Continue reading the transcript:

Gary Nowacki: Hmm. So, it models consumer behavior or consumer preferences. Um, how the heck does it do that? 

Jason Cohen: So, in the last 12 years, we’ve collected the largest sensory data set upon market products that’s ever been assembled. We have standing panels in New York and Shanghai, and we have a team that literally circles the world to a different country or different region almost every other week. And they recruit a set of panelists in that country. 

They have them taste a set of reference products that are available globally. We call that a demographic survey and then we have them taste 80 to a 100 on-market CPG products that are relevant to that market. And we call that a market survey.

And so, we use the demographic survey to parameterize the distributions and perception, the differences between different countries and different groups of demographics—how they perceive flavor. 

And then we use the market survey to reverse engineer the drivers of preferences in that market. And so, we now have done that in more than 35 countries, more than 45 regions around the world.

And what that allows us to do is build up highly accurate generalizable baseline models for those. And so, when we have a new product or a new development or a company wants to move a product from say, the United States into Europe and they say, well, what countries will this product do well in?

And we could say, it’ll do well in Germany. And it won’t do well in Spain, and it’ll do well in Italy. And it won’t do well in Norway. And we can do that all in real time, collecting data anywhere because we have that data set. 

Gary Nowacki: That sounds pretty cool. So, traditional new product development solutions, you know. How is yours better or different? You know, why won’t those work as well?

Jason Cohen: So, we see four deficiencies in traditional methods. The first is that it’s what we call snapshot data. When snapshot data means that it’s based on frequent to statistical hypothesis testing from experimental design. 

So, if you want to know if products are different, you have to run a difference test. If you want to know if they’re similar, you have to run a similarity test. If you want to know if they’re preferred, you have to run some type of hedonic test. If you want to know consumer descriptive analysis or consumer guidance, you have to run that or jars. And so, what that means is that that data is not comparable across tests or across time.

So, if you want to learn something new, you have to run a new test. Every answer requires a new test. And what that does is it means that you have to guess at the answer before you know what the right question is to ask, right? 

So, if you have a product that’s underperforming, or if you have a new product that you want to test, knowing what tests to run before you have any data means that you’re guessing at the answers.

And so, even when that works in the best-case scenario, right? All of that is a cost center. It’s not able to develop large scale explanatory models. But in the best-case scenario, when it works is you’re getting low fidelity consumer responses, which are very hard to turn into actionable insights that you could use to develop a better product or to make a better decision.

And so, in comparison to that, what we can do is everything that we do is predictive. We collect data in a unified format. We can reduce the total amount of data that we need to collect. We can reuse that data to make new predictions for new questions, and we can recycle that data for different countries or different categories into the future.

And so, what that means is that we have an always-on, ever-learning, always up-to-date platform that can continuously use that data. So, we think of that as the core benefit of using a predictive methodology, where everything that we’re predicting for is tied back to specific decisions and actions that a company can do to make better decisions.

Gary Nowacki: So, let me ask you, sort of a brass-tacks question here, Jason. Two-part question. 

Pretty famous McKinsey study done a number of years ago. They followed new consumer brands that were launched out into the marketplace over a period of four, five years. 75% of them were no longer on shelves after four or five years. So, very high failure rate. 

Secondly, you know, we hear from a lot of companies that their cycle time to develop a new product is nine to 12 months, which seems, you know, pretty long, pretty arduous. So how can this technology help with either the failure rate or reducing innovation cycles? 

Jason Cohen: I believe that it could help with both. You know, we see 85 plus percent failure rates across both companies big and small, and it doesn’t really matter the category. And those failure rates are from a multitude of things. 

One is that consumer preference is constantly evolving. Consumers acquire, try new things, acquire new preferences, move into exploratory preference states, and start trying new products. You know, and once a product hits three years on the market, it’s not safe.

15% of the products on the market are pulled every year. And so, when we think about what the AI offers, right? It offers not just the snapshot of what’s currently going on. It offers verifiable, accurate predictions about the future. If you need to develop a product today, that’s going to be preferred, not just today, but six months, a year, two years plus into the future, then you have to have an understanding of how preferences are continuing to evolve and what consumers are moving into those exploratory states and where those exploratory states are stabilizing.

The second thing is that traditional methods are slow. If you want to run a multi-country, CLT consumer pre-launch series of tests, that’s six weeks to six months. Depending on how many countries and depending if you can even run that test. We, you know, we picked up a ton of business during COVID because companies couldn’t run large-scale consumer surveys in the middle of a global pandemic and their options were to shut down their innovation cycle or to launch without consumer feedback or to use a predictive methodology—like us—to take existing data and to predict for new products without needing to go back to a consumer panel. 

So, the ability to replicate and exceed what a consumer panel can do with near-real-time predictions for multiple markets at the same time off of a single test is greatly accelerating.

We can offer feedback in four to six days versus six weeks to six months. And we can prove that those predictions are more accurate than traditional CLT methods. And so, we do believe that that results in much more successful developments. 

Gary Nowacki: Hmm. And on a related topic, you’ve been in the CPG industry for a while, Jason, do you think there’s more pressure on companies to innovate today or less than there was in the past? 

Jason Cohen: I think the pressure is different and I don’t think companies have an equal level of understanding of how and why it’s different. It used to be, the companies could sink quite a bit of money in tens of millions or hundreds of millions of dollars into a single development.

They could say, I’m going to develop one new crown-jewel, leading product. Right? It’s going to be our marquee product. And they could launch that product in multiple countries. They could launch that product maybe globally, right? But that’s no longer competitive. Today, in order to be competitive, you have to be targeted.

And some companies are finally starting to think in terms of portfolios of products versus singular products. So, launching five flavors in one brand where different flavors appeal to different segments of the population is the future of product development. It’s having products that speak to specific consumers, that specific consumers, not just like, but say, you know, this is a product that matches my preferences. This is a product that matches the types of products that I want to consume and the mode of consumption that I want to consume it in. 

And so, different companies are at different states of understanding of that. And so, you know, the pressure is different.

I don’t—I’m skeptical that there will be another billion-dollar product anytime soon. A single product that does more than a billion revenue in a year. But there are many portfolios that will do that. And there are many portfolios that’ll do that across countries with different offerings in different countries under the same brand.

Gary Nowacki: Hmm and speaking of different countries. So, you know, your solution rests on this data set that you’ve been collecting for a long time. You know, what happens when there’s a hole in your data set? What happens if it’s a category or a demographic that you just haven’t been able to test?

Jason Cohen: Yeah, well, two things happen. One, of course, you know, we won’t make the prediction without having a baseline data set that we’ve run our tests on, that we’ve made sure it has predictive power and verifiable accuracy. 

But the second thing is that, as a company, our goal is to totally abstract away any concept of data collection from brand owners and from CPGs. We never want one of our customers to have to think, well, do they have enough data? Is it a large enough sample size? When was the data collected? Right? 

And so, our goal is to do that—all of that—in the background for them. Right? So that they never have to think about that again. So, we have a team, we have a forward-deployed, tactical global panel team that has a rapid grab-and-go methodology that can drop into nearly any country in the world.

Grab relevant on-market products, stand up a panel, and, in under a week, have that demographic profiled and have that market surveyed, right? That’s somewhere between six and 10,000 observations that we can generate in a five- or six-day period in a country.

And so, when we get a request, a real request, obviously a legitimate request, someone says, “I want to do X product in Y country.” If we don’t have that data available, right? We will make it available usually even before that company’s ready to kick off. 

That’s not the kind of thing that we want customers to have to think about. Our goal is to abstract that all away. 

Gary Nowacki: So, when you make a prediction, like you said, “Hey, this new product you’re about to launch, it’s going to do well in Germany. I think,” as you said. “But not in Italy.”

Do you actually go back and test against actual live consumer preferences to see if your predictions were accurate? 

Jason Cohen: So, we go back and re-profile entire markets every three to five years. So, depending on the rate of change in a country, we will go back and we’ll update those models. 

So, you know, France is a long running stable, intense culinary tradition that might be closer to five years. When we think about China, where we have one of our standing panels, China’s evolving so rapidly that we have to collect data there literally every day. It’s one of our standing panels, right? And so, depending on how rapidly a country is developing and how rapidly it’s changing, depends on how frequently we go back and update these models.

And, of course, we retest both similar products that we had from the prior survey and new products that have launched. And the same products that we used as references. When it comes to validation, when it comes to work that we do with the company, our results are heavily tested.

Every single major multinational CPG that uses us, runs us through a series of validations before they stop running CLTs. And before they stop running traditional consumer tests. So, we have multiple double blind validation studies that are done. 

One of the most public ones is the one that we ran with Ajinomoto that was validated by Ipsos. That was for predictions on products from Japan being retargeted and redeveloped for China. But we do a number of different validation studies, both internal to companies and on our own. 

So, you know, we see on a seven-point hedonic scale, we see less than 1% systemic error per preference score, which is less than a seven point error across the scale. So that’s not exactly—you can’t talk about accuracy quite that simply—but it’s approximately 93% accurate, which is far more accurate than any CLT has ever been. 

Gary Nowacki: Hmm. And so, are there certain—you know, based on all the testing you do, this massive database, all the different experiences with different CPG companies. Are there certain categories where you know that AI is going to be able to predict consumer preferences better than other categories?

Jason Cohen: Our focus is on products and categories that are ready to eat and ready to drink. So, as we get into more—less homogenous products. As we get into things like hamburgers or prepared culinary foods, there’s a combinatoric explosion where people like options and they like customization. So, if you order a hamburger and you say, “I want mustard, but no ketchup. And onions, but no pickles.” Right?

It’s a very different experience. So, while we do work on alternative proteins, and while we do work on hamburger patties, and alternative hamburger patties, right? Working on the entire burger is much more complicated. 

So, we would say that CPG, RTD, and ready to drink, and ready to eat are the areas where it’s having the fastest impact or the largest impact, because those are the most controlled.

And then as you go into less and less controlled categories, you know, there’s always a chance that your consumer doesn’t follow directions. There’s always the chance that, you know, you work on an instant noodle and the person uses tepid water, and, you know, doesn’t stir it, and doesn’t wait long enough.

So, um, certainly I think that the ready to drink and ready to eat are the fastest evolving and beverage is always faster than solid foods.

Gary Nowacki: Hmm. What other sorts of—since your company is so technology focused, Jason. What other sort of technologies are you on the lookout for as the whole industry evolves forward? 

Jason Cohen: One large—it’s more of a concept, I guess, than a technology. But there’s been this idea of personalized products that’s floated around in the industry for a long time.

And I always tell the story. I went down to Atlanta when I went to the world of Coke and the idea behind this is right now, what, when companies talk about personalization, really what they’re talking about is customization. You know, to go back to that burger example, customization is where you have to say, I don’t want onions on my burger, right?

But that’s not personalization. Personalization is the company knowing what you like and dislike without you needing to make any active decisions, potentially knowing what you like and dislike. More than you’re aware of yourself—either because you haven’t tried the product or because you’re not a flavor expert.

And so, you know, that to get back to that story. I went down to the World of Coke, and right at the end, right? They have all of those sodas from all over the world that you could try, and they had a Coke freestyle machine, and that’s where you can customize your own soda. 

And so, if you just stand and watch people for a little while, they’ll say, “Oh, well, I like Coke. I like vanillaCoke. So, I’m going to do Coke with extra vanilla.” Right? And then they take a sip of it, and they make a face and they blah. Right? And it’s because they put too much vanilla in it. They’re not an expert product formulator. They don’t, you know, they don’t know how to balance the amount of vanilla flavor with the rest of the soda.

And so that’s an example of technology allowing customization. But that technology needs a brain, right? Someone needs to build a brain that says this is the optimal amount of vanilla for, you know, James. And this is the optimal amount of vanilla for Nick. 

So, I think one of the things that I’m hoping for is that these machines that are going to allow products to be formulated and personalized at the point of sale, or the point of consumption start to become smart. And, and if we could be part of that, that that would be a really interesting way of using this technology right down to the personal level. And so that’s something that I’m excited about. 

Gary Nowacki: Hmm, I’m here with, Jason Cohen, founder and CEO of Gastrograph AI. Jason, any other—looking out on your horizon, your crystal ball, any other trends or tastes or flavors or experiences that you think are going to emerge in the coming years?

Jason Cohen: I’ll leave you with two, maybe. I think one, we’re going to see quite a bit more international products in the United States. It used to be that the United States was generally a place of innovation ourselves. And we exported our products to the world. I think that we’re going to see that trend reverse.

I think we’re going to see a lot of products made in the international market coming into the United States and gaining a foothold. And maybe in some scenarios, even the dominant position. 

And I think the second one is that, you know, we’ve been in an emergent-to-stable preference transition originally from, for increased levels of bitterness, from the 90s into the 2000s, and for sourness from the 2010s until today. Those have stabilized.

We’re now starting to see an emergent-to-stable preference transition for earthy flavors. And I think that there’s going to be a lot of excitement and a lot of development in earthy flavors in the very near future. 

Gary Nowacki: Hmm. Hmm. Yeah, it’s interesting. I, you know, I have a personal bias against too many hoppy beers, and it seems like every craft brewery has just decided to launch five more hoppy beers. So, I hope you’re right about maybe earthiness or other new tastes. 

So Jason, before we go into wrap up, any other words of advice you’d like to share with our listeners or additional information you’d like to share with your peers out there in industry?

Jason Cohen: Well, I would say for anyone who’s considering developing a new product, whether you’re a startup, an emerging brand, or even if you’re in one of the multinationals. You know, the trend that we’ve seen is that companies are saying that doing—running consumer tests, testing the flavor of the product, making sure that we get the product experience, right? It’s too long, it’s too expensive and it’s not predictive. And instead of saying, there’s something wrong with the methodology, they say, “We’re not going run the tests.” Right?

“If the tests aren’t predictive, then we’re not going run the tests.” And that’s saying, you know, that’s throwing up your hands and saying, “Well, we’ll launch it. And we’ll see what happens.” And we see a shocking number of companies do that. 

And so, my advice is that when faced with information that is not helpful, the goal shouldn’t be to ignore the information or to reduce the amount of information you collect. It should be to change the methodology and to change the systems in which you gain that information from.

So, I think that there’s a real role for technology in the food and beverage industry. I think that there is a growing gap between what companies are doing, and what is possible to do. And I think a lot of that gap comes from, you know, increased pressure, reduced budgets, a more competitive market.

And the answer to that is to make investments in better technologies and better methodologies to gain a competitive advantage. That’s my biggest piece of advice. 

Gary Nowacki: Sounds like good advice. And I bet you, a lot of our listeners out there are frustrated with failures they’ve seen in the past.

So don’t give up. Try applying some new technology in a different approach. 

So, I want to thank our guest today, Jason Cohen, who is founder and CEO of Gastrograph AI. Check it out, really neat company. Jason, thanks so much for being on C to C. 

Jason Cohen: Thank you for having me.

Gary Nowacki: Thanks for listening to C to C, where we cover innovation in the food and CPG business from conception to consumption. Just type the letters C-T-O-C, no spaces, to find us on iTunes, Stitcher, Podbean, and Google Play.