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!
Clare Thorp is currently the Senior Vice President, for Creme Global where she works with companies and government agencies to support their decision-making processes through the harnessing of big data and deployment of advanced computational tools.
Since 2010, Clare has held executive leadership positions in trade associations and non-profits including CropLife America, the Biotechnology Innovation Organization, and the International Life Sciences Institute of North America.
Cronan McNamara is the founder and CEO of Crème Global, where he pursues his passion in applying data and predictive models to solve problems in order to help people to make better decisions.
Prior to founding Creme Global in 2005, Cronan worked in financial risk analysis and derivative option price modeling with Merrill Lynch Capital Markets Bank. Cronan then moved to Trinity College Dublin where he adapted the same mathematical methods to the food safety sector while developing the science and technology that underpins Creme Global.
Transcript (Semi-automated, forgive typos!)
John: Cronan, thanks for being on the show.
Cronan: Thanks, John.
John: Alright, great. So we actually have a lots to talk about. So we laid off just for our listeners who may be not familiar with Creme Global. I just like to hear your thoughts on what is it that you'd like to know about Creme Global, about what you do, how you apply data to these kind of real life problems. So maybe we'll hear from Cronan first and then Clare, hide your thoughts.
Cronan: Sure. Thanks, John. So with my background in Physics and Maths and computing, I'm working in the field safety sector, I was always passionate about bringing real data to bear on problem solving. So having founded Creme, it's been a community college. We had some software for food safety risk analysis. And over the years, we've created a platform for managing data and deploying scientific models on that platform in order to ingest data. The data we can get our hands on which helps us to quantify food intakes and consumer issues and chemical levels and products. And we put those in a model and we provide a lot of training and support and services around those models. So to help industry and government and even research groups to understand the intakes and impact of chemicals and nutrients on consumers.
John: That's interesting minutes because we were talking before this call about how, you know, Creme is I think, maybe closer to the objective side of, I'm in consumer science so we do a lot of work, you know the consumer experience, the sensory experience, and the preferences. But it seems like a lot of the work you do is maybe a little bit more on the objective side, that kind of hard data or behavioral data. So I'd like to get to that second, and Claire , I'd like to hear also your thoughts. What are the what would you like to add as far as you see kind of the important contributions that Creme Global makes to your, you know, problems?
Clare: Yes, thank you. I've never in Creme global, actually, for a very long time, ever since they started something. I used to work in the Irish governments and I've dealt with food and safety research and program there. I got to know Cronan and Creme Global at that point, because if we had any issues contaminants to our food, we needed to understand how people were going to be impacted by this. And Cronan's company was doing exposure's science before the exposure sciences was introduce to science and very robust models that will lead us to understand, you know, having an impulsive making decisions we had to implement as a result. And so I joined Creme Global fairly recently. Perhaps one of the reasons I enjoy working with them is because it's largely and very complicated things, and messy data, and robustly put together into a format like model which allows you to then understand what data is on. And it's really provides insights that as an animal scientist would normally have to do experiments like experiment by experiment by experiment. When you can actually harness existing data, you get a much bigger overview information. So that's why I enjoyed this company I'm working with.
John: Yeah, interesting. So something I actually kind of personally interested in is the question of specialization versus generalization, which is that like you are developing tools that maybe are quite general, but you have a very specialized application, right. Because you're, generally speaking, the kind of food safety or regulatory side of things as far as you know, I understand. How do you balance that? Because it is tempting for me to just say, look, you know, I can solve lots of problems involving machine learning or whatnot. How do you decide what problems to work on and which problems are kind of out of the scope of your company?
Cronan: That's a really interesting question. And I suppose it's a business strategy question as much as anything. And then I also same view of our platform technology as being capable of dealing with many risk problems and exposure problems. And I had a background in financial risk modeling, as you mentioned yourself. And so from a business point of view, it's challenging for a company in a kind of scaling company to really penetrate multiple markets. So you kind of need to focus your expertise on your marketing on a topic. And because of the genesis of the company spinning out of Trinity College as a food safety risk company 15 years later, here we are still working in that sector, you know. But I always had the vision of broadening out the company and we have built our technology stock to be able to do that. And all the time we've worked in other sectors from personal care and cosmetics and to even looking at microbial modelling using AI and machine learning. So I see opportunities to expand. And I always like the kind of Amazon story where they were, you know, the book store and now their store, right? A lot of them, I suppose, resources and expertise and a bit of luck to achieve that. I see the opportunity to become a very specialized company. But then to broaden out from there into different verticals and have a platform technology underlying that.
John: Yeah, it's interesting. And maybe Clare, a question I would ask you is, what would you think are the kind of specific cause I find like sensory and consumer science, well, it's true that, you know, apply general machine learning technologies. There are specific problems within sensory science, especially that I think would give you know a general data scientist problems if they didn't weren't familiar with the field. What do you see as the kind of specialized problems within food safety? That's someone who just is an expert in cyber risk management and a general sense, would run into trouble? What do you see as like, where's the subject matter expertise really needed in order to make the most of the solutions that you're providing?
Clare: Well, I think one of the things that's becoming increasingly important in food safety is the ability to have a multidisciplinary approach. And to be able to harness the various newer technologies that are out there. So, for example, if you look at the predictive model that Creme's developed, the microbiome and you have oxidation-regulation doing, we're moving very much in traditional culture testing and finding out that something's already a problem into using really powerful quantities of genetic data and modelling to be able to predict the likelihood of food safety impact. And the challenge with that is that you're having conversations with people from very different backgrounds in very different ways in looking the same problem like having an elephant in the room. Some people look at the kneecaps and people look at the tech. At some point you need to be able to step back and actually see the entire elephant. That's some of the challenges that we're facing moving forward, is when you're bringing in these new approaches and this knowledge, how do you allow everybody, enable everybody to see the entire elephant. So they see it together.
John: And Cronan, what do you think about that?
Cronan: Yeah. I think, I was just thinking that we've hired a very multidisciplinary team, as Clare was mentioning. So we have the software engineering user experience designers, the front-end web development type people to make things user friendly. But in the middle, we have a data science team which also has people at biostatistics, a bit of toxicology experience. And then we have food science expertise. That's a nice, broad team. And when we're working on projects, we'll often have a nice cross-section of people on the project where they can describe the aspects of it. And I like work on my role and draw those flow charts and diagrams. I'm on a whiteboard. You know, like, you know, behind you to kind of bring up there. You know, people can't see on the podcast. But John has a whiteboard behind him, I'm sure you use that.
John: Yeah, I use that many times.
Cronan: Yeah. So, you know, bringing those knowledge together would a bit of them algorithmic knowledge and expertise to say, well, this is how we doing some data sources to make this all make sense. And that's what magic kind of happens, as you say, in a specialized team that can deal with specialized problems and also speak the language of the customer, you know, understand their challenges and then be able to translate that data science back to the customer in a way that they can understand.
John: Right, culture of course, is really important. Every area has its own culture, and I think that's right. I mean, I just see this with like clients where they get one very large company that I work with. They made a huge investment in data science. And then the consumer science team told me they literally couldn't understand the science we're talking about. They had some meetings with them and they couldn't understand them. And so they just gave up.
Clare: I think that's one of the things that Cronan speaking to, is that whenever we sit down to essentially solve problem, it has to have conversations coming in from all sides because you're trying to get the science to speak through mathematical languages and programming. And so if you don't have those conversations and you end up with a disconnect and it's fascinating because I have an online science toxicology risk assessments type of backgrounds. And then I think people who are turning those things into, let's say, a model that will look the users to, I don't know products that get into your yards to control fleas, ticks, and mosquitoes and stuff like that and somehow say, yeah, this is what we need to know. You got to work at, okay, well, how do we answer that? What's the data that we need and what's important to, what actually matters? So how will it function to the yard? Why is that matter? How much of this are you using? What it does not matter? How do you pull those pieces together so that you actually create a model? That's makes sense. And it's fascinating. So, yeah, it is a bit like some magic cooking in some ways putting these ingredients together. But at the end, you come out with a very nice cake or something and make a sense to your discipline. It tastes nice, it smells nice and it works.
John: Yeah. Actually, there should be even more collaboration between, you know, people like you and people like me in terms of the fact that you've got food safety concerns, you have dietary, I'd like someone to approach an obstetrician because I know you all have done a fair amount of work in that area. It's very interesting that not only should we be nutritious for people, but there's also the question of whether or not they're going to appreciate, right? It's all fine and well to tell people that something is good for them, but they don't like the way it tastes then, you know, doesn't really help anybody. Before I move on, though, I do want to talk a little bit about you've touched on it somewhat here with the subject matter expertise. Because I found with machine learning, there are kind of two ways that the subject matter expertise helps. And one is with kind of feature selection and engineering where you're going to, because the thing is, unless you have an essentially infinite amount of data, it will be helpful to have some idea which features are likely to be informative, right. And another, I mean, actually, I run into this with my own client work or sometimes I think data scientists to come, especially from the tech sector where they're used to having enormous amounts of data, don't really welcome inputs from consumer scientists. As far as what may or may not be helpful for the model, they just want to let the model figure it out. But actually, it is helpful if you have some idea which signals are going to be informative and which ones aren't. Then the other courses is interpretability. And I don't know how important interpretability is for your models, but, you know, on my side, where we need to be helping product developers to develop models, you know, to predict for certain groups, what's going to be appreciated or what, I mean there's different types of models you can build. But at the end of the day, usually a product developer is going to have to use the model. And so that was going to have to be some level of interpretability, even if it's sysmetric interpretability. So if the features are, you know, maybe they're very predictive, but they've been engineered to the point that they can't be used by product developer. It doesn't really help people. So I'm kind of wondering, you know, your thoughts like where does subject matter expertise help you all? When it comes to your overall exercise of you know, risk management and advising governments, I suppose.
Cronan: Yeah. Well, maybe I can try and answer that. Just the two different types of models at work. Sort of working in both areas. The type of machine learning models where you try to let the data speak for itself and you tried to keep your scientific opinions to yourself and let the data kind of inform you. And then there's another type of model that we not aware of building up a lot, which is kind of trying to combine that in a more scientific approach to like looking up research, finding the data, finding what people have found in terms of correlations or causations in different finding parameters in papers. I'm trying to put all that together in a model that's a bit more from first principles. So you're building up the populations, you're building up food information. You're trying to put the formulations together. The bioavailability, how do you find that? Well, you might have to research some papers and then put it in a distribution for that parameter and your model. How do you find that with another data source? So it depends on the type of model you're building. And I think having that ability to distill information from the literature and then to translate that into an algorithm is find it an appropriate data set to put into the model to represent that. That's all a very much a combination of scientists who is very strong in the domain, collaborating with maybe a more data engineer and data modeler and on the science side.
John: Yeah, fascinating. So Clare, yes please.
Clare: I was just thinking about. So, for example, if you're looking, let's say you're looking on the food chain and, you know, you hear all these you get all these alerts from various governments and the massive alerts, the FDA they're all saying this is single suit and it's not within compliance. And people will take that data and then within a company and they'll use it to inform themselves as to what should they be worried by, so why should they be buying that ingredients. Then you're looking at that data and just curity that format. You're only seeing the picture. So if you bring in somebody who has an exposure science or a toxicologist or a nutritionist, then they both sort of look at what these large survive and they say, okay, well, which ones actually matter, which one really need to be worried about? Because some of them might be a problem in the sense that they breach a legal threshold but they might not actually be a problem from inspection of the safety threshold, because that depends on how much more expense to. You can look at that and say, okay, well, with this level of contaminants and seeing as we eat this much of this food, we need to be careful of this. Then you can no longer, yeah, okay, fine so there's this much of that. You know, we actually don't eat that very often, very much of it. It's about getting enough into the system. So when you have this was like months of multiple eyes looking when the publications then that thing about what signals actually matter. What do you really pay attention to? Because if you're going to taking risk, you have to be able to prioritize and make informed decisions because every decision comes with a cost and it comes as a benefit. And if you're in a company and you're trying to buy food ingredients and you're dealing with some very narrow margins. You are to be making very informed decisions.
John: Right. And it is risk management at the heart of everything you do that? Or all your models really about predictive, I mean we've talk about personalized nutrition. That seems like a different category.
Cronan: We started working on nutrition and we realized, you know, we had all this food intake information around looking at food safety exposures and chemical exposures. But actually the flipside of that as well, what nutrients are people getting? What are benefits? And there's always a risk benefit calculation that needs to be done in order to make policy changes. So we did a lot of scenario analysis around things like sodium intake and then saturated fat, calories. All these health issues, in fact, in terms of the health risk from things like chemicals and additives in your food, I'm firmly of the view that nutritional problems and diets are causing far more health impact, far greater health impacts around the world and malnutrition, it's over nutrition in most cases and the health is optimizes the count of that. And governments are trying to put pressure, I suppose, on industry to reduce calories, reduce sugar, reduce sodium in products. And in order to have a strong evidence base for that, we're able to complete different scenarios and look at the current scenario. What are people getting sodium from? What are getting the most sugar from? When you do that calculation, you know, you might be a little bit surprised that lots of the things that you might have thought were contributing or actually contributed the most. So things like white bread came up in Ireland. Because of the quantities of people lead, you know, there was a saying it's a factor of what's consumed as well as what's in the products. And that's where we look for that. And then government can say, well, what if we requested a 10% reduction in white bread? We can say, fine, this is what the impact will be for your average consumer, for you're high consumer, for children, for adults and immediately compute dash. And they can therefore go with that policy, with the knowledge that this is actually going to make a difference. You know, and look for reductions in other categories of products. And that's the kind of work we were doing a lot in nutrition. Of course, with all these data sets on consumers habits and practices, we started to learn a lot about nutrients and that consumption. We've got personalized nutrition, project hunger and EU funding at EU Framework five. If the project does call food for me and we were using algorithms in this project, actually allowed people to submit information on their diets. And then we also took a blood sample and swab a genetic swab from all of the participants in the study. And then those participants were split into four categories and they got different levels of advice. And this was all made it through a system so personalized that were getting personalized nutrition advice to improve the diet in order to try and improve your health. So it was the first category was just based on basic dietary information or eat five vegetables a day. You know, you need all of that standard stuff. Not taking into account anything of the data that we've given. Second category was just based on their dietary information that they submitted, along with some information on their phenotype and actually of their age and gender. Third category then included measurements of their phenotype, like body mass index, waist, measurements of blood sugars. On the fourth category included all of that information plus information on the genotype. So we're trying to give them more and more specific personalized nutrition advice. So we learned an awful lot about personalized nutrition to that project and ultimately find a measure that they comply with advice and improve the diet. So we did this over one and a half year period, measured at four times during that period. Each time their diets and their adoption of the advice. That's all well and good. People didn't improve their diets and got healthier diets, but actually having more and more specific personalized advice going to the level that might suit their particular genotype didn't actually make much difference in their adoption of the advice as it was just phenotype and information on their actual diets. And I think that the really nice overlap between what you're doing now and all of that health related advice, you know, as you said earlier, we're trying to make people healthier choices push. Could we combine that with things that people tend to like their preferences? We never included that information in this study. And if we could have done that information in the study and, tailor their diet dietary advice even further for that, tailoring for preference, some sensory type information that you would have in your inventory, perhaps you could have an even greater health benefit to those consumers. By doing that type of them analysis.
John: Yeah, that's very fascinating. I love to stand that actually. Clare, is there anything you want to add before I say anything else?
Clare: Yeah, I mean, that's what I find fascinating is when you ask people what they like and what they don't like. And then you actually measure what they go and do and oftentimes, it's completely different. So here in the company, you're trying to develop a products and you would like people to eat it. You know, sometimes you might know that this is what I would actually like or enjoy. But then question is, when faced with all the other choices in the marketplace, is that the one that we then go and buy? And so it's marrying those two kinds of data. So you actually understand what really is driving that decision making process for the consumer as opposed to what we think might be driving that decision. And that to me, would be very interesting for multiple reasons, partly because how do you incentivize people to change habits? So if, you know, for example, that people like certain smells or sense, where they like certain taxes and flavors, you know, there's some people who like salty snacks. Some people like sugary snacks and some people like certain smells. How do you use that knowledge then to encourage them towards one thing, away from another thing? How do those signals tell you, well, that's not what we do. That's not going to do this and, you know, I mean, something smells lovely and people want to eat it. Some people want to sniff it. I mean, sniffing glue, but apparently it smells nice. So, you know, some signals that we get on action. So to me, this is very interesting overlap between what people are telling us, which you do a great job sometimes it's so diffuse, it's really hard to actually find it then. If you can get that data against and this is what we know, what actually happens, then you can start delving into where there is disconnect, and understand, find to what you're trying to create. Will it be a diety policy, will it be a new food diet. Will it be an understanding of an association between a food and a health actually.
John: Yeah. That's fascinating. I mean one of the things I would say is I think that good marketing is education. That when you actually know which people at a detailed level are likely to appreciate certain aspects of your product, then your marketing can be targeted to reach those people and educate them about the fact that your product has some property that they're likely to appreciate, right? But it is true that a lot of time, even though people would like something if they tried it, even just getting that initial trial is a challenge and you know there is so many, a habit is such a driver of everything, right? And actually, I think one impact of this pandemic is that you're gonna find people are being forced to try things they wouldn't normally try because they can't get the foods that they normally like, right? And, you know, I can tell you, we have now several new brands that we're going to be buying after the pandemic that we would never have tried. But we actually really like. It turns out, yeah, I mean, soda and tonic water with ginger is now my thing. And I didn't ever want that. I just couldn't get sparkling water. But it turns that I really like that. We have a new brand of ketchup. Tonic water supposed to help with malaria or whatever. Yeah, maybe so it doesn't hurt. But the thing I would say about like, Clare, kind of response to what you're saying is that education is, of course, a big piece of that. And I think the work that we all do to try to understand it, more detailed level like what's the interaction between individual differences? People are different from each other. And right now, marketing, like I would say, where you see this demographic analysis where people say, oh, look what, men prefer this product to women or something. Let's focus on men. It's just so crude. It's like such a blunt instrument that we're actually there's like such more detailed information we could be providing that could be going into our marketing. So that's one thing. The other thing I'm actually just a follow up, we run out of time, I'm very interested in the question as far as personalized nutrition goes as to, is there any data on whether like a personalized nutrition, is it true that based on genotype, that some foods are more nutritious to some people than they are to other people. And does that is that actually what turns out to be the case? That there are nutritional programs that are better suited to some people than other people?
Cronan: I think there's certainly a genotypes stock that are more susceptible to illness from, say, a sodium and others who are not susceptible at all. So certain people have to watch their sodium intake, another people can watch they like and they don't have any, you know, hypertension or other health issues from that so I suppose the corollary of that, mostly there on this science isn't, you know, there's not a lot of them very strong nutrition genotype link because it's so expensive to study this and all you end up with is knowledge. You can't really, patent that like a pharm company could patent a drug versus the genotype link one person nutrition studies. There's not as clear cut a way to exploit their results. So it's going to be hard to find the investment taking nutrition, health genotype links. Well, there are signs. No, I think it's true to say that there certainly more, your statement is true. There are foods that are more nutritious to people of certain genotypes than others. But we're not sure. We don't know all of them. Or are a lot of them.
Clare: I think one of the things that is interesting, you just you made that comment that it's really unhelpful to say that women like this were rude so this men don't because everything is on a continuum distribution. And when you're in public policy, you're sort of looking at this normal boneshaker, which is more complex. And generally speaking, because public policy focuses on the population. You're really talking about shifting that normal distribution either up or down, which means you have collateral damage. You're going to have collateral damage those people who are actually doing the right thing and they're the one to tale ends. If you look at the kind of modelling, for example a Creme Global does, they look at distributions. They create distributions with their data. So then you can actually do this on the centralized. Who or what it is you really need to focus your attention on. So you can you answer a message. You can target different sexes within that population. And what you're actually really wanting to do is squeeze your normal distribution curve, either one way or the other way or in the middle. You want more people doing certain things. And, you know, without that collateral damage. So it's fundamentally important that we understand the decisions that kind of approach to modeling data and modeling the information that you get out of surveys. Because, we are all people and we are all different and we all made different choices. And so we have to try and flat to the insights that we're creating on these data.
John: That's fascinating. That's why I do these calls is so I can learn things. And that's like something. The idea that you have more nuanced public policy as a function of this better predictive modelling or informed research that's driving...
Clare: More nuanced public development as well.
John: Right. Yeah.
Clare: Or more targeted marketing approaches because you're not trying to spin a square peg into the round home. And if you're looking at this for increasing the value chain, increasing reliance on you are inspections, specialized foods, you know, there are different citals in society. And who you aiming this at and how do you know you're going to be successful. Have you misjudged, so to speak, or have you nuanced to your data analytics says that you better understands to within that distribution. We are really going to go for this.
John: Right. Interesting. Fascinating. Okay, we are actually over time. So this half hour is blend by. But are there are any, let's say I mean let me hear from each of you just for our listeners, you know, I mean, I think it would be good. First off, how can people learn more about Creme Global? Where should they go to read more about your company or follow you on different social media channels and also connect with you individually?
Cronan: Sure. Well, it's very easy, obviously, www.cremeglobal.com same Twitter handle @cremeglobal, from LinkedIn easy to find on Google and if you want to contact myself. My email is just email@example.com. I'm sure you can put the details on your podcast.
John: Yeah, sure. And Clare, how can people reach out to you?
Clare: Similar way but my email is firstname.lastname@example.org. And those eaten eyes are not very important because otherwise I just use my email.
John: So we'll put links. But you both on LinkedIn as well?
Clare: I'm no longer on Twitter, which to me is quite a relief because trying to get anything into that number of characters, as you know, a quite challenge as a scientist. That's another other topic.
John: Twitter is like the worst for me because there's no completion. If you got that feed, there's no end to it. So I am always, like, trying to get to the end. There is no end. So any last comments for our listeners here? I mean, what do you say, so you're coming from I think more of the as you said, kind of hard science. The risk management. Also the nutrition. You know, some of the more objective measurements that sensory consumer scientists would tend to interact with. What kind of thoughts over the next few years, how do you see kind of more collaboration between people like me and people like yourselves in terms of bringing our models together?
Cronan: Well, I might go first, so I think there is a nice fruitful intersection there that is drawing it out on a little Venn diagram here. We're looking at the consumer from an exposure. Health benefits, market sales point of view and you know a lot more all about their preferences or insights, perception and uncertainly for a company in developing new products or a government's development policies. I think the section of those two pieces of the Venn diagram is where the risk strategy and new products all fit. I think that's a very interesting opportunity. And I think you would be interested in trying to look for opportunities to collaborate on potentials. And potentially some research are something that we could follow up on energy force.
John: Right. And Clare?
Clare: I agree, I think there's a fantastic opportunity now, particularly with the computational tools that we have to join together data sets that traditionally wouldn't ever talk to each other. And to me, I also think that's a really good opportunity there to come together. Again, people with different mindsets and different sciences. And I'm essentially a biological scientist, chemist and mathematist, you know, there are different ways of looking at the same problem, different ways of the same elephant. I think the work that you do on sensory perception is massively important because that drives so much of the decision making that people make. We don't understand that. And don't then integrate it into what people actually do when we're never really going to crack that much. So I think this is a massive opportunity. And it's it's fascinating how people think has always fascinated some.
John: Me, too, I married a psychologist. Psychology is always been around my life.
Clare: It's like economics.
John: Yeah, awesome. Well, it's been great. Thank you both so much for being on the show and I look forward to our continued collaboration into the future.
Cronan: Thanks, John. It's a pleasure.
Clare: Thank you.
John: Thank you.Okay. That's it. Hope you enjoyed this conversation. If you did, please help us grow our audience by telling your friend about AigoraCast and leaving us a positive review on iTunes.
That's it for now. If you'd like to receive email updates from Aigora, including weekly video recaps of our blog activity, click on the button below to join our email list. Thanks for stopping by!