Updated: Sep 6, 2021
Welcome to "AigoraCast", conversations with industry experts on how new technologies are transforming sensory and consumer science!
AigoraCast is available on Apple Podcasts, Stitcher, Google Podcasts, Spotify, PodCast Republic, Pandora, and Amazon Music. Remember to subscribe, and please leave a positive review if you like what you hear!
Daniel Protz is the Founder & CEO of FlavorWiki. He holds a bachelor's degree from Dartmouth College and an MBA from Stanford Business School. Daniel started his career in computer-driven commodity trading before moving into the e-commerce and digital industries. Since 2015, Daniel has been working enthusiastically towards democratizing the world of consumer understanding and building a better purchase experience for consumers around the world. His professional background includes top global firms like Goldman Sachs, Deutsche Bank, and Groupon, where he designed, built, and implemented technology to support inventory, customer service, content publishing, and pricing automation for 1 billion+ USD e-commerce businesses. He’s a keen digital enthusiast and a strong believer that FlavorWiki is building the premier consumer review and discovery platform to support the digital brands of tomorrow.
Transcript (Semi-automated, forgive typos!)
John: So Daniel, welcome to the show.
Daniel: Thanks, John. It's a pleasure to be here. Thanks for having me.
John: Well, that's great. Alright, Daniel. Well, I think it'd be good to talk about your background and also FlavorWiki and what it is that kind of service you provide so maybe you can take us through your journey into sensory science and then talk a little about FlavorWiki and the services provided.
Daniel: Sure. I mean, you read a little bit about my background there. I've started working actually in the finance industry in 1999 and had mostly positions in finance and then e-commerce up until founding FlavorWiki. The way that I got into it was kind of by happenstance working in the e-commerce industry. I wanted to find a way to better use this type of data in online marketing. Right? So I always say, like when you look at databases or marketing CRM, if someone bought a green sweater or leather shoes or wooden share, that's actually all information which we can quantify in a database and understand. But when you start thinking about the intensity of bitterness and chocolate, or even if you're talking about aromas or things like this. These are much more abstract ideas that are more difficult, therefore, to figure out what might the customer like next. And so that was actually the problem that I was thinking about trying to solve which led me to meet and talk to more people in the sensory science area which had very really little knowledge about it at the beginning. And then over time, we developed this methodology that we have and have more or less in applying it to an expanding set of consumer research activities since then.
John: Okay. Well, also for the listeners who may not be familiar with FlavorWiki, what it is, what it does, and what services you provide?
Daniel: Sure. So, we do two things. We have a software as a service platform which we license to food and ingredient companies. They use that for internal product evaluation and consumer testing broadly. So if you can think about it, it's a little bit like a survey monkey on Steroids. It's much more user-friendly and easier to use than like a lab sensory tool. It also doesn't have quite the level of detailed lab type of testing that you would have there. We really focus on the ability to build this what we call consumer journeys that can be delivered on a smartphone. We ask questions. We deliver videos. We allow users to upload videos. All these types of ways in which we could gather information from a consumer out in the world. We build into that including this type of pairing question that we use for building sensor understanding of products, and that includes things like flavor, obviously, texture, mouthfeel, aroma. But we've also used it for emotion, packaging, profiling. Again, like these kinds of abstract ideas where the regular consumer doesn't understand how to scale something on a scale from 1 to 10. That's an abstract concept. By doing this pairing question, we're able to collect that information on the sensory piece, plus the hedonic piece kind of from anywhere in the world we might want to do that. We license that software to food companies. They use it. Ingredient companies actually use it in the selling process. They give it to their sales teams, and then they can get feedback on prototypes. But then as well, food companies are using it to do prototyping with consumers. Many have consumer panels that they even send products to so we can support them. Then the other part of the business is we have a growing footprint around the world of consumer engagement and consumers can get to go out and buy products in the marketplace, or we can send the products in their home. And now we're active in 21 countries and the applications are available in 18 languages. So we've worked in 16 countries and we don't have any facilities or brick and mortar labs. We don't do any of that kind of thing. We try to do everything in the home, on the go, or wherever the consumer is in the world. Again, trying to focus on this leveraging digital in these areas. So there are some things which we don't provide which more traditional research houses are better at. But we really try to focus on the digital experience.
John: Right? Okay. So on the one hand, you have software with somebody if they have their own panel, they can just use your software?
Daniel: Absolutely.
John: And then you separately have the service where you maintain this kind of growing community. Okay, so to what extent is that a standing panel? And to what extent is that kind of new consumers?
Daniel: It's always growing. Usually, we try to find when we do a customer project, about 70% of the users will be new to our platform. And typically you're looking for people that don't do market research often. So we do a kind of gamification and sharing methods to get people to grow the community, and sometimes these things will actually go viral. So we really act in this way more like an online marketing company or like an e-company because we're trying to get people to go buy specific products. About a third of the team is just totally focused on expanding into markets, getting consumers to do tests. We do a lot of our own tests and own surveys and own little types of fun research projects to grow the group because we need to have enough people in place to run these types of projects for our clients when the time comes when they need it, and you probably know. Usually, when they need it, they need it pretty quickly. And so we try to focus on a super-fast turnaround time. I think the quickest we've done a full category appraisal in a market, say, up to twelve products or so and consumers from brief to finish was nine days. So the execution and data collection can be quick. And all the software, all the analytics is automated. We build all the statistics and output everything as automated as we could because that just puts on the delivery time.
John: Yeah. That's good. So you mentioned people buying products in the market. Now someone could come to you if they wanted to mail prototypes, would that be an option or how does that work?
Daniel: Yes. That is a growing option. Actually, when I started the company in 2017, having just come out of running a logistics operation for Group On, I didn't want to touch sheet sample shipping at all. It's clunky. It takes a lot of time and energy. But actually, since then, that's now four years ago, there's been a huge amount of advancement in almost every region in the ability to ship samples and to find service providers that will support that kind of thing. End of 2020, beginning of 2021, we started doing this and now it's actually going really well. I've been very surprised at how much more mature the logistics market is for this type of need and I think it's driven by platforms like eBay and other types of platforms like that where you have a lot more small entrepreneurs that are doing online commerce. There are logistics operations, then you can hire that are supporting these small operations and we look a lot more like that, say, Walmart. We don't have tons of volume obviously, these are small shipments. So that's been, I think, a good development probably coming out of covid partially. There are some things we don't touch, like ice cream. Forget about it. There are some things where even the shipping process would potentially disrupt the outcome, and therefore we just advise the client, like, potentially there's another option to do. There's a different way or maybe you should use a CLT in most cases.
John: Right. That's interesting. Alright, now, what about kind of blind versus branded testing? Because if somebody goes and buys ice cream, obviously it is branded so how do you manage that kind of thing?
Daniel: So we've done a lot of benchmarking between outcomes of branded and non branded. That's I think one of the major activities we've done together with clients is in a pilot phase which is like, look at their output that they have from usually, they have a lot of unbranded tests, particularly in benchmarking. All of these brand companies. They're doing really frequent benchmarking. And normally what they will do is they will give us projects to do as a pilot, which they already know the answer to, right? And so they basically are like, yeah, we know these products very well. We've looked at a variety of them across different markets. So we're going to give this project to FlavorWiki, see what they come with, and then benchmark that our results against what they have which is blinded. And in almost all cases, basically, the conclusions are the same. We have ways in which we ask questions about the product expectation, about the impact of the brand and the package. But at this high level, a number of responses usually thinking about in a test like 120 to 200 responses. We find that it's not as much of a factor as you would think. And interestingly enough, in a lot of cases like in private label brands, you find that the expectation of the brand is negatively impacting the perception of the product. But once they try it, it actually wins on taste. I think it's a different angle. I mean, certainly, there are companies out there and also ones that we work with that really always wanted to be blinded. And they'll even ask for market products to be blinded and we can support that in some cases as well. But for the most part, when you compare the cost and execution time of doing that with the ability to just quickly get out and get a market read and something that's already in the market. Usually, they actually decide to do it branded just because it's so much quicker and it's more cost-effective.
John: Yeah. But one of the things I like about your approach is really someone started off as an outsider to the field. I think that to some extent, I think I've been, I mean, my father is a sensory scientist. So I've been around the field but I am a mathematician. You know, I come into the field, and I kind of willing to try new things to an extent because I have a different background. So you come in and I like that you're trying different things, your experimenting, questioning assumptions. Now, one of the things that you've done that I think is really innovative is this kind of choice-based sensory profile. So can you talk to our audience a little bit about that? I think that's a really interesting technique.
Daniel: Sure. As I mentioned before, I kind of came upon it by happenstance. Then again, I wanted to solve this problem of how can I understand what every person on the planet tastes. Like theoretically, how can I understand what this person that's shopping in my online shop tasted what they like and try to figure out what I should show them next from this sensory or even aroma point of view. To me, looking at this as an outsider, it was pretty clear that scaling things, it takes a long time. It's abstract. People will always give you a sort of a different frame of reference. And so I really stumbled upon some research that had been done in the social sciences using these pair-based responses which I think, it's been around for 100 years. This kind of thing in different testing, like asking people one or the other. And when you couple that type of test which is the research I found on had been done using paper ballots. And I said, well, look, with digital, we could easily build a pretty simple machine that would show different pairs to people. And so if you then sort of show different pairs to people with a set attribute, let's call it ten attributes or twelve attributes, then we could start to figure out how a group ranks them. And I intuitively sort of knew that one person would probably need not be that accurate at first. But what I did is I actually we built this kind of very simplified version of the tool in WordPress actually, this is when I worked in Groupon, we did this, and we found people online, and we asked them to go out in different parts of the world, I think Philippines, Brazil, different places all around the world and buy Toblerone Chocolate which apparently is, so I was told is actually all made in the same factory, or at least at that time was. So it had a pretty similar formulation. And we were just looking to see whether or not these regular people could create repeatable profiles, basically like okay, take this test to teach you, a minute and a half answer these pairs, and then we would take them away and do this in Excel and process it and things like that. And then we would ask them to do it again a week later and a week later, in a week later, a week later. We were giving them a dollar or two or something to do that. And we found that number one, most people could create repeatable profiles even though of course, they could never remember all these different pair responses. It's too complicated. And then the other thing we noticed is that I think you probably won't be surprised they actually became more accurate. They became more repeatable over time. Right? So the second and the third one was a bit closer than the first and the second. And the other thing that I noticed is that people that had kind of like, a food services background because there were some chefs, for example, that signed up to do it. They were actually more accurate right away. Anyway, that was the way we started kind of experimenting with this. And then we realized, well, look, if we sort of look at all of these different patterns of these responses, and we start to try to make it more accurate and repeatable by looking at different types of algorithms that we can run or ways that we can ask the questions in different ways or different formulas we can use in order to tease out the difference, different things we can look at to figure out whether or not there's statistical significance between two things or whether or not it's just random. And that's basically what we've been spending our time on since then, trying to make this whole thing more accurate. But obviously, then building the platform and the other technology as well. It was really like it was a journey. Of course, no one really thought much of it at first. And then we got into accelerator here in Switzerland, which was sponsored by the two largest producer retailers in Switzerland that both have pretty big private label manufacturing. And we got a chance to sort of bench market there against trained panels. We got involved in some research with trained panels with applied sciences and universities, and basically just got more and more trials to see how close this could be to a traditional trained panel. But the real test has barely been with the customers. Like the customers that have all of this data, and they can say, well, yeah, it's basically pretty accurate. It's about 85% to 90% is accurate up to a level of about 15 attributes. And now the nice thing is you can use the attributes of the client. We're not fixed on a set language. We can use different languages. We can do a lot of different things. And we've also figured out how to mix and match and compare products and see what's significant. It's been kind of an outsider's view of how this might work. Over time, we've brought people in that have real sensory and consumer insight background who can advise on how we can and we have to make sure that we respect, like the guard rails of good practice. Also, when we explain it, I think we can explain it in a way now, which usually people understand how it's working.
John: Yeah, I think it's a pretty fresh idea. I like it. I mean, it seems to me that basically what you're doing is you have the choice-based data, and then you compute what would be utilities, basically content utilities. And you treat those like intensity values which I mean, they're not exactly intensity values, but you can treat them that way. My guess would be if you make a PCA map out of that, you probably would get a map that resembles the PCA map that would come out of the panels.
Daniel: Exactly right. That's part of the analysis we provide if you're doing a category appraisal with a products, is we do a PCA analysis and correlation analysis to help the client understand what are the drivers of preference and drivers of negative preference in the product which is what can help the product formulator. And then when you combine that with other types of analysis, like penalty analysis, it looks pretty similar to what they're used to seeing in a market, a project, which means they can use the data and it's pretty actionable for them.
John: Right. Okay. If I make a suggestion, are you already doing, I mean this may be a little nerdy, have you ever done any multiple factors analysis MFA? If you were to take the expert panel data, and so you have two tables, right? You got your extra panel, you got your consumer utilities, run an MFA. That would be quite interesting because you might see differences in how consumers use specific terms.
Daniel: Great. We've had a couple of instances where the client also decided to do a trained panel evaluation of the same products. Right? And yeah, you're right. We're seeing these types of interactions, and you can upload actually your own sensory data into the platform and run similar types of analysis on it. You can also do a traditional sliding analysis. The way that our profiles look is a little bit different from an absolute scale because you can imagine it's all relatively based. And the algorithm also doesn't know any trained scale, but we can normalize it to a scale that the client would better understand because we can look at their trained panel data and then basically sort of rewrite the algorithm so that the output will look more like what there is from a relative basis. But I think your point is exactly correct. Is that really what you care about is doing the correlation and PCA analysis because that's where your conclusions come from and in our data, you can do that directly.
John: It's interesting. Okay, I think that's really good. Alright. And so then there's kind of an AI piece, too, that we've talked about where you mentioned already that you do some research that isn't client funded. It's just your own research for the sake of building out your platform, building our community, and based on that, you have some predictive models that you've been training. Can you talk to us a little bit about the predictive modeling you're doing?
Daniel: Sure. So this is really a function of just wanting to have information to try to make the model of doing the profiling more accurate, and then also the model of being predictive about it, right? It's probably not rocket science coming from your perspective either, is that if you basically say, well, look, I understand the Euclidean space of what a person is tasting here because I understand the PCA. I understand their relative liking. I can start to look at how they react to different intensities of different combinations, and then you can start to make a predictive model of would they like the product or not a different product that's not a net set. And of course, we can test that, right? We can ask them to go try that and see how much they like it. This, I think, has been the most closed loop that we could develop because developing a closed loop with a trained panel, we didn't have so many opportunities to do this because you have to pay for a trained panel. And then the question is, how trained is the panel? But if we develop this model and actually become better and better at predicting liking, then we know that we're getting better at constructing the data, but also we're just getting better at using the information and predicting what the user is going to do. It's interesting, this kind of predictive analysis. I thought also that personalization and all this type of predictive stuff would become very involved like the e-commerce world, for example. I haven't really seen it happening to two months. There's a couple of platforms that do this kind of thing. But I think, for the most part, the retail industry is still the retail industry. It's largely price-driven, and the retailer sort of decides. I'm going to try this product put on the shelf and see if it works. So the predictive element is still largely left to the food companies and the suppliers, flavor houses, and anything suppliers. But I think it will come. I often talk to people about the comparison between innovation cycles in the food industry which I'm super new to and innovation cycles in other industries which I'm not as new to e-commerce. And I know a lot of people that have done a lot of you know had a lot of startups in different industries. I think because food is a pretty emotional product. People tend to like to do things in a habitual way. They kind of are connected to things they ate when they were younger or whatnot. I think just the cycles of change and the demands that the consumer puts on the industry are a little bit longer-dated. It's not like you can decide to take a taxi and then take a car the next day. People don't change their eating and drinking habits as quickly.
John: Right. Now, I think that's a good insight. Maybe if you're a foodie. I don't know if you're a foodie. I'm a foodie, but it's a little different like you and I might be interested in exotic experiences, but even day to day, you know, I like coffee in the morning.
Daniel: Yeah. Me too. I tried a lot of different things as well, and I think most foodies are this way also, like 60% to 80% of what they're eating is kind of core, and then they're experimenting a lot around other things, right?
John: Yeah, that's right. But I'm not going to change my breakfast behavior very easily. Yeah, people keep trying to get us to eat insects, but I don't know if it's ever gonna happen.
Daniel: I won't do that. I think the insect industry in animal feed is a really good commercial idea. I think that makes a lot.
John: Right. Yes, and lab-grown meat also. Okay, apparently there's no lab-grown mouse meat for cats, so that's coming I guess. Cat food is going to be lab-grown. I guess it's some progress. Well, let's talk a little bit about what you're kind of excited to work on. So what are the areas that you see kind of FlavorWiki focusing on over the next year or two? I know it's hard to protect technology past about two years, but what are kind of some of the key initiatives that you're working on?
Daniel: So there are two areas where we're I'm really excited to work with customers. We always say do the best job for your best customers because that's also how you learned the most. And one of them is actually in e-commerce. We're starting to integrate the software more into e-comm platforms. It's not so much related to this predictive analysis that I mentioned before, but it's more related to how can you use this fast iteration and reach that e-commerce has to do the test and learn type of projects. Right? We're seeing some bigger companies start to partner with these e-comm platforms, but you can't just do a five-star Amazon rating. Like how do you bring sophistication to that interaction where it's also easy for the consumer to do it? So for me, I think that's actually where we can show the most leverage in what we're doing because you can also give that to a small company. You can give it to a restaurant. How can we sort of make this type of technology more consumable for people that don't have a background in consumer science? From my perspective, that's where we can have an impact on the industry. And then the other thing that we're actually building a lot more these days is productivity management. We have a lot of pressure to execute projects quickly. Get the data, get the data, get the data. But then you look at the internal like friction between different departments inside a food company. And you see that actually, just the process management that they're dealing with is causing them to slow their projects down quite a bit. And so the more tools that we can build into our platform that help them interact, that help them sort of track a project that helps them keep track of the data. I mean, I think you mentioned you do a lot of this like digitizing all this old data that people have. There's so much information that's been collected and paid for over many years that's just laying around and part of that is just because there's no system to really manage it well. So we tend to respond to the request of customers, and that's the types of things that they're coming to us with her saying, well, I need to manage these things more like a pipeline and make it more efficient so I can get them tested faster, et cetera. And there's plenty of things, of course, you can do in digital tools that help them do that. Again, maybe it's not so much the sensory angle, but a lot of the devil is in the execution, right? At the end of the day, what you're trying to do is make good products and make a business, and the more we can help our clients be more efficient, the more successful they'll be.
John: I think you're definitely increasing the agility of the food industry and that's important and also something else I think is interesting over the next few years is, I think sensory and UX are going to have to merge that UX and to a large extent is I mean, you're talking about the science to a large set of sight and sound, and then sensory tends to focus on the chemical senses. A lot of the same tools are at work, and I think those two fields are going to come together. And I really like that you're bringing so much technology and the innovation into the space. So I think that yeah, I think you're doing a great job, and I would encourage our listeners to reach out to you. Yeah, actually, let me just ask, what are good ways for people to get in touch with you?
Daniel: So our website is actually business.flavorwiki com. If you go to flavorwiki.com, it'll redirect you to the business site. The main site is where we let users sign up. And then you can email me at daniel@flavorwiki.com. I'd say I read every email every day, and I tend to do that sometimes pretty late. It's a habit I have had from my old life. If I get behind, I get too far behind. We try to be responsive and we're always looking for we like tough projects. As far as its effort when companies come in, they say, well, look, we have this, usually it's like a logistical problem or it's an issue with just the execution which is not normal. And those are the ways in which I think we can help a lot because we can be a little bit more agile and tricky. We can get people to do things that maybe would be impossible to do in a traditional setting.
John: Okay. Great. And you're also on LinkedIn? I think that's how we connected.
Daniel: Yeah, sure. You can search for my name. P-R-O-T-Z.
John: Yeah. We'll put the link in the show notes.
Daniel: Flavor Wiki. I think there's probably not many companies.
John: Yeah, all the links will be in the show notes and that's good. Alright. So, actually Daniel we always like to conclude with advice for young scientists. In your case, you could leave it a little more open, young people, young entrepreneurs, what advice would you have for someone who's 25 years old? It's kind of enthusiastic, is interested in, says, the food space text base, what advice do you have for a young person?
Daniel: I always encourage young people to go work for other people because a little bit less, I mean yes, it's less risky, but it's also less risky in the sense that you definitely will learn something more. I've been pretty impressed with how much money has been flowing into the space that's investing in food products. A lot of it's going to grocery delivery investment, which is how valuable that is. But in this alternative meat space, I just think that you're seeing VC money flow into the industry and food in a way that it hasn't before which is going to create I think a lot of disappointed investors over time because I'm not sure they're doing it for the right reasons, but it's definitely creating a lot of opportunities for young people to go work for companies and see what it's like to work for a small company. I mean working for a small growing company, you just can't learn fast enough like when things change. It also happens still for us, we've been around three and a half years. If I think about where we were at two or three months ago, it's a ton of change and if you're working for a company that's evolving that quickly, I think you're learning the most.
John: Yes. That's excellent advice. You know, we just hired our 10th employee at Aigora and we're growing rapidly. And I see it especially people who come from larger companies that come work for us, the speed is so much faster when you're in a small company.
Daniel: It's more fun too, right?
John: Yeah, definitely agree. Alright, Daniel, this has been a pleasure so thank you very much. Anything else you want to say to our audience?
Daniel: No, thanks very much and I hope everybody's having a good, safe summer.
John: Okay, great. Thanks.
Daniel: Take care.
John Ennis is a leading expert in sensory science and consumer research, with extensive experience in statistical analysis and product development methodologies.
