 Welcome to the panel meeting at this GFSI meeting, and it's about the promise and perils of predictive analytics. So first of all, I will introduce myself. My name is Chris Elliott. I'm Professor of Food Safety at Queen's University Belfast, and I established the Institute for Global Food Security at my university quite a number of years ago, with the idea of looking at some of the grand challenges in terms of our global food supply system, that unbelievably complex system that we try to feed nearly 8 billion people on this planet every day of the week. And one of those grand challenges is about what I call integrity, the integrity of our food system. And a lot of what I've done for quite a large part of my career are look at the challenges to the integrity of that system. Some of them can be around accidental contamination of our food system. Some of it is around the deliberate contamination very much about fraud. And when I think about what I've done over the last number of decades, it's generally been very, very reactive, reacting to a major problem, a crisis that has happened. I spent a lot of time, for instance, investigating the European horse meat scandal, which was very much fraud-orientated. A lot of time over the past couple of years looking at massive issues about fraud in the urban spice industry across the world, but also looking at major food contamination incidents that have poisoned many hundreds of people and killed people. And really, over the last few years, I've thought about how can we get away from being reactive to proactive? And proactivity is about stopping these things happening before they emerge as major, major issues. This is where we get into the whole era about data, capturing and analyzing data that can be used in terms of predictive analytics. And what myself and a group of people have done is we are developing, we call it the interest group in predictive analytics for food integrity. And at the end of this presentation, I'll mention that again. It's really the call that the more people, the more companies that join this coalition looking at predictive analytic models, the more likely we are to deal with serious issues before they emerge and become crisis. Now, I describe predictive analytics as looking at a crystal ball, but I have a very special crystal ball. I call it my digital crystal ball, collecting data together that can be used in a way that we can start to really think about using all the wonderful data analytic tools, machine learning and AI to really produce accurate, reliable, predictive models. Now, when we think about predictive analytics solutions, I have been looking for quite a long time and I tell you there's a lot of them out there. My feeling is there's actually just a lot of these predictive analytics solutions are looking for what is the right problem to solve. And what I'm going to do is during this session, we're going to talk with, I think, a wonderful panel of people that we have put together from different backgrounds, from digital companies, from food companies, from regulators to really think about the future of predictive analytics. So I'm very pleased to introduce our panel. The first is Carletta Utton, who is VP for Product Assurance Risk and Security at Amazon. And so it's a pleasure to have you amongst the panel, Carletta. We have Zoltan Saipas, who is VP for Quality, Safety and Environment at the Coca-Cola Company. Again, hello, Zoltan. We also have Donald Prater, who is Associate Commissioner for Imported Food Safety at the US FDA. And what a big and important job that is. So really looking forward to hear from you, Donald. And also we have Nikos Manacelis, who is the founder and CEO of Agrinol, which is a predictive analytics company. So maybe Nikos, over to you, maybe just to say a few words before we start the cross-examination of the panel. Thank you, Chris. And welcome, everyone. We have the inevitable slide problems. They come up in all the live sessions. But we are managing it. We couldn't predict that. But we can predict many of the other types of issues, of the essential in the shoes when I get in the shoes of people that do the job that you do, you guys. And I'm always extremely frustrated, extremely frustrated every time that the recall takes place. And my team tells me that, oh, you know something? We have already predicted that. Because this means that we could have warned people, could have let people know that something was coming up. And we didn't manage to do this. But it is important because this technology is there working. And it's one of the things that we will hear from our speakers today, that these are tools that we have in our hands. These are tools that the innovators and the visionaries and the people that are ahead in adoption are using them to take this here. But we still need to keep in mind the rest of us. I will use an example. I love this example. This is my iPhone. And it's a quite old model. Why is it an old model? For me, the business critical tool, I rely on this device every day for very, very important activity. My calendar is there. My priorities are here. My tasks are here. But I always buy the previous model, not the one that is announced by the manufacturers. In terms of adoption, it would call me a conservative smartphone user. So if you look at the way that technology adoption works, you have the innovators, the bright and brave, therefore, innovators and visionaries. Then you have the cosmic between. And then you have all the people that are more pragmatic, more conservative, and sometimes more skeptical. I would like to involve everyone in the conversation. And this is what I want to highlight as an important issue for our panelists. How do we get everyone in this dialogue? That's all from my side. Many thanks for that. That sets the scene very nicely. I've got a number of questions I would like to pose to our panel. Carletta, the first question comes to you and then the rest of the panel. But what you now have is my digital crystal ball. It's AI part. And what three questions every morning are you going to ask the crystal ball? What would you really want to know to really help your business in terms of driving a better food safety culture? Great question. Thanks, Chris. So when you think about predicting food safety risks, three things come to mind for me. And those three things are sort of the basis of the questions that I would ask myself. In the morning, I'm getting up and I'm thinking about food safety risks. They're all focused on attempting to ensure the safety of our products that we're offering to customers, of course. The first thing and super important is working to ensure that the predictive models we use, because we have a number of predictive models, are actually accurate. Super important, right? We know that with predictive modeling, we're able to calculate most risk predictions before first units are shipped. It allows us to remove potentially harmful products before customers can get them, which gives us confidence in the billions of items that we actually have in our catalog. The second thing, though, is really about the speed. And so this question or this notion around how do we take the action proactively before it negatively impacts customers? And are we doing it fast enough? And I think that's really important, right? Speed matters when it comes to food safety. The third thing is actually a question around, are we getting smarter? Are we continuously retraining our models in order to maintain the high performance that we see, but also to improve the high performance and further drive accuracy for customers? So those are kind of the three things that I think about or attempt to think about every morning, Chris. I mean, that's absolutely super, Carletta, because it's about accuracy, speed, and it should get better and better and better phenomenal. Now, can I ask you, how far ahead would you like to see? Because I always say, if you listen to the weather forecast, they'll get it right today, tomorrow, in a week's time, not so much six months, forget about it. What do you think is realistic? You know, that's a great question. And it's something I've debated with our machine learning scientist, our technologist, and our food safety specialist. Honestly, I don't know that answer. I would like to think we can be weeks in front of an issue that we could even be seasons in front of an issue, right? But it's such a novel space. I know that I've successfully found a trend on an item before a product was subsequently recalled weeks in advance. And so I know it's capable, it's doing it consistently. And then as you do it consistently, pulling that timeframe further and further ahead, that's the best benefit to customers. Yeah. I mean, thanks. That's really good. And Zoltan, the same question to you. You've now got the second of the digital crystal balls. What are your three big asks of it? Thank you so much. And a huge thank you for having me on the panel. A very insightful discussion. By the way, I do love the digital critical expression. And I definitely will scale that in our business. It's very meaningful. I would have the following three questions. Is this a risk for my business? If yes, why is it a risk? Number two would be, has this risk been addressed externally or within my supply chain? And if yes, how has this been addressed? And number three would be, what is the impact if this risk has not been addressed today, next week and agreeing with Carletta in some instances a season ahead? Because every time we have a trade quality incident besides consumer safety, my number one question to myself, to my teams, what could have we done better to prevent it? To Nico's point. Thank you. Yeah. I mean, and thanks because I think you bring out some more really important attributes. It's the severity of the risk. And a lot of what I do is we look at severity of risks in terms of food safety. And we always say, you know, the first is it what's going to be the impact to people? Is it going to make people ill? Is it going to make people die? But then there's all of the additional business risks as well. So how do you start to categorize those in your mind? You know, as such a large company as Coca-Cola is with so many different brands? So at Coca-Cola with our total beverage portfolio, beverages for life, concept and strategy, what we are bringing to life, which automatically goes much more beyond to the sparkling soft drinks and our landscape, our beverage landscape has rapidly evolved to very, very sensitive, microbiologically sensitive categories. We do ask the question we learned to step up our food safety thinking and categorization that way when it comes to daily product, plant-based daily product or enhanced voters, what are the key food safety risks which we need to consider and ask at first hand to protect our consumers and the trade with our brands? So this liquid food next level, next stage food safety culture has been an exciting and a very insightful journey and is a journey for Coca-Cola to ask the right questions definitely when it comes to microbiological, chemical, physical safety of our product and that would be our drivers to categorize the question. Fantastic. So I'm going to pass on to Donald now. You've got the kind of graveyard shift on this question Donald. So you know if some of the things that you've already heard you know resonate with you is fine or is there anything additional that your crystal ball would like to throw up? Yeah thanks Chris and let me just say first of all hello and it's nice to see everyone just looking through the participant list I see a lot of familiar faces so it's nice to see your names and know that you guys are doing well. With respect to this question I think you know one of the things from from our perspective at FDA is that you know we have a significant responsibility overseeing 80% of the US food supply. To do that we have a number of highly trained staff and laboratory and other resources including increasingly data scientists but yet our resources are finite and so we need to allocate them in a way that's the most protective of public health. So if I had this crystal ball ball some of the questions that come to mind are which facilities have increased risk that would impact their ability to produce safe food? Which lots or consignments of food have the greatest risk for having a food safety problem and maybe also where are new hazards emerging or where are old hazards emerging in new places? So those are some of the questions that I would ask that I do ask every day and so very pleased to be on this panel and to hear the perspectives of other parts of the food supply chain and look forward to our discussion today thank you. Many thanks for that Donald and I'm going to stick with you now if it's okay just to go on to the next question and you know unfortunately the digital crystal ball for various reasons isn't working okay and it's not working because you don't have the right data sources so if if you had access to lots of you know open source close source data what would be top of your list in terms of trying to get that additional information to feed the model to make it much more accurate? Yeah thanks Chris it's another good question and of course predictive analytics as we all know is based on data and information. We have an increasing amount of data and information coming our way. I think you know if I look back over the last 10 years at some point that amount and volume of data and information has actually been a challenge. I'm trying to sort through that information to figure out what's the most relevant to be able to use in terms of making predictions and doing predictive analytics. We've been doing predictive analytics for some time but the amount of data is a challenge however one of the things that I note now is with some of these new technologies it can actually turn that challenge into an opportunity with some of the new technologies it may actually the more data the better and so currently at FDA we are using some predictive analytics and I'll talk a little bit more later about some of the new things that we're doing under our new era for smarter food safety but currently we utilize data that are in our own databases so things like compliance history results of examination sampling testing of products information about the food so priority commodity pathogen combinations information that we have about manufacturing processes and some of the risks there we also take into account frequency of outbreaks and occurrence of illness and then and as was mentioned severity of illness as well as important so we have a lot of structured data that we are currently using and have available to us but of course we're interested in acquiring new sources of information including maybe unstructured data that could present lots of opportunities to to improve in the area of predictive analytics thanks thank you and you know I think that's perfect to feed on to Carlisle now because let's face it Amazon isn't going to be short of data isn't going to be short of data sources but maybe some of those data sources that Donald talked about are exactly the sorts of things that you're mining already be it the gray literature be it social media so you know what is it that you want to bring to to make your models more robust our letter yeah so um yeah a little spoof maybe on that question um so you know Amazon does have data you're right Chris um but you know we always want more data um and I would always want more data sources right the more data I have the more robust the sources are for data that I get the more accurate I'm going to be with my predictions but let me tell you just for a quick second what we do and that'll help explain I think the the latter part of my answer is right now we use customer feedback as a key driver to alerting us to issues and we have models that are basically monitoring and analyzing continuously more than 67 million pieces of customer feedback every week and then we can actually take that information in as many as 70 different languages which I think is also very powerful the interesting thing is we of course are going to take the appropriate actions based on that from a reactive feedback standpoint that the critical thing or the interesting thing I think is that we use the data to help to train our predictive models so they get even better we use machine learning to calculate the relative distance between products that we sell and any products that we've ever received a safety related concern right and so where there's a positive correlation we're able to then predict the severity of a potential issue and the likelihood of a similar occurrence what's very interesting here is we treat a prediction from the data as an actual action that something has happened we don't treat them any differently than we would a signal that was embedded in an actual piece of feedback that we got and I think that's the power of the data right and so you know we currently share the information but I think there's an extraordinary opportunity to work further right we want to keep customer data private and confidential but how can we partner with regulators in the industry to find in and use better sources of data I'm open to it I think it's important and I know it'll only help us get better thank you super so I'll tell I'm going to put the question slightly more differently to you so I'm going to talk about the data onion now okay the vegetable and what you have is an onion has got layer after layer after layer and that's the way the data comes in you know I get layers and layers of data come in and we scratch our head to think about how do you actually combine that and come up with really easy to understand dashboard so there's a big challenge and how is Coca-Cola dealing with with the onion problem first of all please allow me to make a side note to to Carletta and Amazon I need to discuss this with her because Amazon does have data and they do know before I myself do know what kind of sports equipment will I buy and when and they arrive at Coca-Cola we have accelerated a journey the so called future state quality and food safety culture and the future state quality assurance under industry 4.0 our challenge and I do know that if I ask most of my colleagues sitting around this virtual meeting room around the globe say our supply chain is complex but we do operate in a franchise business model which makes it even more complex because as a brand owner we would like to see the supplier information the audit status the single source lot to lot information the adherence to specifications the lot testing results the supply disruptions the complaint pattern of ingredients semi-finished goods in across all our bottling partners before a trade quality incident occurs and that is that is still a key key challenge for us how to to compile the data into a meaningful prediction but prediction is only one part of the decision at Coca-Cola we kind of put this under the industry 4.0 in the food safety culture excellence where the next level predictive analytics including the manufacturing end-to-end supply chain distribution marketplace data covering GNP assessment all the complaint patterns reoccurring quality issues across the system are structured into a decision-making process but the name of the game is to bring more and more transparency and that's what we are working on so complaint patterns ingredients going back to the agriculture suppliers to our bottling partner and synthesizing it into a predictive quality assurance data analytics under 4.0 industry 4.0 that's that's our journey yeah thanks and I'm going to stick with you Zaltan for the next question because I think there's been a wonderful discussion about what we really want you know we want it accurate we want it fast we want it to get smarter we want it to generate a risk register in terms of those things that are absolute showstoppers and so forth so really how close are we to that now something that can predict the risk tomorrow a week a month or as Carletta said is in future seasons thank you this is a very impactful and a great food for thought question I personally believe in talking to my colleagues in the industry and within the the coke system that the impact in the near future and I'm talking about three to five years will continue to accelerate in in five key areas in the prediction the decision making as I mentioned earlier and prediction is just a part of it in the applied tools and and we do hope and we see that the cost of prediction is is going down and the the strategy and overall the society where food safety plays a pivotal role I believe we are at the same time close and very far it is definitely a game changer my biggest concern is that maybe the gap in with the mature organization and and the less mature organization may may increase who is applying and how the AI generate data and prediction and my advice would be to us to our industry that let's not leave anyone behind thank you many thanks and Donald the same the same question to you because you know he said already you're starting to use predictive analytics but to get to answer some of those really key questions because you know there's there's issues about food safety and there's issues about fraud and from my experience the data that you collect there there it's like a Venn diagram there are two big circles sometimes they overlap but how far away you know in terms of of your agency do you think you are from having that model that will pick up the big challenges in terms of safety and fraud yeah thanks Chris this is a good question and you know and in certain respect you know I would say we were already doing some of this it was 10 years ago that we developed our predict system for import screening predictive risk based evaluation for dynamic import compliance targeting that's the acronym it's risk-based analytics but what what's different now is that we're really on the cusp of having the ability to utilize more data sources different data sources I mentioned structured and unstructured data I mentioned new technologies I think at the opening you mentioned some of these solutions are still in search of a problem and I think that's a good observation I mean we really want to define the problem well these are complex public health challenges for the agency and we're not chasing any particular technology we'll continue to main you know remain platform agnostic but but there are these new technologies that really flip the the the challenge of having enormous amounts of data into one of opportunity so I think we're going to continue to get better faster more accurate how are we going to do that well we've outlined some of those things in our new era of smarter food safety indeed we have a whole core element that talks about smarter tools and approaches for prevention within that core element we have a section on strengthening predictive analytic capabilities and it all comes back to data how do we use data how do we share data and so some of the key things that we see driving this in the future I think Carlotta mentioned you know how do we share this data between public and private sector how do we maintain transparency but at the same time confidentiality and so those are the challenges that we will have to be able to unlock some of these the real potential of these new technologies and so we're very much looking forward to doing that I mentioned that when we launched the new era of smarter food safety in April 2019 and published our blueprint last year in july one of the projects that we actually initiated was an artificial intelligence and machine learning pilot for imported seafood and so through this pilot we're gaining some firsthand experience in terms of how we can use our own databases how we can integrate databases and how we can potentially look for ways to incorporate more data from external data sources from our regulatory counterparts from our our private sector colleagues and so those are the discussions that we will need to have and and I recognize that this platform this venue and GFSI is really is a good venue for having some of these discussions so really appreciate that today over don't thank you very much and you know unfortunately we are starting to run out of time now but be thankful don Carletta Zoltan be thankful we're sitting in front of a computer screen and not in a room because I wouldn't let you leave I would want this conversation to keep going because it has been absolutely phenomenal to hear about your different concepts and so forth is is really wonderful as you know I really very much appreciate it so just to finish off now I hope those people and there's been you know several hundred people have been listening in here and I hope you've enjoyed the conversation I hope it's stimulated you to really think about this whole idea about predictive analytics tomorrow actually don't we're going to have a session on public private data sharing and which is something that I'm unbelievably passionate about so you know that's a little advert for tomorrow but I'm just going to finish off now before before these nice people actually switch my computer off for me okay because I know you can do it remotely the call for action is why not join this growing coalition of people who really are interested in applying predictive analytics to your business to your various government agencies there is a website www.foodprediction.org please register free of charge there's there's no cost associated with this this is just joining a club a club who are working together on this whole concept of predictive analytics to bring about a safer world food supply system based on the principles of integrity so what I want to say is Donald, Carletta, Zoltan, Nikos thank you very much for an absolutely fantastic discussions today I mean I found it unbelievably interesting very helpful so thank you all very much and I hope you enjoy the rest of GFSI and I hope you join tomorrow's session because I'm chairing that as well thank you very much everybody thanks everyone thanks everyone thank you thank you sorry Carletta I couldn't help myself no worries no worries my apologies too I think I got into your time sorry no no no no those were good questions and I'm gonna look up and figure out what Zoltan needs next so he can buy more things on Amazon and all of that now I think I think Donald that was like that was a good strong end right I think that was that was a good way to kind of bring it all together how do we work together and you know regardless of who we are or what we do right absolutely very insightful and Chris well done Nikos thank you for guiding us a job well done I tell you it's a seriously good discussion really extremely good so I mean I hope you join that food prediction group I mean Nikos and I've been working on it for the best part of the year now and there's really a growing momentum now and you know the more people that join the better so I mean thank you all very very much and Nikos I hope you're happy enough with how it went but I just I found it a great conversation I really did very insightful thank you all thanks everybody thank you very much bye bye bye bye take care