 So, it's hello. Good morning, good afternoon, good evening, depending on which part of the world that you're in. And it's my pleasure. I'll give you a little bit of information about myself in a minute, but it's a great pleasure to introduce you to our webinar today on key food safety trends in 2023. And the question that we hope we will be able to answer for you is what are the top 10 emerging risks that AI models can predict. So, in terms of artificial intelligence predictive analytics. How important a tool could those be for all of us going forward in terms of really trying to understand what's coming along the road because there is a huge and massive difference in terms of presenting food safety crisis food safety crisis. And we all know that there's a wide range of measures that have been implemented on prevention of food safety incidents, but we all know that they still occur. And you know the question is why does this happen. Human error will always play an important role, but we have to look to much wider food, food systems issues, potentially new risks emerging in areas or commodities that hadn't been seen previously. Emerging risks called by changes in agricultural practices and manufacturing practices. And of course, shocks to our food supply chains and we've had several of those, you know, over the past few years in terms of the pandemic, in terms of the war in Ukraine, but also the big, the bigger item around our climate crisis. So these factors really should make food safety professionals worried. And the fact that we have got many hundreds of food safety professionals are not yet participating in our webinar I think that's absolutely correct. So what we want to discuss during this webinar is new ways of thinking about moving from crisis management to crisis prevention. And that's through the exploitation of data. Next slide please. What I'd like to do is think about how we can use AI and predictive analytics, but for those people who are unsure what it what it really means, and whether you can trust these or not. Some people are very skeptical. Some people have been very early adopters in this. And what we're going to say today is a number of case studies that I hope will convince you certainly convince me that this is absolutely the right direction of trial in terms of trying to predict issues from happening in the future. So it's my great pleasure to introduce a wonderful panel that has been put together. Not so wonderful is me. My name is Chris Elliott. I'm Professor of Food Safety at Queen University Belfast. And I've been involved in detecting monitoring and dealing with food safety incidents for a very, very long time. And it's my pleasure to introduce Vera Petrovic Dickinson who has an unbelievable experience in working in so many different food businesses, and beyond food businesses, in terms of trying to deal with the massive issues of food safety. This panelist is Richard Stadler who Richard I've known for a long time and we have sat and talked about quite a lot of food crisis over the years. So I don't think there's anybody better than Richard in terms of thinking about global issues that suddenly appear, and there didn't seem to be any particular warning about them either. Our fourth panelist is Giannis Stoikis, who is Chief Technology Officer and a partner at Agrinus that I've worked with for several years now. And I heard their technology, their ways of doing things, really, really very impressive. So Giannis, I'm going to hand over to you now and you can maybe take us through the next section of our webinar. Thank you so much Chris and I would like also to thank you all for joining for your interest in this webinar and thank you in advance for your time. So I will make a very short intro about what we are doing, how we are doing it, and then I will hand over again Chris to you so you can post the most important questions that you would like to be answered together with the other panelists, with the great panelists that we have today. So some things, a few things about Agrinus, we are the data and analytics company that uses AI to predict food safety risks. And the promise what we are promising to the partners and the organizations that we are working with the food companies that we are working with is to provide a reliable reliable risk forecasting for critical raw materials and ingredients by thoroughly using thoroughly tested and accurate AI models. We are measuring what this promise and what we are delivering by monitoring continuously, which is the prediction accuracy of each deployed model, but also by monitoring how many of the market recalls that have been announced and they have hit the market, the market have been early highlighted as an emerging risk by our models. So working with many food companies and many large brands, what we hear as a main concern, when we are discussing about the AI technology and the use of AI technology and how this technology can address the very important challenges that we have in the food safety. And food integrity in general, the main concern that we hear is if we can really trust these AI models. And the other thing that we are, that we hear and we are discussing a lot is that people are saying that they don't understand the experts are don't, they don't understand how these models work. So the confidence for these models is not so high, especially if you cannot understand them. So, we believe, and it's very important for us, it's a priority to deliver this new technologies, the results of the AI models in a simple and easy way, in a way that can be trusted, it can be explained by the experts. So we have developed AI models and we have delivered them through dashboards through live dashboards. And the main goal of this dashboards is to answer the question about which are the emerging food safety trends in the global landscape that we should keep an eye on. And to do so, we are, we have developed a methodology for deploying such models, which is tailor made each time to the specific ingredient, the specific region, and specific hazards that may be linked and may affect the ingredients and the raw materials that we are sourcing from different regions. So the first step of this methodology is to create a tailor made data set for each ingredient. And then the next step is to understand very well which is the question that we want to answer and based on this question to select the best machine learning, the best AI approach that can provide an answer to the business question to the problem that we are trying to solve. After that, we have an iterative process of parameterizing the machine learning algorithms of training and testing the deployed model, the developed models, because the main goal there is eventually to deploy the best performing model, using the data that we have at hand, but also to having as a very important thing to answer the question that we need that we want to provide a solution to so the methodology. The goal of this methodology is to actually taking the different parameters that we may have in a model to be able to deploy the best performing model through a dashboard. Following this methodology, we have been able during the last three years to forecast some of the important increasing trends, increasing risks or emerging risks in the market. So I have some of the examples of these successful use cases that we have also presented in previous webinars so the one use case is that we have forecasted very early, the increasing, the highly increasing trend of ethylene oxide in sesame seeds but also in herbs and spices, which was an issue that was a very important issue that dominated the year of 2021. But it started from 2020 and we had early signals for using for eto in herbs and spices from 2012 and 2014. So we use this information in order to highlight that there will be such an increasing trend in the market in the industry. The second case is the prediction of the increase increased plan for salmonella in chocolate products, which was forecasted in October 2021. Six months before we had this many, many incidents announced about salmonella in chocolate products, not only in Europe but also in the United States. And a very recent case is that based on the information and the trends that we are and the data that we are collecting for heavy metals in different ingredients in different materials like cocoa are models highlighted in May 2022 that there is an increase in demand for cardmium in cocoa and a couple of months ago in December 2022, we had the first consumer reports and lawsuits to companies regarding the occurrence of heavy metals in chocolate products in finished chocolate products. There are also crises that have gone under the radar of AI models and I have also some examples here, some cases here. I will explain later on why this may happen that we may miss an important issue in the industry. One such example is a new risk, an unknown risk that was linked to the peanuts traces in soy lecithin, which was last year, but also the case of ETO which expanded to other food categories like food additives. This was something that was forecasted and actually this created an important issue for ice cream products, but also the very recent Swedish salmonella outbreak that is right now active. These are examples of cases that have gone under the radar of AI models. And as I mentioned later on during this presentation, I will explain when this may happen. And this I will hand over again back to you after this short intro about what we are doing and how we are doing it. So, I'm very I'm looking forward and very eager to hear which are the key questions that we would like to answer which what is very important for our participants today but also for our panelists. And already there's been some questions have started to come into the chat box which is fantastic and please feel free to keep asking your questions. We'll have a dedicated slot in terms of a Q&A session, not only for the panelists but also for our participants so type your questions in. One of the pieces of homework participants were asked to do was to complete a questionnaire and the questionnaire part of it was around where do you think the biggest risks lie in terms of future food safety incidents. What we have here is from small numbers in terms of things like poultry and meat right up to the very serious issues around things like cereal and bakery products. When myself and Janice when we started to really analyze this data because we you know we like to analyze data in a huge amount of detail. I found it incredibly interesting because what I do is I've got my own systems for collecting information and targeting information and forming my own, I would call it a risk register in terms of emerging issues. What I found was that a lot of parallels between the risks that I was picking up and the participants where as well, particularly around cereals. And for a lot of that to me was coming from massive issues about about the massive disruption in supply chains around the world because of what's happening in Ukraine and Russia. I was I was personally surprised at the second ranking and risk was put down as milk and milk products, which I was not picking up. And I'm really interested to find out why that might be the case. How is my data sources, you know, lacking in that, and that might be a subject for discussion for later on. So we can see from this, you know, some of the things that Janice talked about herbs and spices, massive issues and ongoing issues there, oils and fats again, really big issues about about supply and demand because of all of the supply chain shocks there. So I think I think it's a really very, very good and very insightful list of what's going on. So, perhaps we can go to the next slide now. So now what I would like to do is to invite Vera, if you would like to come in and you talk about using predictive analytics to build stronger supplier and material controls. Yes. Thank you Chris, and thank you for inviting me today to this wonderful panel. As you know, I'm a huge believer in AI and predictive analytics have been for many years. And you may ask me why I really truly believe this is going to be a game changer for food safety and quality risk management and for our better decision making in business. And I have a few questions as a business leader on how do we make stronger and better decisions in the area of supplier quality management or material controls. Another question that comes to mind is, what are some of the regional food safety risks that may be impacting materials or suppliers in my network. This really will drive some decisions around analytical performance requirements that we may be enforcing on suppliers for materials, or additional verifications, verification practices for suppliers, for example, increased audits. What are some of the slowly brewing emerging food safety hazards that may be impacting global food safety internal food safety strategies. So I may decide that with the new emerging risk, I need to invest in a better piece of equipment in my lab, or maybe I need to hire a an additional food safety expert to have on staff. Another question that pops in mind is, well, can I actually anticipate future supplier or material performance, so I can make decisions on a long term relationship with my suppliers. Or perhaps we want to have conversations with R&D on reformulation of a product that may be using a material that is predicting to be not performing to our standards. And finally, I'm sure as many of you, you probably have the same issue as I have for the last two to three years due to COVID implications on supply chain. Can we be one step ahead of potential material shortages. And really, this information will drive meaningful discussions with procurement on the diversification of supply base. These are just some of the questions that I have in my mind now over to you Richard. Okay, thank you very much Vera hello to you all and it's a great pleasure for me to be on the panel. So, essentially just to second on what Vera said, I think it's really essential that we try and exploit all the available data we have. In our unit we use machine learning to consolidate aggregating connect data. And basically, these algorithms are used to translate input data into priorities based on occurrence of hazards in the different materials. Now, if you look at the first two points I've put in down here in terms of some questions we would like to answer. The one is about risk when we see a lot about hazards but what is the real risk from a safety compliance and even reputational perspective. And I've added here also are these models sufficiently granular also going into the second point there, because we see a lot of advances in the sciences as we go along new research. And one thing models probably need to do is get deeper into the data, extracting it from tables which is a little bit more difficult. Maybe supplementary material, do they consider, for example, risk assessment data exposure data. So you're getting into that mode of risk, which is, I think, very essential here because my hazard could be there, you're seeing that it's happening. But is it really equating into something which is of a greater concern. And they're staying on the granular tree I think there as well. A lot of the issues and you talked about it in Oxide. A lot of it happens locally originally. And of course that can spread. You've got global supply chains. And so suddenly it'll pop up in many regions, particularly when you're buying materials and produce from a supply producing in a certain country. So there again it's not simple to to understand the hazards but you have to basically then rate those hazards differently in the materials from a low rating to a higher rating depending on where the issues happening and how broad that issue is across your supply chain. And one thing models as well models need data you need to have it there and sometimes you won't have that data and third point there I put down is about correlating associate data on climate change so we all know this it could be linked maybe to soil and water quality but I'm thinking now of an example thinking of the polyimperfluoroclated substances the PFAS would call them. If you look at that we have today. A lot of changes in the regulation I'll talk about that in a minute on the fourth point but what we're seeing here is we don't have maybe have a lot of data in the foods and maybe just looking at water data and soil data and occurrence data and usage of PFAS in the industries. If you have a food manufacturer or you have a farm close to something which is more industrial or usage you may have water contaminated and you're using basically industrial water municipal water or process water in your in your unit in your facility, then you could contaminate your food product. So having that one data set which would be basically water you could translate it into food stuff potentially so basically connecting those nodes doing that interconnection of data is really essential for these these tools and these models to do. And the last point on singles in the regulatory landscape that's a quagmire I mean you know we don't want to go into that two details but staying in on the PFAS example now. We have regulatory changes coming up in the, in the EU we have the EPA on health advisory levels where you're going into the partner quadrillion level. First you know we talked about the billion maybe nowadays pot per million 30 years ago and three decades in this now in quality and food safety now we're talking PPQ. So we've got to go to for PPQ for one of these substances again, what does it mean it means labs can't even measure. And so how are we going to manage that with with AI and data to understand how to connect and how to react because you'll see more issues than that. So your models will see creation of certain trends, and we're going to have to mitigate we're going to have to be able to understand how we're going to manage it from outside from an industry side so these are all the challenges. And maybe I'm putting too much on the table here what I expect models in the future to do, but what you're really shown is really nice trends and trends which are very, very pertinent and very helpful for us. Thank you. I mean, many thanks for our very thanks Richard. I mean some really insightful information, and you know, lots of questions coming from both of you as well which I think is phenomenal really good. And again, questions are starting to come into the chat box please feel free to ask them. What what I will do is the really most difficult questions. Yeah, you just get ready I'll put them in your direction okay. In terms of food safety trends the AI forecasts, perhaps jealous you can tell us not like what you think now based on the algorithms that have developed, what do you think really the key issues that we're going to face into our. Thank you so much Chris so it's, yeah I have. We all have a very hard work to do there are many questions and they are very interesting. All the aspects are really really interesting. So I will try to share here what AI models are forecasting for 2023. For the main categories of that are of concern for all of us for the for us here in panel but also for the participant. So, I, one challenge was how to present all these categories the trends for all these categories so we have created here a table table that is summarizing the main issues for the categories I will highlight some of these issues of course there will be a recording so at any point you can stop here and you can, you can check them in detail, and I will highlight mainly those categories that you have mentioned that are very important for you and that are of concern for the next year. So for milk and milk products, what the AI models that we are deploying are saying in the in this tailor made approach that we are following is that there will be an increase of known risk in this industry, which is the case of Listeria. But there are also imagine the risks like the case of coronavirus that have been highlighted and have been pointed out by the models for 2023 and and trying to answer this on how we can have this risk to be original and to have a good granularity of the models are also focusing and they are providing information about for which geographies we will have high risk for this specific categories so this is in this case this is the United States, France and Mexico. For cereals and bakery, I have, I will analyze further the results of the AI models in my next slide so I will not comment this risks here. And I will, the same stands for herbs and spices so I will go to fats and oils. In fats and oils, yet there is this very important issue right now, during the Ukraine war and all the problems that we have in this region. So, there are some three very important issues that have been highlighted as an increasing risks for the next months. And this is the use of colors like so done for and the ethylene oxide best says like ethylene oxide. And one of the emerging issues that have that are is highlighted by the models is the presence of mineral oil in fats and oils. And of course this is at the level of category but it can be also drill down to specific oils and fats. In the case of oils it can be sunflower oil or it can be also pile more so the models can also provide answers at this granularity because I heard also the point of granularity which is very important. And in terms of the geographies the most risky regions for this kind for this category of ingredients and materials are the regions like Ghana and Syria. And for nuts and seeds for cocoa I will have I will further analyze it in in my slides later on for nuts and seeds. And the models are predicting an increasing risk for aflatoxin but also the presence of eto mainly in seeds. But we will have also emerging issues with some pesticides that have been identified for the first time to be exceeding the limits that I'm showing here in this slide and the geographies with high risk profile here are the United States but also India and China. And for fruits and vegetables. The models are predicting an increasing risk for pesticides and aflatoxin. And here we have again the emerging risk for mineral oil in fruits and vegetables and the regions with high risk profile. And in this case in the case of fruits and vegetables are Turkey, China and the United States. As I mentioned, I will further analyze the case of some of the food categories that were very important. Based on the questions that we had from the panelists but also based on the, on the preferences and the interests of the participants. So I will start with the results with some more details about the forecasted risks for cereals and bakery products. In the case of cereals and bakery products. You can see on the on the top left side of this slide we can see which is the trend of the incidence and which is the predicted trend of the incidence the forecasted trend of the incident. But also we can see a line, a curve that is showing which is how the model performed in the last 12 months so we can also see which was the performance. We see here that it was very difficult to to catch to model this high peak that we had for in cereals during the, at the beginning of 2022. But in general, the accuracy of the model is quite good is at the level of 75%. And the way that it predicts the trend. It's very close to the actual trend that we had during this period. And for the next period, the trend, the forecasted trend is that the incident level of the incidence will be quite the same. However, although we will have a high level of incidence but with no increasing trend for cereals and we will have some hazards that are likely to increase to increase and this is mainly the case of pesticides. So, there are the models are predicting that there will will have more issue with the use of pesticides in cereals. In terms of the regions, the most the risky geographies based on the AI models will be the United States but also Mexico and India for cereals. We are talking for signals here. Let's go and see also the results for herds and spices, as I promised so for herds and spices. Again, we can see which, how our models are predicting the trend. So, as you can see, during the last year, 2022 and 2021, we have a high level of incidence already so the number of the incidence is at the higher level that in previous years. This trend based on according to our models will remain and the performance of the model specifically for herds and spices because this is a model that it has been developed only for herds and spices using global data has a good performance the accuracy is not the accuracy is 81% bit more than 91% and the main hazards that our models predict that will increase during the next months are pesticides, salmonella, but also the case of it all the ethylene oxide will still we will still have more issues with ethylene oxide for herds and spices. One of the very interesting cases that I mentioned at the beginning of this in the successful cases at the beginning of this presentation is the case of heavy metals in chocolate products and cocoa and chocolate products. And here I would like to with the results I would like to answer the question that Richard made about correlating different associated data and correlating different and using different data types in order to confirm or to identify increasing and this is I will try to answer this in this specific use case. So what we have done here is that we have used two different data types. One types of info one type of information that we have used to estimate the risk during the last for more than 10 years was the global incidents that we have the frequency of the incidents and the other type of information type of data that we have used is the frequency of heavy metals occurrence that are about the limit of quantification in cocoa. So as we can see here, this is the plot the graph, the chart of overall risk using these two associated data types, the overall risk in of heavy metals in cocoa and chocolate products during the last 10 years is highly increasing. And also the AI models are predicting that there will still will still have this increase during the next 12 months. And if we go and try to explain why this happening and we can if we focus and see which are the insights from the global incidents point of view. So we will see that after 2016 the frequency of these incidents of having heavy metals in cocoa or in Finnish in chocolate products were increased by 360%. So this very important increase. And if we see, if we do the similar analysis for the occurrence of the heavy metals in cocoa, using global lab test results for concentration of heavy metals. We will see again that the number of the cocoa samples in which the concentration of heavy metals were found above the limit of quantification has doubled in the last five years. And this was mainly due that the high peaks that we have in this in the concentration that was measured was mainly due to the presence of cardmium in cocoa. And I would like also go one step further and answer the question for the regional for the regional risk for the granularity of this model can we see for which regions for cocoa from which regions, we have higher risk, doing using all this data, and are using AI models. We have identified that the high risk regions are based on the data that we have is Ecuador and also Columbia, and this will increase during the next months. So this is how the risk that we are estimating which can be the overall risk can give us a very good idea of how the risk will increase overall for a specific material and then for for a specific hazard for a specific issue like heavy metals. This can be also become very specific and point out to us from which regions we have higher risk for this specific material. And one of the questions that we receive very frequently very often when we talk with with the companies that we are working with or with organizations, or working with is what we can do with this knowledge. So we were one of the, we are sharing here some of the most important things that we have agreed. All that this is a very important knowledge and that we can use this knowledge in order to communicate the risk inside internally in an organization to share this knowledge in a fast way. So not only the food safety departments and food quality department but also procurement departments know about this, this emerging issue. We can also use this knowledge to practically adjust supplier practices but also we can adjust some specific preventive measures like the, the testing plans that we have or the audit plan that we, we have. In some cases, we may also use this information to make a very hard decision about changing suppliers and using suppliers from other regions with less risk for the specific material. As mentioned at the first part of the presentation. There are also crisis that have gone and will go under the radar of the models and this we have tried to identify and point out here when this may happen. And finally, what, when we cannot be, we cannot predict things is in cases when we are talking about really new and unknown risk for which we, there is no previous knowledge, and no reports announced, no research studies, no research results. There is no additional regulatory framework so this is something very, very difficult. I will not count the case of PFAS in that based on the based on your comment Richard because PFAS we have already some data for PFAS. But the regulatory framework is trying to provide an answer to this increased risk that the data are highlighting. And this is very difficult. Another case in which it's very difficult to predict things is that we cannot predict. In many cases the social and economic situation in specific regions, so we cannot predict how the GDP of specific region will go. This has, this is dependent on many, many different parameters it's a very complex thing. So, we can use this as a parameters, as to correlate this data this associated data about the risk, the country is that we have based on the socio economical situation, but not to predict this situation it's very hard to predict this situation. We can use this, this parameter as risk drivers, these parameters as risk drivers that may increase the likelihood of a risk for a specific material. And also, something that is very, very much linked to what I have explained with socio economical situation with predicting the socio economical situation is that it's very hard to predict events that are dependent on many different parameters that cannot be predicted so this is something. When, when we have a chaotic situation, it's very difficult to predict something. And if we have again some new behavioral people on how they handle an ingredient or how they handle how they are behaving in a processing in the processing framework or in a processing. When they are part of the processing of the food. This is something that is very difficult if we have no data of such behavior before. So these are, these are the points. And these are the cases in which we cannot provide questions using the models based on our experience so far. I think that some of this of the technology will get more mature we will get some more data and we will be able also to provide questions for some of these cases. And with that, I will hand over back to you, Chris. So we have a reflection by you the panels but also questions by audience. Yeah. Thanks again, Janice. I have to say there's lots and lots of questions coming in really good questions. So I think that's fantastic and please, please keep typing your questions into the Q&A box. And while you're gathering your thoughts together, maybe I'll just turn to Vera again just to think about, you know, from what you've heard, but also from your, you know, your knowledge in this area, what your reflections are in terms of the robustness of these predictions. Yeah, thank you so much, Chris. Great session so far. Really a lot of learnings and, like I said, I love this topic. I just wanted to share a couple of thoughts. AI is of course a very powerful tool. But it is just that it's it's just a tool. It certainly has been unlocking things that we couldn't or cannot as a human as human beings do right so in particular with the amount of information that is coming at us. But I have to just encourage everyone that no matter what predictions AI may be driving or proposing to you, it's still merely an input, like with any forecast, you still have to have robust internal discussions on what your final business recommendations should be. As Yanis mentioned, there's so many contributing factors that may change your direction, right. It's the geopolitical scenario that you didn't expect or maybe changes with your supplier organizational structure or something else that's happening that AI simply would not be able to pick up. So to me, if I were to drive a comparison. If you look at this picture, AI is basically like the light that's shining through really thick forest is just showing you the direction that you need to go. You go into, but your really path, you and your business partner partners need to define what that looks like. So it's a great tool, but it's not the answer that is not the final answer for everything Richard. Thank you, Vera. Absolutely. I mean, it's, I think it's AI. This is one element in the toolkit of early warning that we have and that we really can exploit. So, yeah, for some reflections, you know, we, we have a lot of data out there. And this we have to definitely use better interconnected data as I said before, identify signals and trends. But overall, you know, it's really key at one point still to have the expert available, because there's a lot of knowledge there. And there still has to be a human judgment on those on those predictions. And I think that's, that's really, it's really key. So what needs to be filtered out? Can it be filtered out? And is it reliable? Even some of the data that you find there in the science, in the web, you still can question some of it. So it needs a critical eye in my opinion. But I think that's something we need to definitely factor in. The point on source data and reliability, that's exactly what I just said, verification. Again, it's going to cost resources. How much do you want to invest in that to verify, but I think it's key. Some of it can be done very quickly. Other things will need to be, you know, they will take more time. It might even end up in a program, particularly if you're sourcing in different places and you've got to dig deeper. And that might might be, yes, necessary required. And so also, you know, you're looking at decisions which are in real time. So you've got to react fast as well. I think that's that's also of key importance on on the surprising issues. I mean, what I've seen here and what you've shown, I mean, there's topics, which are clearly reemerging. And as I said, you've got so much data there for data we we sometimes even with those with the cadmium with some Nella and with with things you've highlighted mirrors, etc. We tend sometimes to forget, you know, and we don't understand what the new triggers are. Is it based on the material? Is it based on the sourcing country where things have changed in agricultural practices, or have they changed in manufacturing practices. So I think that's, that's important also to follow up and not forget those and don't put them out in, you know, into isolation, or forget them. So this is really key that we keep monitoring them as you've just shown. So yeah, those would be these would be my reflections. So thanks very much, very much, and Richard again, because I think those reflections are really important, and they overlap a great deal with with mine as well. And I think I'll take the chairs prerogative and not spend time of me talking. I would much rather now to get the audience participation, because there's some really good questions coming in here. Unfortunately, Janice, most of them are for you about the data that you have presented. Now, again, there are multiple questions and apologies if I can't deal with all of them. There's just a couple, and I think I know the answers to them. And if I'm correct Janice you just say yes or or if you want to add to it. I mean one of the questions about the granularity of the data. And you talked about potentially based based on what the, the forecaster showing you changing suppliers, but I don't think your, your, your, your data is granular to suppliers. It's about countries where people source things from, and then you would have to think about changing your supplier that sources from a different geographical region. Is that a correct interpretation? Yeah, it's very correct. We have also data for suppliers, but most of the data have the source, the origin country of the material of the ingredients. But we have also most of the incidents that we have. They are also linked to suppliers so we can project these predictions this forecast this risk forecast also to suppliers. Now, I'm trying to combine a couple of questions here, because it is about the various sources of data that you use and they're asking what are your sources of data. And as you increase your data sources, the amount of testing that goes on is increasing, or actually you're just picking up trends and more testing rather than food safety trends itself so perhaps to try to cover those collectively. Yes, so that's a great question because we are the ones, we belong to the group of the people that we very much believe, first in data, and then in what's the, what the models they models can predict using this data. So we are expanding continuously the data in different data types to cover also different types of information that are there and are linked to parameters to risk drivers that may increase the risk for specific ingredients and materials, but also for the information types that we have already for the data types that we have already like the incident so the laboratory testing data, we are continuously adding new data sources new countries that are publishing information. And yeah we have a growth of the data of the data set that of this large harmonized data that we are data set that we are creating that is at least 10% every year. And this is actually very helpful because having better data we are, we all understand that we are very dependent on the data. The AI models cannot predict cannot forecast cannot be built cannot be tested if we don't have data if we don't have a good data. So the increase of this data is of great importance also to improve the performance of the AI models. And yeah we are always talking about not only about collecting the data downloading the data but also about harmonizing all these different data. Which is hard work but it can be done. There are technologies that you can utilize and can show this issue. Thanks for that. There's a number of questions about will you supply CPD certificates and I'm quite sure you'll be doing that to all of the participants, but very important part of our professional careers. And also some people are saying, will the presentation will be available online and again the answer to that is yes absolutely this will be this will be streamed for quite a while. There is, I guess, probably one of the best questions because it's the question I asked myself quite regularly. I think, Richard Vera, you will ask yourself the same question. What's going to be the next big scandal. What is going to be the next crisis of a scale of melamine. And you know, is that something that can or cannot be predicted. Is there a specific scandal or do you want to answer the question of if it can be predicted or not. So I mean, I guess if we all knew what the next melamine scandal is, we wouldn't be on the webinar we would be doing other things getting ready for it. In terms of, I think a better question is, or perhaps more fitting to the data. Can you predict the level of the crisis that might happen, based on the red flags that you are identifying, you know, you know, to me, a problem with a with a with a mycotoxin is a kind of one type of risk. So if you're finding a compiler factor in food products and the numbers are rising that that's a completely different scale event. So can you scale the severity of the potential incidents. Yeah, this is what we can do so we can, we can estimate and we can anticipate which how how big will be this increasing trend. And then for mycotoxins if the mycotoxins due to the climate conditions due to the climate change will increase a lot during for the next couple of years, and we have the data for the climate change and the data for for early periods with mycotoxin this is something that for sure it can be predicted by the models and this this is something that can be done. So something that can be forecasted is some new issues new hazards that have been identified sporadically but they are affecting broadly different types of foods, but there is an underlying connection. So the regions are based on the conditions that this ingredients or materials have been produced. So this is again cases that we can model and we can predict well with AI models, but still, your point is very good that we are very much dependent on data so we cannot build models if we don't have the data. So we, and there are there are a lot of data there. They need to be collected harmonized, and there are also a lot of data internally in the food company so we need to unlock also this sharing of data between sharing data in an anonymous way that can be used by the model to forecast the next very important scandal. Thanks again for that. I mean, there are lots and lots of more questions and I think we are running out of time. So I'm going to ask you another question because it was the same question that I had in my mind, because you talked about things that were on our screen and that you explained it very well. What about things that you predict are going to be a problem that actually aren't a problem. And how many times will you raise red flags and you know companies and businesses will put lots of efforts into trying to deal with something, but it doesn't materialize. So let's start with the second, the second part of the if your question which is about how we can and we are discussing a lot this with with the companies that we are working with how we can become and how we can help the companies become more agile when they see an increasing risk and how they can be, how they can confirm this increasing trend and how they can activate the preventive measures so this is something that we are discussing a lot. This is this is one of the very important things because we hear about mineral we hear about heavy metals but we are not doing we know about that. We see also the predictions but we are not doing any actions about it. So this is something that I very important thing to work on during the next years and also yet there are of course we have our models are quite sensitive so they may be cases that are that there is this problem there but they may be over highlighted so they may be over emphasized. So we will what we are doing in these cases is that we are collaborating with the company and we are digging deeper to identify what this problem, what this emerging issue was which was the region if we have other data that we can utilize to confirm this issue. So these are the actions that we are doing when we see these kinds of things that are not very clear are the red flags or green flag but it may happen yet. No, many thanks for that and I mean, thanks for all of those answers. So maybe just to wrap up will quickly go on to mean the reflections very Richard and maybe a few myself. No, I just wanted to mention a couple of few things. And it's probably more call for action, I think we are here on unlocking something incredible for our for food safety and quality in terms of application and the benefits that I can bring us. But really my call for action, I don't think that we will go far if we don't have the right early adopters in our leadership community, who actually are interested in playing with it, interested in doing some real life pilots with companies like So, and the more we do that, the better our prediction accuracy of the prediction is going to be. And also we're going to together will be able to evolve this machine learning to the levels that we want, we want to. I think it's really important that we continue to drive webinars like these because we need to continue to educate our public about the data and analytics in general and AI, and remove some fears that I see they're very, very real. So final, final food for thought. I really think that this is predictive analytics is the missing piece for us that will allow us as business as businesses to make more accurate predictions or decisions. So if you think of it this community actually the precision of our decisions will mean reduction of recalls or creating a safer food supply supply for our consumers. So it's a big deal. I believe in this in this tool, and really hope that we can partner up as as leaders of food safety and quality community. Oh, thanks Chris so some of the reflections I put together into your to your previous point on sharing data anonymous I think they are already initiatives where this is starting in Finn for example. But I think we can only grow that I think that's so important and going back to what we did in food and Europe at least in the activities we have there we put together some of the data on food ran and throw rate and water. And we really shared that information. And I think that's really key, so that also the public understands you know what are the risks what what the hazards that we find, and how they interpreted so the data has to be definitely generated and shared much, much more. And I think that's something we will definitely look at because in the day you know food safety, we're all interested in food safety wants that food to be safe. And it's not something which is proprietary or which is confidential I think it's important that we share anonymously this is key. And then you build those larger data sets and then of course your models will be far more accurate and predictions. And I think the quality I put down because we've seen examples and cases where this is really lacking and we would like to dig a bit more deep into some of the sources verification that requires of course industry concerted actions but possibly to to verify to look at those early indicators this is important because it's part of an early warning tool, essentially so we want to read fast you can test it's not always too expensive you see things happening. There's a time of the essence here because it can easily spiral out of control if you're not acting fast and you're not looking at your sources and understanding how to mitigate at the end of the day. Decisions taken to these considerations I think what you've shown this this definitely flows into how we basically do risk classification of materials. So, it's essential in HCCP studies, finally after verification in a broader context, and then some of those parameters may even flow into our peers the raw material purchasing specifications I mean of course mineral oils will be an approach to the specifications as normal. And as you dig deeper and you go into ways to to manage, you can set up purchase specification and then of course, finally into a contaminants events plan, which you shouldn't do once every year you should really do it more frequently to see that you really capture those hazards or those risks let's say which which are popping up in real time and that you can react to those accordingly. So monitoring and surveillance plans this is where that information flows into. And then finally it will, if you have a strategic research program in input safety that will obviously in some cases and in our case definitely some of these contaminants are on our programs, which can span from from short three months six months up to up to a year or even two years to understand how we can manage them and how we can mitigate because that's the end of the day you want to see a risk in terms of food safety compliance you've got to drive it down. And that's where of course you need heavy resources but again where resources can be shared through Horizon 2020 programs and others. And I think they also we need to do a lot more terms of research together with the different actors. In the space. Richard, that's really very very helpful thank you for that. So I guess we're a couple of minutes over now so I'm just going to wrap up in terms of where we are. You know, the progress that has been made over the past three years and I think it's been very well demonstrated in terms of the models that are being built are becoming more and more robust, more and more reliable. And certainly it is proving to be, you know, a really important and insightful tool in trying to predict what issues are coming short term and longer term as well. And I think also the, what we've tried to do during this webinar is this process of demystification. You know, what is this black box that gives out out information. I think very touched on really well you know training is very very important. This is a new skill this this is this is something you know that we want to add into our ways of being able to manage things. And it will only come about through standardization and validation. And again, you know, I think on behalf of agronome. The more people the more companies have become involved in this, the more reliable the data becomes becomes you know shared shared issues in ways that food industry really quite struggles to do. I think the case is that that were provided today by by Janice I mean we're quite excellent. So I think if you allow me just to kind of a very very quick summary just to finish up with. As I look through, you know this presentation and you know I'm lucky because I've looked at it several times now. What I can see in my mind is, I can see all of those things as a professor of food security that we talk about every day. We're seeing the impacts of climate change here. We are seeing the impacts of the Ukraine war. We're seeing the impacts of fraud because I can see fraud and food safety. They're coming across each other. We have a very, very complex global food supply system. We know now that is really showing to have frailties because of the pressures it has been put under. And some risks. We do not know what are coming. And again, in the chat box I could see people say this happened 20 years ago, and exactly right. But for many of us our memory isn't that good and we forget what there was a massive issue in Salmonella and chocolate 20 years ago, and here it comes back again. The validation of the issues that are coming, I think to me are very, very strong. And I'd just like to finish off with, do we really need to fully understand how predictive analytics works to trust it? Since the last time you watched your TV, and you saw a weather prediction, and the weather prediction said tomorrow the weather is going to be horrible. And do then is you change what you're going to do tomorrow. Do you actually know how that weather prediction was made? I don't, but I trust it. I would like to thank absolutely thank the panel that we have today, Richard Vera Janus. I think the information, the insight, absolutely fantastic. I want to thank you all for that. And I want to thank the audience for joining us today. I think the questions really shows the interest in the topic. So wherever you are in the world, take the rest of the day or the rest of the night off. Don't worry too much because you know predictive analytics will be there to solve your problems. Thank you all very much. Thank you so much. Thank you. Bye.