or other systemic issues that might be present in the source data.
Pete Bishop: yeah, that makes sense. Is ⁓ is just for big business? Like I I'm just trying to get a sense of scale. Like, can any kind of use ⁓ synthetic at level or does is this kind of big problems only and big money only kind of thing? ⁓
Ray: I guess it depends on who you ask. And there are many versions of synthetic data. approach has been to, we're a bunch of data geeks who've spent our time in this world for many years. And sometimes we say to each other, we're building the tools that we wish we always had. And so the data and the tools we've built on top of this to access that data, because it's really important to access the data.
Pete Bishop: Yeah.
Ray: are available to anybody and are built for non-data scientists. And we have a platform that sits on top of our data, which is really easy to use and yet really powerful. The analogy that I will draw is that the platform that we have is similar to our do-it-yourself tax platforms these days. You go it or Wealthsimple or whatever to do your taxes, you input your data,
Pete Bishop: ⁓ yeah.
Ray: You know, there's a bot that pops up and says, you know, if you increase your RSP contribution, you could save this much. Or if you gave that much more to charity, you could do this, that, or the other. And there's a very dynamic iterative interface. And we don't need to know tax law to file our taxes anymore. We allow the AI and the platform to guide us. And so we have built a platform on top of our synthetic data that guides and helps the users through the process to answer whatever question they have. whatever questions or whatever marketing problems that they're trying to solve without an analytics background. And so they can model cross-media plans without really needing to know advanced statistics. We have a GIS module which allows for mapping of traffic patterns without really needing to understand the principle of centroids. and of how the postal codes overlap or don't overlap. It just does it for you. It just helps you do it, let's put it that way. So it is accessible to everybody. We've tried to make it accessible to everybody. ⁓ in fact, most of our clients, and I don't wanna shill too much, but most of our clients are small to medium size agencies and small to medium size brands. Literally, I just spoke to a...
Pete Bishop: Yeah.
Ray: a brand new client just before coming onto this call. And she wants to activate 15 to 20 of her, of her, her team, everything from a media buyer to a junior analyst, to a, know, to their head of stats, everybody to use it so that it makes the data and data driven decision-making accessible to all. Whereas you know, it was always a great theory, but in practice it was not. Not so.
Pete Bishop: Yeah, I it it you know, it's funny 'cause when you hear it all explained, it's it feels like such a noble brainer, but I also get why people are so reluctant 'cause it's it's a big mind shift to s you know, for the people who are thinking, you know, the robots are taking over everything and This is a definitely one of those moments where you're like, ⁓ my goodness, we're all getting replaced by robots here. But can really see the advantage of doing this type of thing and like you said, you know, removing some of that fatigue. I remember going doing surveys where like, wow, we can't ask more than fifteen questions because people just get tired. So, you too bad. it's like, Well, you could ask a thousand questions ⁓
Ray: And you know, in post COVID, I'm in attention spans, attention spans, like, I don't know about you, can't sit through a 10 minute YouTube video anymore. Give me 30 second version.
Pete Bishop: Right. No, it just seems fascinating. So do you feel like that's the the the barrier in thinking, just getting your brain rewired for this moment? what would what would be some of the like reasons why people wouldn't do this, I guess, or that you hear of like what are the common reasons they're like, ⁓ about this, what about that?
Ray: Yeah. And there's a lot of what about this and what about that. And it's valid. The main one is like, how valid is this data? it is a mind shift for sure. you know, the first EVs were introduced in 2002. And I remember hearing the first electric vehicles, battery vehicles, and I was like, ⁓ that's awesome. Well, but I have a car already that I just bought three years ago and What about charging infrastructure and what about reliability and what about this and what about that and change takes time, period. Change takes time and this is a big mind shift as you have already identified. And so will take time. know, the early adopters will adopt and they'll create use cases that will, you know, that will that will convince the mass and then the mass will, you know, then convince the late adopters and, you know, so the cycle goes with anything new. My advice to listeners, my advice to everybody though is, is got to jump in and just start to use it. It doesn't have to be in a big way. You don't have to base multi-gazillion dollar decisions on it.
Pete Bishop: Mm-hmm.
Ray: But you need to be familiar. There are so many different sources and types of synthetic data. You need to understand them. You need to understand the use cases that are right for you and your organization or you as an individual because it is moving quickly. And you you use the analogy before when we were talking, Pete, you know, if you drop an iPhone 17 on somebody to begin with, you would be lost because you wouldn't know how to use the functionality. But We started with iPhone twos and then they went to iPhone threes and then we went to iPhone fours and everyone got incrementally better until we're now at, I don't even know what we're at now, 17 or 18. ⁓ And the advanced, you know, early adopters are using those, well, or Discord or using Discord, you know, in ways that us Luddites can't even imagine. at the leading ⁓
Pete Bishop: Yeah, I think so.
Ray: But if you don't get started, then you are going to be behind the curve because the early adopter companies are embracing it in a very big way and are gaining meaningful competitive advantage from it, from synthetic data.
Pete Bishop: Yeah, it's it is fascinating. I don't know if this will make the podcast, but I do need to talk to you, I think, a little bit about travel research, because I'm in the middle of it and I would love your perspective on it. it's just kind of like the new travel behavior using AI ⁓ and how they're booking differently.
Ray: Yeah. Well, and we actually have two cruise companies, destination X, like destination Toronto, destination ⁓ types of, I guess those are tourism boards as clients. You know, one of the reasons why I think I might've, so Arima has a population of Canada, a synthetic population of the U.S., a synthetic population of Germany, and a population of the UK. ⁓ And the reason we built a synthetic population of Germany and the UK to start with, because they may seem like odd choices, was that one of our clients said, we need to target tourists from those countries to come to Canada, and we need the synthetic population to do that. One of the other things that we're doing is building a synthetic population of fans. So sports fans. sports fans are not just sports fans. They consume on media. They consume gaming. They travel. They consume in stadium.
Pete Bishop: ⁓ right.
Ray: you know, they buy merchandise, they are definitely multifaceted and understanding that group is, some thing that one of our very large clients has identified. And so they are funding our development of that database, but travel is definitely one of those pieces as well. When, when we think about travel, gosh, my eyes were completely wide opened wide when I saw the travel. ⁓ behavior around this World Cup. had appreciated before, never appreciated before the extent to which people will travel to follow their team during the World Cup. And, know, and so that's
Pete Bishop: ⁓ I believe it. Yeah. Like fan fanatics.
Ray: But maybe not even when I saw the pictures of the Tartan army in Scotland and the number of people in Scotland or witnessed the number of people of the last game was the Portuguese in Toronto and they closed down the 427 because they were jamming up the 427 waiting for Ronaldo and the team bus to drive by. It's like, what the heck? This is crazy, man. ⁓
Pete Bishop: was I was in Scotland when they won their first game, which they were not expecting to win any of them, and th they gave everybody the day off the next day 'cause they knew that no one was gonna be able to work.
Ray: Well, or just to see the pictures at the Red Sox games of them taking over the stadium. I mean, was so much fun. It's not over yet, but it's so much. It's over for Toronto anyway, but it's so much fun. It's over for Canada too, I guess.
Pete Bishop: Yeah, I imagine that There must be so much data pouring out of those things too, just like realizing how like I thought it was really neat to see how Canada got on board this year. not like they haven't in the past, but this year it feels especially big. it's just like what's driving that and why are we so excited about it this year? Obviously hosting, but you know, there's lots of other things I'm sure at play. It's ⁓ yeah, it's it's it must be a fascinating job just in general, watching trends and seeing what's what's ticking and what's not. There's all sorts of insights I imagine you get.
Ray: know, one of the things that we say to our detriment is that use cases are limited only by imagination. And to our detriment because people are, you know, in this type of world, people are not good at imagining things. You know, we couldn't imagine where the internet was going, you know, when the internet, when we first heard about the internet, but, and this is kind of the same, but imagine, and we do, have mobility data. on top of our synthetic people. And so now you can understand travel patterns, time of day where people, you know, spend their evenings. We have a golf course client who use data to model before and after the round. Where are people going before the round? Where are people going after the round? ⁓ So we can understand. Actually, we an auto dealer who does the same, models the
Pete Bishop: Mm.
Ray: before and after what dealerships are, our audience visiting before they come to our dealership and what audiences, where do they visit afterwards and where are they going? And that way we can better shape our marketing messages and our out of home placements and ⁓ offer, our better because we know they're visiting Toyota before they come to us. And so we can target Toyota. mean, that kind of thing is possible when you have synthetic population data.
Pete Bishop: head's spinning right now, and it's not just because of my hangover. ⁓ the you it it it ⁓ I do it's I I love the thought around the only limit. I've said this a lot. Like the only limit right now is your imagination. I it really feels like that with the tools that we have at our disposal lately. Is I I feel like I feel like I am
Ray: It blows your mind. You
Pete Bishop: constantly not thinking big enough because you can you can do these things and this is such a great example. And I think it's been around for a while. Like I we did talk a little bit around you know this notion of creating a persona and then asking it questions. I've seen people create like their board of directors, like a virtual group of them and feed them all their personality you know ticks and everything like that and then present to them and try and like work out some of their problems ahead of time before presenting to their board. Like things like that. feels like a like that steroids is what's been built here, right?
Ray: You that actually brings up a good analogy, I think, Pete, you you asked about will primary research ever go away. And so let's, let's call synthetic data a flight simulator. And, know, modern flight simulators, you can spend hours and hours and hours learning how to and you can pretend to crash the plane and you can cut out engines. You can do all kinds of scenarios in a very safe environment. but nobody's going to give you a pilot's license having you be in a real plane, flying in real world and having a certain number of hours in a real plane. So synthetic data is kind of the same. You can play all you want, but at the end of the day, I think most people will still want to test, know, test the new packaging in a test market or, you know, try out a new, you know, test communications A versus B or, you know, you know, test in the real world before you give the pilot's license, right? Air Canada is never going to give somebody a pilot's license who's never flown anything but a flight simulator, but the flight simulator is so valuable for training. It cuts the training cost, it cuts the training time, it increases the value of training, but all in a very safe environment. so data and we call population simulators are the same. They're just... place that you can. test and accelerate your your bad ideas out ⁓ all of those kinds of things.
Pete Bishop: Mm-hmm. ⁓ that's a great analogy. Yeah, I love that. It does feel a safe spot to ask stupid questions or or find probably ask questions than you would have, right? Because it is so accessible and and probably a little more efficient to try things out and experiment and maybe get to find those and insights that you would never have otherwise found because of the time and effort it would take to to do that, right?
Ray: it's synthetic populations and they're not allowed to say, don't know, you could find things that you may not have ever occurred to you before, right? ⁓
Pete Bishop: Right. yeah. No exactly. let's on question 'cause I'm I am curious about this. Just if you can fast forward five years from now, where's wh where do think Arima's going? Where where is all of this going, do you figure? Like what is it what does it look like?
Ray: I think that it will be much more wide, things like what you said, like recreating the board of directors ⁓ ARIMA was asked to recreate the house of representatives in the U S I think those are extreme cases. there are data integrity reasons why those small audiences will not get created, but the use of synthetic personas and particularly the use of synthetic populations will become, I think a lot more commonplace than they are now. Earlier this year, May, Gartner came out with a report that basically said that by 2030, 80 % of companies globally will use synthetic data at scale for decision-making, up from 10 % today. it will accelerate very quickly. It will become more commonplace. It will never, never replace primary research because you'll never give somebody a pilot's license without them actually flying a plane. But it will become more commonplace and the of use cases and the quality of the data will just only get better with time.
Pete Bishop: Mm-hmm.
Ray: Now my advice to everybody, as I said before, always just get started now because 2030 is only four years from now. And that's not very long.
Pete Bishop: Yeah. and like you said, there's no pilot these days that doesn't have to spend a certain amount of time in that simulator, right? So it's become commonplace. but all these things just take time and once the once that little You know, the fear of this or the anxiety around it kinda you get over that little hump, then it grows so quickly, right? And th there's like we see this all throughout our careers for the most part. You saw when the internet came in. I remember to pick on Scotland again, but there at one point ⁓ the government decided to put a computer with a a modem into every household in Scotland. And a lot of them just looked at it like the devil's box and it just sat in a corner and no one used it. ⁓ Right? But once you know you get over the hump of like, ⁓ okay, I get this now and it's starting to feel normalized, then you just see it kind of connect and swing up and it feels like this is in that moment, right? Like it's it feels early days, it's not common knowledge and a lot of people are probably loving it and running with it, but it hasn't quite ticked over in a lot of people's minds that this is where it's going and ⁓ you know, we need to be part of it.
Ray: Yeah. Going back to, once again, going back to the flight simulator analogy, you know, in the early days of flight simulators, the scenery was really blocky and the trees didn't look photorealistic and the cities were not properly rendered. But, you know, the latest iteration Microsoft Flight Simulator, it's like, like that's my city. I can fly over my house. And it you know, you can set as many guardrails or non-guardrails as you wish to simulate flying and they just have gotten so much better and synthetic data for marketing purposes is going to do the same. It's just going to get so much better over the next five years and that photorealism we'll call it will just improve vastly.
Pete Bishop: My brother, he had the flight simulator on the Commodore sixty four and he would literally spend 'cause I think it was almost real time, it'd be like, ⁓ if you want to fly over to Europe it's gonna take seven hours he would start it and it would like literally take that long and he would just be staring at these two colors with a line in the middle. I was like, How are you doing this? This is crazy ⁓
Ray: Yeah, exactly. now you can look at the ⁓ know, all kinds of crazy stuff.
Pete Bishop: Yeah. And all the dashboard and everything, you probably dial in whatever model plane you want and all that stuff. Yeah, I can just you can probably nerd out heavily on this. But it goes to show just the how that stuff changes grows and I can just imagine where you're at in the next little while. But it's ⁓ yeah, fascinating. It's like honestly still wrapping my head around it, but I just feel like it's a secret treasure trove that everyone needs to know about. So this has been fascinating.
Ray: Yeah. You can't. It's been, it's exciting. It's an exciting space to be in.
Pete Bishop: Yeah. Well, thanks much, Ray, for spending some of your time. I know you're busy today, but I I appreciate the chat. and hopefully we can connect soon.
Ray: a pleasure Pete, I'm glad you reached out and I really enjoyed our chat, thank you. You know, I think there is one thing that I want to say is that I think we're still at the stage of buyer beware. And so there are synthetic data sources with do which with with what I will call dubious provenance. And so there are some synthetic data sources where they will say, well, you know, you just tell us who your target audience is. Females, 35 to 50 who have body conscious images.
Pete Bishop: Mm.
Ray: a negative body conscious image. And they will then, you know, they will have an interface that allows you to ask questions of those, but you don't know where that data is being drawn from to give you those answers. And so there is very much a buyer beware mentality that I think users synthetic data have to have today still because, you know, ⁓ knows the term ⁓ hallucination and so there is synthetic data hallucination. If you don't know the source data, if you don't know how the persona the population is made, you don't know where your answers are coming from, it is not a good thing to be just blindly trusting them when you have multi-million dollar decisions to make based on the ⁓ outcome.
Pete Bishop: Right. You know, I feel like that's good advice for right now, just understanding your data. Like there's a lot of people especially when you're dealing with even anthropic or open AI, just understanding where were things trained, where is this stuff coming through, right? I had a friend that was I Hopefully not repeating myself, but I had a friend who was every time he was getting a response out of chat, it was it was saying something like, Thanks for tuning in on the end of all this stuff. And he's What is this coming from? Or and he's like, Yeah, it's because it was trained on YouTube data. So most of the time it's talking about episodes, and that was kind of how it would sign stuff off. So when you're like that's where from, right? And you just don't realize, ⁓ that's right. Yeah, it's just it is a product of its data. And if the data is garbage in, garbage out half the time, right? So it's really important to know Know where it's coming from because think these days that's just part of being a literate in AI. So and this is no different by the sounds of it.
Ray: It's no different. if the person providing your synthetic data or the synthetic data source is a little bit vague and cagey about what the data source is, then I would say run away, right? You want people who are open and you need to understand because there's, you in a, you know, particularly for, you know, agencies or brands or whoever, like I said, there's, there's big money decisions at stake here ⁓ and you to
Pete Bishop: Yeah. Yeah, and your reputation's on the line, right? Still is, right? No matter what.
Ray: Mm-hmm.
