Thierry Fahmy - New Frontiers for Sensory
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!
Thierry is a serial entrepreneur who created XLSTAT in 1993 and who founded Addinsoft, which has been selling XLSTAT since 2000. Along the way, Thierry completed his masters and PhD in statistics and, since 2002, he has worked full time on XLSTAT. Since then Addinsoft has grown to 25 employees operating from both France and Mexico, with 2019 its best year ever and 2020 looking even better.
Transcript (Semi-automated, forgive typos!)
John: It's an honor to have you on the show, Thierry. So thank you for joining us.
Thierry: Thank you very much, John for inviting me.
John: Okay. Wonderful. Now, I think one thing right off the bat that I find very interesting about your background, Thierry is similar to mine in that, my background is in math and so through really actually a lot of connections with my father, I ended up in sensory. But in your case, your background is statistics and so then you had to, you ended up finding the sensory community through sort of a different path because as far as I know, your father was not a sensory scientist. Is that correct?
Thierry: No. He was a PhD in Physics.
John: Oh Physics. Okay.
Thierry: I should have had connection to sensory analysis before because I was in engineering school for my Masters that was dealing a lot with the food industry. But happens that my connection with sensory analysis is due to an exchange by email with Hal Lesley in the early 2000's. And he wants you to have it convenient tool to do his teaching. The trainings he was doing all around the world. And he suggested me to add some methods to XLSTAT and that head our connection to the sensory analysis are very tight sensory industry.
John: That's very interesting. So you had built XLSTAT as sort of an add into Excel that was already doing statistical, additional statistical and I guess analysis or just extra support. And then Hal contacted you? Asking you if you could add sensory specific functionalities.
John: I see. And what were some of the, so it was early 2000's you said?
Thierry: Yeah. I think it that was 2002-2003.
John: Oh, I see. So really, so in the early days you've been in the XLSTAT at this point. Right from the beginning.
Thierry: You know, the initial idea of XLSTAT was to have a very convenient and powerful to do statistical analysis in Microsoft Excel. There was no reason to add specific methods for sensory analysis in the beginning, but the fact that we had added pretty early the principal components analysis. It's appealing to the sensory community. I believe that this is the reason why Hal contacted me. Well, that, of course, all works related to that original linear model. So these, we had the, you know, the basics that the sensory methods were requesting.
John: I see. That is really interesting to me. So do you remember, like early on what were the types of methods? This is general question. Have you seen the requests change over time? Do you find that the different things? You know, what is the kind of path that you've seen in terms of what the sensory community is looking for?
Thierry: I think it's little by little we see the methods mixing more and more different methods that are coming from statistics, mass optimization, computer science. And this is getting bigger and bigger. And I'm very optimistic for the sensory community and the sensory sciences access for the mathematical part, because I think this industry should grow significantly. I would compare it to the Six Sigma community. The industry discovered that they could save a lot of money by having better processes on the production side. Because they can avoid failures and they can improve quality. And this converts, this is obvious to anyone that it converts into money. Well, the sensory analysis, it's more complex because people do realize that they can work better and do more, make it do more, generate more revenue by working better. It's not as obvious. You know, the production facility that face suddenly it's obvious for a manager to know that he has a problem that cost him something that someone who produces some yogurts, not obvious for him that he's not producing the yogurts that are expected on the market. He might be selling some, he might be very profitable, but he it's difficult for him to realize that he could be doing much more if he was really doing products that match demand. And sensory analysis is therefore that I'm really confident that more and more companies and not only the food industry will realize that. Of course, you have already some car makers and the cosmetic industry that has realized that. But there are plenty of areas where sensory analysis is not used and where it should. That would be some revenue gain for the actors in these verticals.
John: Right. That's very interesting because I've started to see. Well, you mentioned car manufacturers. I know that, you know, one way that sensory is coming up in the car manufacturing were it's just the sounds of things. Right? They're interested in you know, the sensory analysis talk to my sounds.
Thierry: Sound and in touch.
John: Right. Yes, exactly. And so that's really interesting because, you know, I guess you have a kind of insider's view of who like which sense. If we assume that it's a kind of a natural experiment that as people grow their sensory department, they eventually buy XLSTAT you get to see by the sorts of companies that are buying, you know, the sensory packages with an XLSTAT who in fact are adding, you know, a sensory group. So what are some trends that you've seen then in terms of the sorts of companies that are buying the sensory capability with the next XLSTAT?
Thierry: It's really a very large majority of food industry today. A little bit cosmetics. A little bit consumer goods. But it should be much more.
John: That's fascinating. Yeah, I agree.
Thierry: I think it would not be easy for me to tell you an industry where it shouldn't be useful, where it would be used. So it's really the food sciences community that grow that sensory analysis to what it is today. But it's important to know.
John: Yeah, that's interesting. I see it converging a little bit.
Thierry: Why not even a bank, you know?
John: A bank? You mean, for example, the ATM ownership, how you say it in French but the automatic teller machine?
Thierry: Yeah. ATM's. Anything that is being used by the customer, credit card.
John: Right. Yeah. I mean, if there's a sensory experience.
Thierry: Design of, how you say the branches where people go.
John: Right. The smell in the lobby. That kind of thing. Yeah. That is really a big point right there. Yeah. I mean I do see it in my own work, you know. Well it's certainly institute of perception. We were starting to get into user experience. You know, that you see the tech community is really interested in UX, right? And that really is the sensory experience of using the app. You know, it's a specific type of sensory experience, right? And of course, there's design aspects to that. But some of it is just, you know, is like the same tools get applied for trying to optimize the layout of, you know, landing page.
Thierry: Yeah. You know the design, the UX product design tends to be very qualitative. And we're bringing the quantitative to that. And it's not competing with the quanlitative approach. It's complementing. And the problem is that, you know, people have tried to avoid using or doing Math's, it's very useful.
John: Yeah. And other front, actually, you know, something you see more and more of is like text analysis where people are trying to bring together text various means of text analysis. Is that's something that you see? I mean, what are the current capabilities? What are your plans for XLSTAT in terms of text analysis?
Thierry: We added a text mining set of tools. I think two years ago. And, I don't have myself a vision for that because I'm not really into this subjects, the text mining tools. One of my colleague is an expert in that but I haven't worked on that, so I cannot truly speak about it. So we made it involve if people want to do more or request, and ask us to add some features. We would be very happy to. It's not the main thing today. Adding more methods for structure.
John: I see. The structure. Right. I would say as a statistician, once you get into text analysis, you'll probably find it fascinating. That if you ever actually start to study it, you'll probably like it a lot because it is amazing.
Thierry: You've been working on this subject?
John: Yes. I've done some work in text analysis and it is amazing how much of what you would think is kind of I want to say a soft subject, but subjective is in fact very quantitative analytic underneath the hood. You know, in terms of trying to figure out which words tend to get used together or, you know, which words are used in the same ways and sentences are basically, you know, in terms of the grammar like cat-dog. They get used in the same sort of way and there's a relationship between cat and dog where any sentence that you say with cat, you could take cat out and put dog in, the sentence would still make sense. Right? So there's a lot of these kinds of analysis that are very interesting. So I predict that when you get into that, you'll be like, oh, wow, this is a neat subject.
Thierry: I think right now I don't need more subjects. I have a pile of books to read.
John: Well, fair enough. Fair enough. So as far as the requests that you're seeing now, what are you seeing now from your clients in terms of, because you're in a great position to spot trends I think, in terms of like what people are wanting. So what do you see in terms of the trends?
Thierry: I think people need some tools that simplify the changing of methodologies, so we are really into that. I think there is also some more to do with ketadata that check all that apply. And so we are working on that area. And I think also some things to mix from different areas where they've been traditionally applied. And they could combine nicely to provide some tools for the sensory analysis. We were adding new tools at a fast pace. Last year, when some methods to cluster data sets not only variables but data tables of variables and some methods for ketadata. Sometimes we are a little bit late on some subjects for unknown reasons, but we added the status multi-table methods last year and we're just about to announce the sorting assessment. It will be airing in January.
John: Okay, so stay tune. So do you find that there is more of an interest now and tools that might be considered coming from data science as opposed to statistics in terms of some of the machine learning methods that might be? I mean, you mentioned cluster analysis, which of course I mean, it's not always clear where the line is between data science and statistics. But do you find that the sort of the types of tools that people want are kind of evolving along with this like rising into data science?
Thierry: We're working in this area also. But not really in the view of adding this for sensory analysis. You know that all these methods are really black boxes. And I think that we are in a sensory analysis is a science where you should still try to understand what's happening and why people tend to prefer this or that. So you need to be able to really interpret the results. And I'm a bit afraid that leaving everything to black boxes would make people understand less and lose some competence.
John: You'd rather have a model where there are some coefficients you can interpret or something like.
Thierry: Yes. And also, when you choose your model, if it's a quadratic or linear, you need to understand why it's linear or quadratic. It must make sense. Otherwise, you have to go back to the field and get some more data.
John: Yeah. Yeah. I mean, that is a big discussion right now for sure. And a similar one that I'd actually like to get your view on is kind of open source software. Like what would you say XLSTAT, what is the advantage of, because obviously there are advantages to proprietary software like XLSTAT. So what would you say would be the advantages of using XLSTAT over R, Python or some of the other tools you might use to perform analysis?
Thierry: They have various certain reasons. Let me try with the one that I prefer myself which is not the nicest one business wise but we always try to make our computations faster and faster and faster and that requires that the code is very well written and we're working on this very seriously. I can tell you that you do not have a single tool, even in open source that is able to compute for a sensory analysis as fast as we do.
John: I see. So it's performance advantage.
Thierry: Exactly. Performance advantage. Sometimes we identify some problems. So recently someone reported an issue on speed with a very specific setting for ANOVA. And we've been working immediately on it. And today, even in this situation, we are able to face our competitors without any issue very often where we are much quicker than R. And so this is one thing, the speed. And what specific in sensory analysis is that you have multiple why's. And it's not very frequent in statistics that you want to run a series of ANOVA's for a series of why's and this specific situations that no other software is holding it. So this is one thing. The second thing is support. If there is a problem or if you have a methodological question, or if you want to change to methods that haven't been changed so far, you contact us and we'll do it. We've been doing that for a very long time. I've seen several packages even you don't have enough to do the R, so many packages for R. Some are maintained by excellent geographers. Some are PDR Clip. Some are with no one behind because the people just beat them for their needs at time and they switched to another subject. And they are not there to answer the question because answering the questions of customers is a job. It's not something that you can do on Saturday when you have some spare time. We have a team to do that.
John: And what you said about changing methods together, that it's almost like you're taking your there's a bit of statistical consulting that's going into every you know, every version or every which I say instance of XLSTAT that's out there in the world contains a little bit of you as a consultant inside it in the sense that you're guiding your user through maybe a series of analysis that makes sense. You're changing things together. Whereas you're really on your own if you're using in R package. I mean, if you're me, it's probably okay. But like, if you're just someone, you know, the world you're trying to figure out, okay, what do I do next, etcetera. You know, the guidance that you all are offering is definitely valuable. So I can see that, as you know, there's support, like you said, you have a problem. But even just in the use of the software, there's a kind of support. Because you've already worked out the work for.
Thierry: You know, it's so easy to do bugs and to change things not properly. We invest a lot on that, too. I didn't know that when I started XLSTAT but producing a software with versions being released on a very regular basis, I would say on average, where we release an update every week. You need to have a real software factory that is completely automated with thousands of tests to make sure you are not regressing and while correcting a bug, you're adding to other ones. So that is very strick process and very organize process. And some software behind that is co-digest to run the robots that will be doing the job.
John: That's fascinating. You may have heard that there was a bug discovered, an important Python library. Yeah. You're gonna love this. There were a 150 scientific papers were invalidated because of this bug. Yeah. So the next time that you're trying to, you could use that if you need to, for marketing purposes. Why you should pay for a software. That's interesting. I mean there are pros and cons for sure. Now what about scripting, you know, is it the case that your software is mainly a kind of single use or how does it how does Excel interact with if someone wants to write a script so that they can reproduce some research later? What is that look like in terms of like the current capabilities of XLSTAT?
Thierry: Today you can do it with DBA. You can change pics with the DBA. Recently a customer sent us a macro that had thousands of DBA instructions to automate the several test on their data. It was related to I don't know that the word in English, but it's like creams, you know, you put on your skin. You are comparing things. The macro's we're running for hours. And if we if we had done it to ourselves, it would be just a few minutes.
John: Oh, very interesting.
Thierry: So we're encouraging customers who want to automate things, you ask us to do it because we are really pros at coding and we can also identify some weaknesses of the code when we see that the code of customers so this is really a service about wanting to do to offer that people are not coming to us very often, so they script and that's good that they can script, because that should be a proof of concept that for industrializing the process. Because XLSTAT that is not developed in DBA. Only the interfaces are using the DBA layer of Microsoft Excel, all the rest is developed in C++.
John: C++. Yeah, which can be much faster.Yeah. Okay that make sense. Well, amazingly we are almost out of time, Thierry. I mean it's just amazing how these calls can fly by. So before we wrap things up, I would be interested to kind of hear your advice for our listeners. I mean, just a general advice for sensory scientist. What would do you as a physician think that people should be focusing more on, what do you see that you know would be areas of emphasis that you would recommend?
Thierry: My advice to anyone who is using statistics, is to try to plan experiments before analyzing data. It's a general problem wherever you go. People would be coming to you. I've got this huge set of data, I want to create some insight from the data. The problem, that is not very often. It's not designed properly. And the first that should always be design of experiment. It's not the part of statistics that I prefer but it's really a mess. Let me add something. The sensory scientist are not the worst because they are very often producing balanced experiments. They know about it. That should be that rule number one, design your experiments.
John: Right. Yes, I agree. I think it was Fisher who said bringing a statistician in after the experiment has been run is like bringing a doctor in after the patient has died. That's what the statitiscian might do is tell you what the experiment died of. Yeah. So that's a good one. Alright. So kind of a conclusion here, Thierry, do you have any more thought you'd like to share before we wrap things up? Any other bits of advice or things that you see that you think are interesting that you want to draw attention to?
John: Okay, so where can people find you? First off, when they want to invest in XLSTAT, where should they go? And then where should they go to find you?
Thierry: To our website, xlstat.com. Maybe they can contact us through our support channel and we will be very please to help them. Exchange with them. Listen to them. Very important and hopefully meet them because now we are more and more trade shows and conferences. I'm very sad, I will not be able to be at sensometrics. But two of my colleagues will be there presenting some new features. I hope I can make it to the next sensory meeting. I love the community there. I really like what people are doing. And I really hope that my vision, that sensory analysis should be deployed as much as a statistical process control and Six Sigma has deployed.
John: Yeah. I know. I think that's great. Thierry, I think you've done a lot to advance the quality of statistics within sensory. So I thank you for that. I think it's been really helped the field. So just to kind of wrap up, if people wanted to contact you personally. Can they find you on LinkedIn? Are you on other social media?
Thierry: Yes. They can find me on LinkedIn. I'm replying, and accepting people.
John: Yeah, the networking is so important. When will you be at EuroSense in Rotterdam?
Thierry: Yeah. Would you be there?
John: Oh, yeah, I'll be there. Great. Okay. Excellent. This has been great. Thank you very much for your insights. And I look forward to seeing you in person in Rotterdam.
Thierry: Thank you very much, John. Have a nice weekend.
John: Thanks. OK. That's it. Hope you enjoyed this conversation. If you did, please help us grow our audience by telling a 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!