• John Ennis

Frank Rossi - Skills for Success


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, and PodCast Republic. Remember to subscribe, and please leave a positive review if you like what you hear!


Frank Rossi is an R&D Director at PepsiCo in Plano TX, where he leads the statistics team supporting PepsiCo’s global snacks business. He is an Accredited Professional Statistician with more than 30 years of industry experience including positions at Kraft Foods, Campbell Soup Company and General Foods. Frank is coauthor of the book Statistics for Food Scientists: Making Sense of the Numbers which was published by Academic Press in 2015.


LinkedIn

PepsiCo



Transcript (Semi-automated, forgive typos!)


John: Welcome to AigoraCast. Conversations with industry experts on how new technologies are impacting sensory and consumer science. Hi, I'm Dr. John Ennis, co-founder of Aigora and host of AigoraCast. This episode, I enjoyed speaking with Frank Rossi, an R&D director at PepsiCo. In this episode, Frank shared some fascinating insights on the skills that are needed now for success on the statistical side of sensory. Hope you enjoy this conversation as much as I did. And remember to subscribe to AigoraCast to hear more conversations like this one in the future.


OK. Welcome, everyone, to another episode of AigoraCast. Today, I'm happy to have someone I've known for a long time, Frank Rossi, who I have been friends with, I see he's been at multiple companies as I've been at multiple companies. So it's been a fun kind of trip that he and I have been on together over the years. Frank is an R&D director at PepsiCo in Plano, Texas, where he leads the statistics team supporting PepsiCo's global snacks business. He is an accredited professional statistician with more than 30 years of industry experience, including positions at Kraft Foods, Campbell's Soup Company and General Foods. Frank is co-author of the book "Statistics for Food Scientists Making Sense of the Numbers," which was published by Academic Press in 2015. So, Frank, thanks a lot for being on the show.


Frank: Well, thanks for having me, John.


John: OK, awesome. So, Frank, before the call, you and I were talking about some of the things that we've just seen over time. And something you mentioned I thought was very interesting is that in your current position, one of the duties that you are, one of the roles you play is that you help in the hiring process. And I think it will be interesting to hear you talk about how you've seen the sorts of skills that it takes to succeed, kind of, you know, because you're at the intersection of the kind of food and statistics space. Like, what skills do you see that are really relevant now in that area that maybe were less relevant, how things change in terms of the skills that people need to be successful?


Frank: Yeah, yeah. And so, you know, if I go back and think about the skill set of people like me when we were entering the industry, you know in the early to mid 1980's, it was more about theoretical statistics and applied areas, your quality and manufacturing things. And so now, though, people tend to have much better coding skills, maybe a little bit less theoretical statistic. But again, these emerging areas of machine learning, artificial intelligence, these are type things that obviously didn't exist thirty five years ago. But now even a student getting a master's degree in statistics is going to start is going to have some level of experience in those even in things like, you know, text mining and things which at one level isn't statistics at all. Right. There's no numbers in that. But it's very just interesting to see how that's very different now, because it's still called a master's degree in statistics, right? And it just it it's fascinating to me what they are learning now relative to say what I learned 35 years ago. And it's not just about the technology.


John: I see. And so what do you. OK. So there's a few ways to go. This one is, I guess, the mindset. So do you see a mindset shift in terms of the people who are coming out of support? As someone who has a master's in statistics and they're joining your group? What are they bringing that's different that a statistician would have brought thirty five years ago?

Frank: Like I mentioned, that there's a little bit less of a theoretical underpinning and more about more of these advanced data analysis skills and also just much much better coding skills. I mean, if I go back to my myself, you know, we learned SAS primarily and I personally first used it on punch cards. So obviously, it says a little bit about my age. But the other thing is, is that, you know, those were that was code not to not to do very fancy things, but basically do a simple linear regression or a multiple regression or something. You had to read in data. You didn't have to do much to it other than to write the code that's going to write the regression and say what a model is and things like that. So it's not the kind of coding that I would think that that that these people now have, which is, you know, there's so much more involved with data cleaning, data management. You know, you're working with very different kinds of data sources. And so that just the need for good coding skills is much more critical than it was say thirty five years.


John: Yeah. I mean, this is something that, you know, of course, like I've had to get much better at coding, so my degrees in math, you know, and then I went into neuroscience and I was programming a lot during my postdoc. But the last five or six years within sensory science, it became apparent to me that was really going to have to learn data science in order to excel, that that's you know, the size of the datasets. I mean, it's just so many things. Like in my case, it was statistics for regulatory engagement. We were doing these big studies to support a client who is interacting with the FDA. And I actually worked all the way through hour for data science by Hadley Wickham. Appreciate.


Frank: I suspected it'll be a classic or considered a classic at some point.


John: I think it is already. Yeah. Yeah. But anyway, that's a sad topic. But there was I worked all year through that book and now it's like I can't even imagine working without those skills. You know, just today I was having to clean a data set that I mean, it just is. Yeah. The whole area has changed. I would say let a lot of my time is simply spent manipulating, you know, all sorts of datasets to get things ready, that the algorithms. Right. When I finally get to running algorithms, it's like, oh, what a relief. Now, where is the easy part.


Frank: Yeah, it really is the hard work. It's always been difficult dealing with data, but it's just very different fields now. You know, in the path, we would have to, well, often we were just entering it by hand. So, you know, I go back to paper and things like that. I mean, having I remember in my first role in the food industry, we had someone whose job was to basically enter data. So often paper ballots and things like that. It could be any type of data, could be someone's lab notebook or something like that. But even then, it would come in a format and we'd have to do some things to kind of work with it. But that's still nothing compared to what's going on today. You know, one other thing that I think is kind of interesting is, you know, if I go back, you know, because almost my entire career since 1984 has been in the food industry. But I did have three years working for a statistical software company when software was going out from was becoming more user friendly. So instead, you know, right at that point where we stopped entering command type things and started using point and click interfaces and it seems like.


John: Which year was this, this early 80's?


Frank: It would have been 87 to 90. And I worked for a company that no longer exists, but we were the first commercial design of experiments software package available. And this was in the days prior to PC's being very very ubiquitous. And the idea then was, you know, people were still having to deal with command line things as opposed to menus and dropdown and things like that. And so there's been this interesting shift in, you know, tools like mini tap jump where you're no longer writing commands and code, but you're actually, you know, doing pulldown things. And now we seem to have gone all the way to the other area where, well, to really get things done, especially in this modern world, you need those coding skills again. We spent a long time trying to keep people from coding, and it seems now that we're actually getting back to. Coding skills are really important. And it's, you know, just this general concepts that are because in a way, it doesn't matter which language are python. A lot of these overlap. A do loop is a do loop. But the idea of how do you manipulate data, clean data and then run various analysis? It really is. It's we're kind of moving back away from point and click. There is always be a need for that. But I think there's still, again, that that sudden emerging need for coding skills is something that I think has really changed in the last five to 10 years.


John: It is really fascinating because, I mean, this is a huge topic, the whole question of coding versus dashboards or user interfaces, this kind of thing. I mean, what I found kind of in my client work when I started Aigora, I thought, okay, I'm going to teach all sensory science just to code. Well, that is not really how it's how it's working. It's actually turning into a lot of dashboards, but they're custom dashboards. So it's kind of at the intersection of a user interface and coding. And I see that as a theme. You know, I'm sure you've heard about the whole Fourth Industrial Revolution concept, right, that you've had like it's kind of ill-defined because it's not clear what's different between the third industrial revolution, which is the computing revolution that happened in the 80's. Now there's something new is happening. And so people are calling it the Fourth Industrial Revolution or Industry 4.0. The question is, what is that new thing? Well, you can name technologies. You can talk about, OK. You know, elastic cloud computing or you could talk about, you know, the machine learning models that are readily available or you can talk about an augmented reality. There's like a lot of new technology that's coming online. But it seems to me that the thing that's most different is in the 80's, people had to adapt the way they were working to computers. Now it's the other way around. We're trying to take trying to get the computers to adapt to the people, right? And I think that's where people like you and I, when we have coding skills, we can build things for the people that we interact with, like to support them, right?


Frank: I think that's a huge part of it, because really, you made the comment about dashboarding. And that really is you know, it's, it's the place where people are gonna now interact with their data, but they are not going to be the ones doing the coding and things to get it there. You know, we have many I've seen many different instances where, you know, data is continuing to update, you know, be it just, you know, monthly, you know, manufacturing data. Right. We we collect data, you know, on various things. And just, you know, the people who want that monthly update, what's our performance look like? They don't need to write the programs to do that. They just need to have a tool in place where they can visualize that data, get various types of summaries. Maybe there's other types of analysis and things. But the idea they are the users of the data, whereas people like you and I are the ones who are enabling them to use their data as opposed to creating summaries for them, creating analysis for them, that's what a large portion of the earlier part of my career was. Giving people those kinds of things. Now they're getting to interact with it. But I see our roles are enabling them to interact better with their data. They know what their data looks like. They know what all those things mean. In a large company like PepsiCo, a lot of different types of applications and I couldn't possibly understand the data, the data sources, as well as those people who live with them day in and day out.


John: Yeah. And that I should be just another topic, which I think is really interesting, which is the importance of subject matter expertise. Which is something I see is becoming even more important. Would you agree with that statement? Frank, do you think that.


Frank: Absolutely. Absolutely. And this is this is why I think throughout my career, the companies that I've worked with, we tend to work in teams anyway. I'm a statistician. I sound like I know more about some of those subject matter thing. Just because I've been in the same industry for most of my career. But I really I don't understand that data. The one which the subject matter experts will. Of course, there's a broad variety of data. Even in just say that the R&D space that they like PepsiCo or Kraft. And so those subject matter experts have to help us to help them because we just we can't possibly know all that. Like I said, I I know enough buzzwords to make it sound like I know what more that I know what what I'm talking about. But I tell people to I said, don't let that fool you. Don't don't assume. And that's why people like myself.


John: Right


Frank: People that are doing similar things on my team. We have to talk to people a lot to get that subject matter activities open up to us so that we can address.


John: And actually, it's interesting because that's where we think about the coding. Like you and I are not software developers, per say. I mean, I think we have some skills that I do do software development, but I don't think of myself primarily as a software developer. Right. Like the old days where the lines were more clear cut, where you might have SAS has their software that they make and you can request features. But at some level, you're not empowered to implement these feature changes if you wanted to. Whereas now, like for me, I work a lot with shiny dashboards. I'm not sure you know what you are using, but I think that's a common tool that's used, you know, to allow us to build maybe not the level of, you know, industrial level software that will be made by a company like SAS. But we can make something that that is informed by the conversation with subject matter expert so that the thing that's created really suits their task, right? So there's this something I think very special about that, where you've got like the power has been shared more or something. I'm sure I have had to really think about this. How how would you say this?


Frank: It's interesting you say that because so we do and I always have to phrase it very gently when I say it. I said one of the things, you know, if I'm talking about. What does your team do?


Frank: I often mentioned, you know, I obviously mentioned things like designing experimentation and building models. And I'll say things about when we provide training. And I also say things like, oh, we do application development. And I have to be very I say that I said that, I'm using that very, very lightly, because what we're doing is creating things like you mentioned, like a shiny interface to do a routine data analysis that's powered underneath with R, but we don't want people to have to download R and take a piece of code and run it in R and get things. We want to make it a little bit easier for them to interface with. Same with things like, you know, I mean depending on what the application is. Again, I'll use that word very gently when I say application. But the idea of, you know, it could be a visual basic front end to an Excel spreadsheet. If you don't need any particularly sophisticated in that terms of calculations, a lot of it is also about where you're deploying it, you know, in a big company like PepsiCo or similarly with Kraft. These are global companies. They you know, you need to pay attention to what the resources are going to be for the various users for these applications. I guess we can't put my little air quotes on applications every time I say it, but, for the purposes of our discussion. Every time I say applications, John, think air quotes, please.


John: OK, we'll put that at the top of the show. Think air quotes, please.


Frank: And so, yeah, I mean, that's that's something that I think is definitely emerged. But it is a big part of what me and my teammates do, because there are routine data analysis that people do. Again, some of them are not even sophisticated enough that you create a dashboard, you know. But all of these kind of fit in the same place, I think. And that'll be bad air quote applications piece.


John: Applications, Jacob's lightweight applications.


Frank: Yes. Lightweight maybe is a good way.


John: Lightweight. I mean, I totally relate to that. It's very special because a lot of times people don't need a full, heavy duty piece of industrial strength software. They just need something that facilitates the use of some model that maybe they would have been able to run otherwise, or they don't have to keep going back to the statistics group all the time. The statistics group just say okay, here you go. And then you guys, you all can spend your time.


Frank: Yeah. Yeah


John: You all can spend your time learning new things and and a building out like what I love about a dashboard is you can keep putting more functionality into it. Right. That if you can continue to grow. Yes. So very exciting.


Frank: And if it's done well, it's not even just for that individual user, but they can then share what they're learning with their data to their peers, their management teams, things like that. And if it's done well, it's not even just for that individual user, but they can then share what they're learning with their data to their peers, their management teams, things like that. You know, these are companies are collecting data for certain reasons and they're trying to get information out of it. And often it's a broader you know, there's levels to it, but there's a broader team that needs to understand the implications.


John: And to that point, are you also supporting automated reporting? Is that something? That's a huge thing that I do. There's so much time there historically, so much time has been spent and sensory making tables and chart.


Frank: Yep


John: And if you can if people that I mean, it's amazing. We are lying.


Frank: Yeah. Yeah. It's interesting when you say that because there's a couple ways you can approach that. And some of it is, you know, obviously the tools in that sensory space. So if you think about train panel data. Right. The software tools like Compusense and their competitors, they all are starting to provide more of that because I guess they sense that's a way that they can kind of differentiate themselves and add value to the to their user group. And so there's that level that we don't really need to get involved with because it gets, you know, the software that's that's being used. They're introducing more and more of that. I think there's always going to be things that are very, very specific to companies, the way in which they operate their individual needs, that no generic software package is going to be able to provide. And that's where you end up with, OK. You know, our management team needs it to look like this because that's what they're going to understand to get that well.


John: Right. Yeah. Yeah and that's great. I mean, there's the officer package in R. Which I'm a big fan of, which allows you to make fully formatted PowerPoint doc. Yes. Slides or word documents, you know. And then there's that you can also get Excel documents through Open XLS. There's really nice tools for that. So I think it's really promising. Now, I do want to ask you on the flip side, Frank. Like, it is great that all there has been all this progress. Do you feel that something has been lost, though, with these new, you know, new graduates where OK, maybe they're good at that. They have a broader set of of tools maybe or or skills than they would have had 35 years. But do you think that I mean, sometimes I feel like the scientific side of statistics is getting lost in the shuffle a little bit.


Frank: And that's what I had mentioned earlier with this, that, you know, that there seems to be less of a theoretical understanding of in some instances, it's it's pretty basic types of statistical concepts, you know, designing experimentation, which is, you know, I've spent my entire career not just in this industry, but in research and development. And so, you know, designing experimentation, you know, when you're making new products or working with new innovation or new new equipment, you don't you know, big data doesn't exist. And you need to understand the basics of designing efficient experimentation. And I've you know, we I get interns. You know, I've I've interviewed for intern positions. And I'm surprised sometimes at what some of those what used to be very basic statistical concepts they're not necessarily exposed to before they're getting into some of the more fancy. I don't want to quite say trendy, but I'll just say modern data analysis techniques and approaches.


John: Right. Yeah.


Frank: And I think that's a shame. I think they're both important in it. I think it probably very much depends on the specific program. And what their focus is, the specific, you know, university statistics department and what their focus might be. Some of them might stress more theory than others. And I would hope they're you know, we don't want to lose that aspect of it, too, because not everything is going to be about, you know, more complex and fancy things. There's a lot to be said about some very basic statistical concepts and how they're applied.


John: Right. A well-designed experiment is going to give you a lot more information.


Frank: Yeah. Right.


John: Yeah. You know, I mean, you'll be able to see what's going on much more clearly.


Frank: Yeah


John: And that sometimes that is, you know, the most effective way to speed things up or to reduce costs or whatever is through. I mean, I think we should always look to good science first and then support it.


Frank: And and I think that people need to understand that throwing a very fancy modern data analysis technique at poorly collected data isn't going to solve the problem. You know, that it just it's just it's not going to come out in the watch some way.


John: Yes.


Frank: That that's really there's there's just some very basic things that and that's that really gets down to even just cleaning data. Right. And I mean, if you don't do that, well, then that is going to have implications further on down the line. And you're you know, any data analysis technique is still going to be somewhat hampered by bad data. And just that type of skill is still on. It's always been important and it will continue to be important.


John: Right. Right. It's like that Woody Allen movie where, you know, I guess Woody Allen is in a restaurant he overheard these two women. Do you know this where the two women are eating? And one man says, oh, this food is terrible. And the other was but at least there's a lot of it. Oh, it's dead. It's terrible.


Frank: But I don't I don't think I heard that. But I could see that in a Woody Allen movie for sure. And that really captures the sentiment right there. Yeah.


John: These days are terrible. But there's a lot of it.


Frank: I have to see if I could find that video clip and inserted into some presentations that I give and probably get into some kind of trouble with the studio or something like that. But maybe maybe they'll they'll never know.


John: Exactly. So, yeah, I mean, that's the thing with my fear of biased data, then it doesn't matter how much of it you have.


Frank: Yeah. Right.


John: Yeah. Yeah.


Frank: It's funny because when so you had mentioned this statistics for Food Scientists book, and that's actually based on a class that Victor Murchú who worked with me, a craft and I still deliver at Rutgers University through their office a continuing professional education once a year. And one of the topics is around sampling. And we kind of ended up with something to give people, something to think about. And I said, well, you know, sampling's about the Marines, we're looking for a few good men. It's about it's about a few good. Well, you know, well chosen data points as opposed to many that have potential bias is I would rather have fewer data that's of good quality and that little, yeah, I think that phrase has become, it's probably dating me because I don't think that


John: We need to have a few good people. That's it.


Frank: Well, there's that. And also the fact is, I don't think there's used to be in commercials and stuff, probably even going back to my youth. But now probably people are quite following along with that. So maybe we have to bring all that up to remember. Yeah. Maybe we'll have to bring back that clip, too, and see how many times I can get in trouble with various studios and things like that when I start using people's clips.


John: Yeah, well, fair enough. Yeah. So I think what we actually are amazingly almost at the end of our call here. So let's talk a little bit just about what do you think is most needed in the next few years? Because I do think there has to be a return to a focus on the scientific method. Doesn't mean I definitely. That's gotten lost. Yes. It's great that we all can have basic data science skills. I think a data scientific workflow is really important. Reproducible research you shouldn't have two years emphasize completed. It shouldn't be like a murder mystery to try to figure out what happened. It should all be well-documented. Right. It should have to reproduce everything that you did. So there's been a big progress there. And of course we want to use machine learning. But what, what would you say are the things that let's say this, the statisticians who are supporting sensory science should be focusing on over the next couple of years in order to do their jobs even more effectively?


Frank: So there's been a big progress there. And of course we want to use machine learning. But what, what would you say are the things that let's say this, the statisticians who are supporting sensory science should be focusing on over the next couple of years in order to do their jobs even more effectively? Are we just trying to do something interesting and fancy or are we trying to solve a business question? And there is this temptation.


John: Not even a research question, but.


Frank: Very much so, a business question. I missed those. That's. That's where the space that I operate in. And so just because you can do something that start to say, well, should you? Because if it's not adding value in some place, then it might be just cool to try out. And but yes, then we have to make decisions about where to put those resources, because the one thing that I think there's a potential for in this space is, oh, we you know, we're gonna hire all these people that have these skills. They're gonna have these great expectations of what's going to come out of that. And without really thinking about, well, what do we really need and what are the business problems we're trying to solve? And then let's get the right people involved with that, because if you just throw people at data, even even, you know, even assuming all the datas of good quality and that if it's not really focused on a specific business objective, eventually people are going to say, yeah, I don't really think I need that skill anymore or it didn't pan out. That didn't give us anything ultimately helpful and drive in success.


John: That's related, of course, the design of experiments piece, right, because


Frank: Yes


John: This is very similar concept that to know what you're trying to accomplish and then get this, you know, pursue your research in a focused way.


Frank: Almost the first question that I ask people when I interact with them in a you know, on these teams is about what is their business objective. They have to be able to state that pretty clearly because otherwise I'm not sure how much I can help. And I think that doesn't change if, you know, we're looking at the intersection of data science and sensory and consumer research.


John: That saying, OK, Frank, what's great? So we're out of time, so we're just going to have to stop. But I really appreciate you being on the show. Where can people find you or be the best place someone wants to reach out to you?


Frank: Certainly my LinkedIn profile is not just up to date, but should be easily accessible. And within that, there's also the specific like email contact me too. But of course, people can email me through LinkedIn, too.


John: OK, this sounds great. So we'll put that information in the show notes when we publish this. Frank, thank you so much. This has been great. And I look forward to talking to you more in future.


Frank: All right. Thanks, John. It's been a fun conversation.


John: Okay, great. Thanks a lot.


Frank: Thanks


John: 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 reviews on iTunes. And if you'd like to learn more about Aigora, please visit us at www.aigora.com. Thanks!


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!



©2020 Aigora