Fixing the neighborhood: what's actually broken in data quality with Karine Pepin

Listen

Watch on YouTube · Apple Podcasts · Spotify

Summary

Karine Pepin, co-founder of The Research Heads, on the real state of data quality in market research — why the biggest threats remain misrepresentation and click farms rather than AI agents; how LinkedIn alarm around data quality is amplified by companies monetizing it; the ground-truth subset technique for internal validation; why aggregate-level detection but respondent-level fixing creates a painful mismatch; the house-in-a-neighborhood analogy for why securing individual surveys cannot fix a structurally broken incentive ecosystem; CASE advocacy for transparency and identity validation; incidence-rate divergence as the most telling quality signal; why the industry should agree some low-incidence audiences cannot be done online; and the competitive dynamic between a $200 LinkedIn-recruit CPI and a $45 online CPI. Plus Jonathan Goodbread unpacks AYTM's newly released data quality benchmark report, the 4P framework in practice, the 5:44 median LOI, 47% pre-survey fraud removal, Krosnick's cognitive-load research applied to survey design, and the call for industry-wide transparency.

Guests

Karine Pepin — Co-founder, The Research Heads · Supplier-side
Jonathan Goodbread — Head of Data Quality Strategy, aytm · Supplier-side

From this episode — top claims

• False marketing claims about panel size — 'nobody has 50 million people on their panel' — are a disservice to the industry and harm smaller providers who may have fewer but more active, responsive, and attentive panelists; what matters is engagement quality, not lifetime sign-up volume. — Ep. 77, Karine Pepin
• Data quality is a house-in-a-neighborhood problem: researchers have been building more secure individual surveys (better locks, security systems, guards), but the wider ecosystem remains fundamentally broken because both suppliers and respondents profit from volume rather than quality — and unless incentive models change, the neighborhood cannot improve. — Ep. 77, Karine Pepin
• Data quality in market research is not materially worse than it was five years ago — the heightened alarm on LinkedIn and in industry discourse is partly an artifact of more companies now trying to monetize data quality solutions, which amplifies the sense of crisis beyond what the underlying problem warrants. — Ep. 77, Karine Pepin
• Data quality must be an industry-wide effort requiring transparency from every participant — panel providers, consultancies, and tech firms alike — because research that does not reflect real-world scenarios hurts the industry's credibility, individual reputations, and organizational reputations. — Ep. 77, Jonathan Goodbread
• AYTM's data quality benchmark report reports the same statistics the GDQ publishes every quarter, computed internally from AYTM's own data, so that buyers can directly compare AYTM's quality metrics to industry-wide benchmarks in full transparency. — Ep. 77, Jonathan Goodbread
• AYTM's data quality approach is organized as the 4P framework — prevent (good survey instrument design that respects cognitive load), protect (removing fraud before it reaches the survey), purify (cleaning out what gets through), and prove (a per-link data quality report showing who was removed, at what stage, and why). — Ep. 77, Jonathan Goodbread

Full transcript

0:00 When you think about the ecosystem and the neighborhood, for that to change, we need to fundamentally change the incentive models. Because the way it works today is that both suppliers and respondents are profiting from volume, not from quality. So, unless we can like change that completely, I don't think the neighborhood can fundamentally change.

0:00 Hello, fellow insight seekers. I'm your host, Molly, and welcome to the Curiosity Current. We're so glad to have you here.

0:00 And I'm your host, Stephanie. We're here to dive into the fast-moving waters of market research, where curiosity isn't just encouraged, it's essential. Each episode will explore what's shaping the world of consumer behavior, from fresh trends and new tech to the stories behind the data. From bold innovations to the human quirks that move markets, we'll explore how curiosity fuels smarter research and sharper insights.

0:00 So, whether you're deep into the data or just here for the fun of discovery,

0:00 grab your life vest and join us as we ride the Curiosity Current.

0:00 Today on the Curiosity Current, we're joined by Carine Peppin, co-founder of The Research Heads, a qual-quan agency focused on helping brands transform insights into bite-sized research assets, and a fierce advocate for data quality and the participant experience in market research.

0:00 Carine's spent more than two decades on the researcher side of the industry, where data quality is not an abstract problem. It is something she has to see in the data and ultimately defend to clients.

0:00 Today, we're exploring where data quality really stands in our industry, why catching bad respondents only solves part of the problem, and what researchers can actually do when the data and their tools tell them different things.

0:00 Carine, welcome to the show.

0:00 Thank you. Thank you both for having me.

0:00 Well, we're really happy to have you and excited about this conversation today. If you don't mind, we would really love to start with kind of the 30,000-ft view. If you read LinkedIn, and I know you do cuz you post a lot on LinkedIn, I I read a lot of what you post. But also sometimes if you even just read the news, data quality kind of feels like a five-alarm fire. Everyone is sounding the alarm about bots, about fraud, about attention being in a race to the bottom. You're someone who works with this hands-on every day and who's very plugged into data quality at the industry level. So, where do we actually stand as an industry? Is it as dire as the feeds make it out to be or do you see it differently?

0:00 Well, I think it's pretty bad, to be honest, but I also think that if you hear more about it on LinkedIn than you used to, it's because there are more companies trying to monetize data quality now, which makes it sound worse. But as far as I'm concerned, compared to the past 5 years, I don't really think it's gotten worse. I think our biggest threats are not AI agents. They continue to be people who misrepresent themselves and click farms. So, I think yes, we need to uh to work a lot on data quality, but I don't think it's actually even worse than 5 years ago.

0:00 And uh that's a great place to start because it sets up some of the tension that I want to dig into. At the 30,000-ft level, we can talk about the state of the industry. But for a working researcher, data quality isn't abstract. It's something you have to spot in the actual data and then defend to a client. So, can we get into the weeds a bit? Uh sometimes data looks perfectly reasonable at first, but starts falling apart once you dig into the cross-tabs. What tells you that you might be looking at a quality problem rather than a genuine finding?

0:00 It's a good question. So, I think there are a couple of things, but first is how much of the data is surprising to you? I think you need to have a good ratio of things that you're confirming and things that are new. If there's too much new, then I think that's the problem. But assuming that you're just seeing this in a few spots, the way that I handle this is that I try to find a ground truth. I think that's that's something that's super hard for researchers is that we don't know what there's nobody that's going to come at the end and give you a report card and said you were right, right? Like you just go with your instinct. So what I like to do is create myself a ground truth. So I identify in a data set a subset of the sample that I'm confident they're real people, they're attentive.

0:00 That's my ground truth. Whatever they say is the truth. And the way that I find these people in the data set, there there different ways. So sometimes when you blend sample, you could have a subset that was done face-to-face and a subset that was done online. It's like, well, the face-to-face people are verified, right? So what are they saying? You could you could be a LinkedIn recruit. It could be a panel that you have more confidence in or it could even be markers from a fraud detection software if that's all you have. So for example, there is a fraud detection software out there that follows respondents across the ecosystem and computes a persona. So some of these personas are really bad and some of these personas are really good.

0:00 So you can take the good persona data and see how they responded compared to the rest. So in an ideal world the two align and you're like, okay, well, that's a real finding because the good people are also saying that. If it doesn't align, then this is when you start digging more into the data. And I think what's super challenging is that you can identify these problems in an aggregate, you know, at an aggregate level because the data looks wrong or it's just looks like there's noise, but you can only fix that problem at a respondent level. So then that's why you have to start digging into the data and find like is it completely random or is there like a group of people, a panel, demographics, like something that is causing that to happen.

0:00 Kind of related to that, talking about digging in at the individual level, you use this great metaphor of data quality being a house in a neighborhood problem. Uh individual researchers who are at the individual survey level keep adding better locks, better alarms, but the wider ecosystem, the safety of the neighborhood, is not necessarily improving. What would it actually take to start fixing the neighborhood?

0:00 Yeah, so I like this analogy because researchers in the past few years have been building better or just more secure surveys, right? More secure houses. We've We had door locks, you know, trap in the surveys, but now we've added a security system. We've added like security guards, like, you know, but if you live in an unsafe neighborhood, you still don't feel safe even if your house has barbed wire wire all over it. So, I think when you think about the the ecosystem and the neighborhood, for that to change, we need to fundamentally change the incentive models. Because the way it works today is that both suppliers and respondents are profiting from volume, not from quality. So, unless we get like change that completely, I don't think the neighborhood can fundamentally change.

0:00 And you know, we can have better pockets of neighborhood, you know, some panels I know are trying hard, you know, to clean their neighborhood, but because of the way exchanges work and and how people are blending sample from many suppliers, sometimes it doesn't feel like you're in a safe neighborhood at all, even though there might be pockets that are safer than others.

0:00 No, that absolutely makes sense. And I think this might be a good opportunity to talk a little bit about the work that you do with Case for Data Quality. Would you Can you talk about that?

0:00 So, Case for Quality, um a few years ago, before I join I joined them, they were one of the first to actually do studies about data quality to really understand what the state in the industry was as a And they also did not only just look at quality across suppliers and across sources, but they also tested fraud detection software. So, do they actually work? So, it was it was a really big study. And one of the key finding of the study was that the average respondent attempts 21 surveys in a 24-hour period. And there are people that complete just more surveys than humanly possible. This is something we've known for 5 years now. Right? But we haven't really acted on that. So, what Case for Quality advocates for is first of all transparency.

0:00 Um transparency about where is the sample really coming from, you know, how many people are a block, how many people are clean on the back end, all this information that suppliers should be providing. So, that especially if you use the same supplier for programming, sampling, and and doing data processing, and all you get is a clean data set, you have no idea what happened in between. So, transparency. And the other part of this puzzle is, I guess, validating people's identity. Because at the end of the day, that's sort of like the only thing that is going to to help improve this is making sure that people are who they say they are. And the only way to do this is to verify who the respondents are.

0:00 So, you know, historically have double opt-in panel, and you still have them, but it's so easy to create yourself a fake email account and join, you know, 50 different use 50 different, you know, email address to join the same panel, that sort of thing. So, that's another big thing that um we've been advocating for.

0:00 Makes good sense.

0:00 Yeah, so I'm sure that you've been in these situations where you are faced with a a bunch of respondents and multiple fraud tools flagged respondents that you thought were legitimate. Um so when the technology and your own reading of the data disagree, how do you decide what to trust?

0:00 Yeah, it's uh it's very much a story of my life um is to look at all these fraud detection flags and try to make sense of it. And I think at the end of the day, um you know, we're all trying to do the right thing, right? There's going to be false positives and there's going to be false negative as long as we don't identify we don't validate people's identity, this this is going to happen, right? We don't know who they are, so we're just going to have to use all these signals and try to figure out if they're good or fraudulent. So, personally, and I know that there's like a a bias there, but if to go back to the neighborhood analogy, if I'm using a panel or sample source that I'm very confident in, you know, I'm I'm walking into a a good neighborhood here, I will be more likely to give the benefit of the doubt to the respondent.

0:00 Because I'm like, well, you know, I'm in this good neighborhood, it's a stranger on the streets, but I'm sure they're fine. If on the other end, I know that this source has a lot of fraud to begin with, I'm in this bad neighborhood, now I think all the strangers are out there to get me. And it's very difficult to give people benefit of the doubt when you're in that situation where it's like, well, if there's already 40% that's fraudulent, they probably are. But at At of the day, we're just all trying to do the right thing for the data, for the respondents, you know, who have spent time doing the survey, and you know, for the suppliers because we don't want to reconcile people for no reason. But, this is the kind of stuff that really keeps me up at night.

0:00 Yeah, absolutely. You know, me, too. And kind of in the same vein, you know, when we think about an individual data set where, you know, you talked earlier about the role of exchanges, we know that there are different panel sources in a lot of our surveys, right? So, when that's the case, I'm curious like what tells you that the source itself may be affecting the quality of the data? And And really where this question is coming from for me is I think about it like uh in the context of like sampling strategy. Like there are times where if I divide my data by or do a cross tab with source as as the banner variable, they may look wildly different, but that could be a reflection of the sampling strategy that was used because I know that I can go and get high-income niche uh respondents from a particular panel, and that's the group that I have left, right?

0:00 And so, that was intentional, but it's not always easy for an end user to have access to that information. So, how do you think about that?

0:00 Yeah, so So, for me because I I try quality metrics, um you know, I use my own not my own, but I I manage my own like fraud detection software and processes. I don't uh rely on the suppliers to do that. I can see how the data's coming in. I can see the fraud, you know, markers and flags. But, I think what, you know, part of it is understanding what is the sampling strategy, right? Because one of the thing that I think is actually more telling than just purely looking at, "Okay, percentage of virtual machine, percentage this percent and that." Is actually looking at your incidence rate. If there is a source that has a vastly different incidence rates and you're not aware they're targeting, then it can only be two things, right?

0:00 A leading invitation or people are lying. But either way, like you know this is a problem. So it it's kind of important to understand how the sampling is done, but I think it is about kind of combining all of the signals that that you're you're getting because you know you know, I'm sure that it will easily blend into your survey. You know, you might not realize that incidence is 50%, you know, and uh that's a problem, but the data looks fine. You know, you can't really spot that after the fact. So it's kind of monitoring throughout the process, you know, while you're sampling, you know, the screener data, all of that, not just look at the completes at the end.

0:00 Yeah. No, that's a great tip. And I mean, it sounds like a lot of what you personally do and probably advocate for is if we're going to be living and walking through these neighborhoods that are not so safe, there's a level of personal responsibility that just has to be taken right now. Like it's the onus is is sort of on us to be aware of the sampling strategy, to demand a view into sample sourcing and things like that.

0:00 And yeah, and it is something that Case for Quality has been advocating for, just like we we talked about a few minutes ago.

0:00 Makes sense. Well, Karim, we have a segment that we run on every episode called the current 101 and we ask all of our guests the same questions. Um in your experience, what is one trend or practice in market research or data quality that you would like to see stopped or you could do sampling, whatever works. Um what is one thing you would like to see more of?

0:00 I think stop would be uh less false marketing claims. Nobody has 50 million people on their panel. That's just not possible. And they are certainly not verified or validated, however you use that word. And I think that's really doing a disservice to the industry and to smaller sample providers who might not have, you know, that's 5 million, but the size really doesn't matter. It's how active are people on the panel, how responsive are they, how attentive are they? That's what matters, not really that if people say the 50 million is probably because they have 50 million people who have ever signed up for this back on the past 35 years. That's irrelevant today. And what I would like to see more of that's a dream because I mean, that seems like it will never happen.

0:00 But I would really like if we got together and agreed on what are the targets that you cannot do online. And I'm using online research because there's so much traffic that goes through the online ecosystem, obviously. But there are some target audience that are not online just sitting there waiting patiently for survey. But we never say no, right? Because the supplier says we're feasible, the agency says we're feasible, and the client thinks we're feasible because we will always get data. Like getting data is not the problem, right? Like there will always be fraudsters who will qualify for these low incidents. But we know, like we know better. And so we can't keep giving the impression that everything is possible.

0:00 And I think that's really not setting us up for success for quality because we know that, you know, the lower the incidence, the worse the quality is going to be. So I think if we eliminated a lot of that kind of um of work or audience or target or whatnot, it would actually solve some of the quality problems um that we have.

0:00 And so for you, it's I mean, cuz I was going to ask you how do you define that? Like what audiences are you talking about? But you really just said like low incidences you probably don't want to be doing online. Is that fair to say?

0:00 Well, that's exactly why we'll never agree because because we'll start discussing because but you know, I think me to be for example, like that's the realistic like sure you can find IT decision makers, but really when you understand the ecosystem, right? And how much people are getting paid and the routing system and all of that how many legit IT decision makers will just sit there and wait for a survey. So if you want to do this right, we got to do a LinkedIn, you know, recruit and all of that, but I think what becomes difficult is if you're the one proposing the LinkedIn recruit at 200 CPI and your competitor is proposing online at $45 CPI. And how do you explain this that they can do it, but you can't?

0:00 It creates a really challenging situation.

0:00 Yeah, and I think this is a not something that will be fixed by one agency or one supplier, but it's really kind of trying to get together and be like, okay, let's be honest about what is possible here. But like I said, like we're so far from anything from anything of that nature, but we're just talking here, right?

0:00 We are, yeah. Love it.

0:00 Okay, well, we've covered both the 30,000 foot view and the practical realities of what good data quality work actually looks like. So to bring it home in your view, Corrine, is the industry actually heading in the right direction? Where are we still missing the mark? And of everything we've talked about, what is the most important work still ahead of us?

0:00 Well, I think it's fixing the neighborhood, right? As we talked about, I think there's been a lot of work done to kind of improve security around the surveys themselves. You know, I counted 23 different fraud detection tools out there. We know the main ones, but there are other ones, too. So I counted 23 of them. And on my end, all I can do is curate the traffic. So this is what I will continue to do as as best as I can, dealing with my false positives and my false negatives, um and trying to use better sources uh as much as I can, that sort of thing. But, I realize that this is just a band-aid solution, right? This is not actually fixing the problem for everyone, but I think we're just at this stage, I guess, of the problem that you know, if you can fix your own house and make your own house secure, well, that's kind of half the battle.

0:00 So, I don't know how we tackle the bigger ecosystem issue of how just like structurally we're incentivizing volume and not quality. And unless until that changes, it's it's really not going to change.

0:00 So, until that time, it sounds like we all need to focus on home security as much as we can. Fair to say?

0:00 I think so.

0:00 I don't want to be negative Nancy here, but let's just be honest.

0:00 No, I I think that's really fair. And and good advice, quite frankly, for for anybody who is doing research and really cares about data quality. Well, Karin, thank you so much. This has been an illuminating conversation. I feel like, you know, we should check back in with you in a year and and see where we are.

0:00 Thank you for having me.

0:00 Thank you, Karin.

0:00 Karin, thank you so much for joining us. I think what's especially useful here is the reminder that data quality doesn't begin and end with catching bad respondents. It shows up in the sourcing and the tools we use, and ultimately in whether the data itself holds together when a researcher starts asking harder questions and really drilling in.

0:00 Yeah, thank you, Karin, for joining us today and for sharing the practitioner's perspective on a problem the entire industry is still working to solve.

0:00 Now, we want to shift from Karin's perspective as a researcher working directly with data to what we're seeing um at a more macro level. Jonathan Goodbread, who's been interviewing with me today, is AYTM's head of data quality. And John's now going to unpack our newly released data quality benchmark report.

0:00 Thanks, Stephanie. I'm really excited about this report. You know, as I I came into AYTM, there was a real need for us to make a statement in the market on data quality. And as we evaluated methods to do that, we thought the best we could we could do that in full transparency was by reporting out the same statistics that the GDQ reports out every single quarter, which is where this report came from. So, we take the GDQ values that they report, we we find our own internally from our data, and we report out so that people can compare us to the rest of the industry. The analysis is grounded in our 4P framework of data quality: prevent, protect, purify, and prove. And I'm really excited to share the details of it here with you today.

0:00 Awesome. Well, let's get into it then. Jonathan, looking across the benchmark report, what findings tell us the most about where data quality problems are entering the research process?

0:00 Right. Well, the the thing is, the data quality problems can really enter at any phase of the research process, and that's why we have a framework of the four Ps: prevent, protect, purify, and prove. And so, prevent tells us that we need a great survey instrument that people can really engage with. And so, it's a measure of how much cognitive load and the kind of survey design we're doing in in order to ensure we get great quality respondents were paying attention. And in that sense, we've got a 5-minute and 44-second median LOI here at AYTM, which I'm really happy to see because we know that respondents are willing on average to give us about 10 minutes of their time. So, a 5-minute survey on average, and that doesn't include IR checks and things like that, that tells us that we're well within that window, and that we are in good custodians of people's time.

0:00 And so, once you built that instrument, you got to protect yourself from the fraud, right? And so, there's a whole chunk of people and you don't even want them to see your survey. And we've got 46% removal for that from our third-party providers, but if you add in PaidViewpoint, which is our panel, uh you get 47% and I think that's the more fair number. And that was a really gratifying number to see. Because if you look at the research right now, a lot of the research is saying for high-incidence work, you should be pulling out 30 to 40 at the gate. And we do, but when the incidence on the uh audience goes lower, that number has to go higher. And so, the fact that we're hitting 47% tells me that we're doing the work we need to do to pull the fraud before it hits the link at all.

0:00 Then have our purify step, which is, "Okay, we protected ourselves. We created this great survey instrument. Uh now, what are we going to do about whatever fraud is left over?" And we see uh a a 5.4% post-survey clean out rate. We we get the right people in and out of all of those human beings that answered our survey, about 5.4% of them we toss out for data quality challenges. And and so, we're finding the fraud that's left over, which is so great. And then there's proof. And proof is what's really really important in all of this, because you can make all the data quality claims that you want. Uh it doesn't matter if you can't back it up. And we put a data quality report on every single survey link, showing you who got removed, at what stage, and why.

0:00 And that way you know that the first three stages have given you great data quality.

0:00 I love that. It makes a ton of sense. And I I guess one thing that I would wonder if you could unpack a little bit is what are some of the signals that we see in the data that would let us know that these are respondents we should move once they're remove once they're in the survey.

0:00 Yeah, absolutely. And um it's it's going to depend on what's in the survey in the first place, right? But when we remove from a survey, what we're looking for is inattentive people who got through. We're looking for bot farms or groups of people working together. And then there's AI agents. And you know, th- those aren't a threat yet. When agents are fast and cheap, we'll have a different conversation. But, they are slow and expensive. So, agents taking your surveys is not yet a big thing, right? And so, we look at indicators that show us that these kinds of people are in our survey. And so, we look for people with extraordinarily long LOIs on a survey that does not require it. We look for people who give us poor open-ends or, in the worst case, open-ends that we can literally tell came from an LLL.

0:00 And so, we we look at some of those behaviors of the respondent. Um we look across respondents, and we we see if there's patterns in how they answered and the way in which they answered our verbatim fields, right? And so, we use the data that they provide us in order to get a sense of how they're interacting with our survey. And as you note bad behaviors in the data, those are the individuals that you are going to want to remove.

0:00 That makes sense. You know, the report also raises questions about study design itself. Um and we haven't talked very much about that. What do researchers need to rethink about survey length, uh survey content, and respondent experience if they want better quality data?

0:00 Oh, yeah. You know, since 1991, we've known from Krosnick that the more cognitive load that's in a survey, the more people are going to satisfice going through the survey. Uh and so, it is our job as researchers to actually put out a survey instrument that people are going to engage with and that people are going to feel compelled to give us their honest answers. And that means accounting for respondent cognitive load. And there's many ways to do that. One we use is the length of interview, right? So, we we again know that respondents are willing to give us about 10 minutes of their attention for a survey. But, we also need to look at the cognitive load of the individual components of the survey instrument.

0:00 So, if you have something that's very cognitively taxing like a max diff exercise or perhaps a conjoint, those belong at the beginning of the survey because that's when the people have the most cognitive energy to put towards your exercise. You put them at the end, you're looking for satisficing. So, you also need to make sure that you are not just stacking these hard exercises one after another after another because people's eyes will glaze over and they will stop participating. So, what you really need to do is design a survey of an appropriate length of interview, 10 minutes or less. You need to put your cognitively loaded exercises up front, the easier questions in back. So, as they get more tired, they have easier questions to answer.

0:00 And then you're going to come out with a survey instrument that people really want to pay attention to and work with.

0:00 It really highlights that like everybody has a role in making data quality better, right? From panel providers all the way through to the researchers that are, you know, boots on the ground working with the data to tell the story.

0:00 Yeah, absolutely. Data quality has to be a an event in which everyone participates and I do mean everyone from across the industry, be you a panel provider or a consultancy or a tech firm that services the broader research and insights community. We all have to be involved in this because at the core of what we do is our research and if our research does not talk about real-world scenarios, if it's just puff of insights in the wind instead of being grounded in real-world views, it's not useful to anyone and that hurts our industry, that hurts our personal reputations, it hurts our organizational reputations. If everyone doesn't participate and demands that our data quality gets better, uh we're going to be in some trouble.

0:00 So, um my viewpoint on this is that we need transparency from every player in the game. Sometimes that transparency isn't going to sound great, but we have to accept that. We are in a data quality challenge, let's all work together, become transparent, and figure out ways to get out of this.

0:00 Absolutely. Well, Jonathan, I think this brings us full circle. Corrine gave us that view from inside the data set and the benchmark report really gives us another way to see where these quality challenges are actually showing up.

0:00 Yeah, absolutely and reporting like this is so important. I love that we're doing this at AYTM. I would love to see other organizations start doing this, too. And quite frankly, I would love to see us all compare numbers. And it's not that I think that it's important to know who's best and who's worse according to which metric. It's that I think as we come together and we understand our panels together, that we're going to be able to handle this data quality challenge more effectively.

0:00 Yeah. Jonathan, thanks for helping us break down the report. And to everyone who's listening, thank you for being part of the Curiosity Current. We'll see you next time.

0:00 Thank you so much, Stephanie. It was a pleasure to be here.

0:00 The Curiosity Current is brought to you by AYTM. To find out how AYTM helps brands connect with consumers and bring insights to life, visit aytm.com. And to make sure you never miss an episode, subscribe to the Curiosity Current on Apple, Spotify, YouTube, or wherever you get your podcasts. Thanks for joining us, and we'll see you next time.

Produced by aytm

Curiosity Current is made by aytm, the consumer-insights company. Guests speak for themselves; the synthesis is ours. About aytm