 So good morning, good afternoon, good evening, wherever you are. My name is Neil Marshall and I'm the webinar for today's session on demystifying food risk predictions for large food manufacturers and also some details around the case for ETL. So today we're going to have a nice discussion with two esteemed colleagues with myself. And I'm just going to take you through some background information before we get to the meat of the webinar. So in this digital era, AI and big data are dominating the conversation. There's hundreds of systems from robots to advanced software platforms to virtual reality devices that are all aiming to improve food manufacturing. Changes are already taking place at many large manufacturers where major digital transformation projects were already well underway. It's a change that will eventually affect everyone in the industry, whether you're a tech enthusiast, a tech pragmatist or very conservative about technology. The food safety landscape continues to become more and more digital. And of course with that, this brings a lot of pressure on us food safety professionals who need to either upscale or learn how to get the most from this technology. It's a constant challenge. So if I think back to my days at Coca Cola 18 months or so now I guess for me digital technology and suppliers were always trying to knock on my door trying to sell me solutions and convince me of the benefits of their it their solution or their platform. But they often didn't understand the challenges of the problem they needed to solve for me. Because as I always used to say, what is the problem you're trying to solve, and then you give me a solution. So to begin that successful transition, we need to move towards digital transformation. I knew we had to evaluate, adopt and deploy the right suite of tools to make our lives easier and to make the job easier and make the production safer. But we also had to quickly understand the hidden meaning behind some of these buzz words that have been around in the industry for a while now like blockchain, big data and predictive analytics. And that was really the first critical step to understand the language the lingo and the terminology. So we have a profession that's full of critical decisions we have lots of challenges we have things to resolve and debate and discuss all the time. Because the industry is full of ambiguity of uncertainty shades of gray and digital technologies can help us play a critical role in improving and assessing those risks to make better decisions. So when I was still working at Coke. We often had to consider the different data dimensions and aspects continuously because of the amount of production and volume and change in the landscape. For example, deciding how the organization should react to image risks or defining strategic technical framework to help and minimize future incidents. So exploring that product integrity for the beverages, which were obviously delivered to millions and billions of customers, whilst also protecting the brand and the reputation of the company from maybe defective ingredients or suppliers who didn't follow all the processes was obviously very critical important and gave me a lot of fun as well as part of my role. Nice robot picture there for us. What can AI and predictive analytics do for people like us and for the food industry. Or should we invest in this sometimes quite expensive technology and change the way that we and our teams take decisions. A lot of people say you know how reliable are these models the AI models, and what kind of food risks, can they really predict. There's a lot of discussion and debate about that. And I'm sure Janice has come on to tell us more in the future of this webinar and explain a little bit more. But one of the things that I guess often asked is, what are the real life examples from industry at the really safe time. And how can you, you know, deploy those using the predictive analytics to get improved results. Maybe I want to help us to delve a little deeper into this question. So we set up this webinar with my friends from agronau. So we could pose these questions, and hopefully get some of the right answers which I'm sure we will. So with that we come to our esteemed panel. You can see on the right. Sorry, the left, that's myself, Bill Marshall I now run my own consulting business. My name is Alex Coca Cola. In the center we have Nicos, the CEO of agronau, and also Janice the CTO and co founder of Agno. So in this session, we'll try to demystify the food risk predictions and illustrate some practical ways in which they can be used to announce decisions around food safety for food safety quality professionals. So I asked the two people that I consider our world class and superbly qualified to talk about this topic. Nicos Manasuelis and Yanis Stottis, sometimes known as John from agronau. As I said previously Nicos is the founder and CEO of agronau. The food safety intelligence company that predicts risks to help inform prevention. He has 15 years of expertise in the intersection of data and technology for food and agriculture. And he's frequently working with industry, academia, and international organizations on ways in which state of the art technologies may be used to help solve critical challenges in the food supply chain. Hi Nicos, welcome to the video. Yanis is also the business partner and CTO of agronau. The food safety intelligence company that predicts risk to help inform prevention. And he's the creator of food archive and the platform, which we're going to also review today. The food archive platform has been described by many people as the Ferrari of the food risk intelligence solutions. And it's been used by many global manufacturers now like the color company on Agra yearly group and you may leave amongst others. So we talked about definitions I mentioned AI big data and these things already. What is artificial intelligence. So with that, let's open the debate and open the discussion. I'm here for the demystification part that you're here for the demystification to put a bit the mystery out of the war. I like going back to the definitions when doing this. So thanks. Thanks nearly for the intro. You're welcome. When I go back to the definitions. I go back to the dictionary definitions that help us understand what does a specific phrase mean, not forgetting that there is a lots of science behind it. So, Merriam Webster defines AI as a branch of computer science, particularly focusing on simulating intelligent behavior. And if you wonder, what does this mean? What does intelligent behavior mean? It practically means the capability of a system, a machine, a device to imitate what one would call intelligent human behavior. That's the famous Turing test that you're behind a door or a wall and you don't know if you're interacting with the system or with a human. So, what exactly is this intelligence and human behavior? I went to a second definition to demystify that part. And Britannica gave me some very interesting examples because it talks about the ability of a system, a computer software system or a device, a robot, to perform specific tasks, tasks that you would associate and link to an intelligent being, or to follow and execute a process that has some intellectual characteristics of humans. Many typical examples are the ability to reason and explain why you propose a specific action or the ability to observe many, many different data points and generalize or to go and refer to previous experience so that you can take an action. And one of my favorite applications that I have spotted a couple of weeks ago is this particular AI powered machine vision solution by a company called landing in AI. There is the traditional part of image processing that is using a camera and a computer to identify a product that has some deficiency or some defaults. So this is something that employs lots of complex software with no AI up to a point. And then it puts in place, it complements this with machine learning algorithms that are trying to either understand more quickly by generalization or to even predict what is coming up in the flow of products. Or do other more intelligent things by the enhancement of these AI models. What does this mean? What does adding predictive capabilities to such an application mean? It means that there is some kind of model that is trying to imitate intelligent human behavior. And that's exactly the backbone upon which I was based to develop a working definition of AI or food risk predictions or prevention. How can we define this? It's purely based on what we have seen so far. Food risk intelligence or AI refers to systems. So a computer system or other type of device that can imitate the intellectual process that a human follows so that a risk can be identified, evaluated, and then prevented. What does this mean? Intellectual processes that a human follows, for example, discover the meaning that is hidden in different observations and data sources. What would a human do by observing different things and then try to understand what is the meaning there? What is the risk that I see if I'm observing prices going up, the supply breaking down. Climate conditions changing in an area where my suppliers reside. What do I understand? How can a system mimic, imitate this intellectual process? Or learning from past experience? What I have seen happening many, many years as an expert when I'm on the profession that needle and many of you are. How can a system be put in place to imitate this process of using past experience to try to predict something that's coming up? And if we go a bit deeper and we look at the three dimensions that Neil also mentioned, we can even look at the type of process, intellectual process, but the food safety professional is following. Either when monitoring risks, so scanning, investigating or identifying and reporting what is happening outside in the world that is affecting our supply chain. Second, everything that has to do with the way that a human is performing an evaluation, a risk evaluation or assessment or scoring activity. How do I take different data inputs from the outside world, from my supplier audits, from the ingredients risk that I see emerging or any other types of data inputs so that I can connect them together and calculate the risk associated with an ingredient or a supplier. And then the actual processes that a human expert follows when trying to mitigate or prevent a risk. What do I follow as a signal? What do I monitor as a signal in our production environment and or in the outside world that tells me that something is happening. And this should trigger an action from my side. And how can we put in place a system that can actually imitate this intellectual process that I'm following as an expert. So this is the way that I would formulate in a more specific way, the areas and the decisions. And if we go to the specific examples, I am using three well proven excellent solutions from the market that are used to support these decisions, risk monitoring, a widely used and proven system. If we can stay to the previous slide please. That is reporting on the sport, something that is happening, alerting that something was reported and you need to be aware of it or using all the data so that we can report trends, increasing or decreasing trends. So this is what I do with the incidents and other system that is using this data and adds the risk calculation layer. How can I use the information about the incidents about an ingredient or material, a product category or a supplier and put a risk model in place to calculate a score. I use this core to understand whether there is a higher priority something I have to do or a lower priority something I have to do a third example that goes to the direction of prevention. How can I design my house up plan based on hundreds or thousands of similar has a plans in the database so that I can appropriately foresee the control points in my process, all these types of software systems. Are the types of systems that are supporting these critical decision areas. And you will ask me because. Okay, but where come AI value. So I go back to the challenges. I go back to the to the shortcomings that we hear from clients and colleagues when this tasks are being executed I will use one example. A typical example. There is an incident that is being announced if we can stay at the, at the previous slide, please. There is an incident that is being announced by an authority. It arrives at my mailbox. It brings a bell because it's similar or close to something that I read a couple of weeks ago. It's a different announcement. And I'm not so sure should I investigate it again. So they spend hours investigating whether this is something that concerns us or not. That's one of the typical shortcomings where we see AI coming and helping. Imitating the process that I would follow as a human. I'll give you the example that I would come in and dig deeper into this type of early threat monitoring and identification. In which my intellectual processes monitoring as many sources as possible. Reading and processing as many incident reports as possible. Putting together all this information to understand if this is a relevant incident that I should react upon. Or if I see a hazard and a messaging hazard risk or threat that I should be reacting upon. That's the traditional horizon scanning or surveillance or foresight activities that I will follow. So in this kind of use case and in this kind of task where can I play a role if we can go to the next slide. Let's see first how we do something like this today. From a survey that we did with more than 100 companies in our network. We asked them how do you perform this kind of task today. More than 60%, almost 65% are doing it manually. There is someone, a person, a dedicated person or a group of people that spend time visiting official or trusted and trusted sources of information. Either a couple of them like RASAP and the FDA or many of them. 65% of the companies that participate in the survey, they do it manually. And there is about 25% that is using a software system. In the majority of cases, a third party service, and a few of them developing something in house. But not in all cases, this is a whole software system that has intelligence in it. So what we see in the market is that there is still a way ahead until we come to more intelligent systems for exactly this kind of threat risk monitoring and identification. So how does intelligence look like? What is exactly the way in which we should expect in the future AI to come and not only help us automate tasks, but even imitate the thought process, the intellectual process that the food expert follows. I have some examples to share. I started with trying to extract meaning from lots and lots and lots of data, lots and lots and lots of incident reports. This is the case where there is an event that something has been recalled because Salmonella was found. It is announced in different ways in different formats in different languages with different types of content in many, many authorities around the world. And someone has to sit down, read all these announcements, aggregate them in something that makes sense and refers to this particular case. To understand and extract whether there is a specific supplier that we can link to this incident and if there are any relevant suppliers that are also affected or associated. What is the particular product category or ingredient category or specific ingredient that is affected? And what is exactly the hazard or even serotype of hazard that led to this incident? All this process that we take for human hours to perform by putting in place an intelligent software system. We can make it faster and we can make it even more extensive and give something to the human decision maker that has already saved lots and lots of time. Another example, forecasting what will come next. Taking advantage of all the historical data that come in a time period. In this case, the numbers of incidents for a particular product category and how they have been historically going and identifying the way in which they either increase or decrease and build a model. In this case, a time series-based forecasting model that can tell us what is the estimated number of incidents in the weeks to come or in the months to come. A different approach, a different model for a process that you could approximate as an expert, but when a system comes and puts in place some technology, it becomes even more specific and gives you an estimate. That you can work with. And a third example, trying to calculate, estimate the likelihood of a new risk to increase, to come up or increase or decrease. How can you incorporate in a model the different signals that you would take into consideration from the market as an expert. That would lead to you suggesting, for example, to your team that I feel that we have increased likelihood for this particular risk to come and hit us. I'm not so sure, but I estimate that the likelihood is higher during this period and we have to do something. How can we put a model in place that will do this calculation for us and come with an estimation of this likelihood and can provide this as an output. These are different examples of different thought intellectual processes where we are trying to put a model with some relevant data in place to imitate the human process, the human behavior to imitate and make it quicker because such a system is a hard worker and can work efficiently, tirelessly, process lots and lots of data that will take us lots of hours or days to process. In many cases, more extensive, especially if it tries, if it starts incorporating and processing thousands or hundreds of thousands of data points and information that is coming in different languages and bringing them so that it can provide an outcome and it gives something that is proven to be reliable. Some results of the model by the way that you train it and you test it, you see that it would predict good things in the past. So everything here comes with these components being very, very important. The problem that we're trying to solve the process that we're trying to imitate, what kind of model should we put in place and how should we train the model so that we get reliable and accurate results. The specific examples I will let them to Janis to explain what I would like us to think about right now is what would be the difference or what would be the usage scenarios in which we would put such a technology in practice. So imagine that it's as simple as having this in an app in your phone, like we have it in weather prediction. What is exactly the task in which you would put this technology in practice? What would be the process that you already follow now that you would like it to support? With what kind of data would you feed it? What is the outcome that you would expect? This is a very crucial step down the road of taking advantage of this technology. Neil, I'm taking a break here. You've been talking for a while now, yeah, so you need a break. Yeah, but I think there's a few questions and comments in the chat already about the accuracy of the data and the confidence level of the data, but I think we'll come onto that as well later. But can AI really predict the next recall before it happens? That's the difficult question for you. It's good for me. I like to ask difficult questions. I don't have to answer them for a change. I've written an article with this title, if AI can help us predict the next recall. If you're asking me if there is a magic black box or a digital crystal ball that out of the blue will tell us next week where exactly and when exactly the next incident will happen, my answer is no. At least not yet. What we can do with this technology is that we can go and think about the way that you would estimate something like this yourself. So how do you decide? Now, if you go back to the days of you being in such a position, what kind of signals did you take into consideration so that you can take this decision that something nasty is going to happen? The usual ones that everybody looks for, the consumer complaints, the recalls, the industry information, the scans of the horizon, you know, whatever you could pick up trade associations, et cetera. That's the same inputs everyone's looking for to use to base our import. So that's exactly the way that I would approach it, that if we can take these information sources that you would consider and as an expert, reason, generalize, base your thought process on past experience and then arrive to an estimate of the likelihood of something coming out. Can we create a model that will replicate this and imitate this? Yes, we can do this. But if we can build a huge model that will take everything as input and will produce the perfect answer as a result, no. Sure. Okay, I think we need to move to our friend Yanis now to keep to the timeline. Thank you, Nick. Thank you, Neil. Thank you, Nikos, for the mystifying the AI, the prediction. So hello to all. It's great to have you all in this webinar. I will share some experiences from working with several companies and trying to put AI in practice. I would like to share the most important things that we have learned so far. We have identified, as Nikos also mentioned, and Neil at the beginning of this presentation of this webinar, we have identified three critical scenarios in which AI cannot run. In this presentation, these slides, I would like to share which are the most critical decision scenarios that can support good companies to take very important decisions that they have in every day. So the first scenario has to do with the ingredient risks. We are not calculating the risk that is associated with all the key ingredients in the short term in the, which is today in the midterm, next weeks, or in the next month, and in the long term, and next years. The second type and the second scenario for decision support has to do with suppliers risks. So this is about using ingredient risk and other important indicators to calculate the risk and to calculate the ranking, a risk ranking, and being able also in some cases to incorporate predictions. And the third scenario is the one that has to do with risk mitigation and risk prevention. And this is about moving towards a more proactive decision making process. So for these three scenarios, I would like to share some experience with you from putting in place AI. So how do we analyze each scenario? By focusing first of all by focusing on the business question. For us it's very important to understand the business question. And after understanding the business question, the next very critical step is to select the right data to identify which data we can use to answer this question. And then, which prediction method, as you all hear, and you read, there are several methods but it's always very important to select the prediction methods that fits well to the type to the nature of the business question to the nature of the problem that we need to solve. So is this a classification problem? Is this a forecasting problem? Should I use machine learning methods like decision tree or regression, sub-regression, or should I use deep learning? So all these are questions that you can answer if you understand and if you know which type of the method, which method fits well to the type of the question that we need to answer. And then we need of course to train the model that we will develop and to test the model so we can deliver an accurate prediction to the people but they need to take the critical decisions. This webinar has promised also a very specific use case, so a very known issue during the last months, which is the ethanol oxide case. So I would like to walk you through a real life use case. Which is about the decisions within a client that manufactures snacks products in which sesame is one of the main ingredients, together with other ingredients like grain and spices. So this client used for his protons, he used a system like Pudakai to continuously monitor all the risks for all the ingredients, and this is how he managed to identify early and increasing risks for ETO in sesame seeds. And using all the calculations, the dynamic calculations for the risk, the system was able to highlight to the client that there will be an important increase by 400% of the risk for chemical contaminants like ETO in sesame seeds. Using this information, the client performed data information analysis for the sesame and this means that he used, he was able to use all the information about the previous incidents, previous recalls, border rejections, historically going back more than 20 years, so he could identify if such a chemical hazard has affected before the sesame, an ingredient like sesame, and which is the hazard profile based on the specific regions from which you can source the sesame. And of course he focused there on the regions that he was sourcing the sesame in order to identify which are the hazards. So in the short term he had the monitoring and he had all the data, the historical data for the hazards so far. He was also able to use large data sets like the residues monitoring programs for from 34 countries. In order to identify if the ethylene upside has been previously identified in the lab tests for in order to know this monitoring for programs and there were several cases, several samples in which the ETO has exceeded the official limits or was close to the official limits and it was definitely identified during the lab tests and this was even three years before the incident. So he knew at that point that this is not a very new issue that this is something that he need to pay that needs action. He used also the predictive analytics to identify which is the emerging threat for about the sesame of ETO specifically in the sesame so the system highlighted that there will be an increase in this specific hazard. A very important increase in this specific hazard. So he was also able to see some weeks and some months ahead and to identify which should be the next action and if he needs to take preventive measures to manage this issue. But when actually ETO in sesame seeds first emerge, so you probably all know that this was something that was initially announced by some by RASAF and then by some authorities but some local authorities and this was at the beginning of September 2020 and this was indeed emerge very fast, as you can see in this diagram, starting on September on October November we had already a very massive amount of incidents that have been reported and of course this change very much the profile there is profile of sesame of ETO in sesame seeds. But was it really a new issue, was ETO a new issue for the industry. So if we can go to the next slide. Our data says no, clearly no. This was not a new issue. This was something that client identified using historical data that this was an issue that was reported even from 2008, then in 2011 2015 2018 2019 and the last two years it was reported also for things coming from the specific region from from India. So this was not an issue. And client had the data to justify that this is something that was there and it's emerging now. This is also, this was also a known issue in the food safety research so there were early studies and publications that were reporting this kind of issues in the industry, and that there is a need to pay some more attention to that. This is one example of such publication but there are several publications that were reporting that so using all this information of the ingredient level of the ingredients level. And we are talking here for sesame seed but of course, coming from for the specific regions there are several other ingredients that may be relevant like herbs and spices and later on there were other ingredients that were affected. But using all this information, he was sure the client was sure that he needs to take some preventive measures and let's see what he, he did also for his suppliers so for the suppliers client used our client use continuously the monitoring part. So he was able to continuously monitor all his suppliers from the specific region. And he was also able to check suppliers from alternative regions from other regions, so he can for some period he, he, he was thinking to change even the region of sourcing this kind of ingredients. And this kind of monitoring and supplier check was very helpful for him. And then he was also using the predictive analytics to see which suppliers will be affected by this increase in tissue of video in sesame, sesame seeds. And how this, this issue will also change the risk profile of this supplier. So the system using the information of the predicted of the of this increase of the hazard of the ETO hazard highlighted to the client that the risk profile of these suppliers has changed. So he was able to start thinking about preventive measures or about alternative alternative regions for sourcing the specific ingredient. He was able using all this information for ingredients but also other external factors like inspections like warning letters, combined with predictions, but also combined with his internal data about quality compliance about audit scores. So in this combining this different in terms data, he was able to have a dynamic real time risk scoring and ranking for all his supplier suppliers in his supply chain. In ranking the supplier he was able to identify which are the suppliers that are at high risk. So for instance, the supplier that he had for sesame was at high risk. So he needed to take an action about that. After all this analysis, having all this analysis for the predicted trends for the historical, the historical information but also all the dynamic risk ranking and risk assessment of the suppliers. He increased the sampling, the sampling frequency of all the sesame seed that was supplied by the specific region because he managed also to do this analysis for the specific region. And he explored also alternative sourcing locations for the specific ingredient. So this helped him very much to take timely the preventive measures. And after a few months, by studying the predicted ETO incidents in sesame seeds he realized that it was predicted that this will significantly decrease. And he again reduced, he decreased the test sampling frequency and the sourcing of the ingredient of the specific ingredient from media and has been restored. So he got back to the previous situation. So which are the learnings from this specific use case. First for the ingredient risk. What we have learned is that the daily monitoring of the hazard analysis was very useful for the client to see if ETO has previously affected sesame seeds, and which is the trend of this issue for specific regions. In the last term, it was very important that using the launch data sets, the client realized that ETO has previously affected several ingredients from specific region so it was not a new issue. In the next term, the predictions help him to see how the major issues will involve within the next few months, and if he should activate extra preventive measures for the suppliers scenario. In that term, we learned that supplier check and the live food safety profile was very helpful to check also alternative suppliers to find a solution for the availability of the of the ingredient reducing the risk. We also learned that for the midterm, the client was able to combine heterogeneous data sources and get a real time and dynamic risk ranking for all his suppliers and this helped him very much to adjust the lab testing. By increasing the sampling frequency for the sesame seeds so this was very helpful in order to mitigate the risk. And for the risk prevention scenario, we learned that knowing an increasing trend can really help in taking immediate mitigation action. It also made more frequent and dynamic the process of allocating residue prevention measures measures and also the budgets that we are using for that. Yes, what we have learned is that we can become more proactive if we use such large data sets and we use predictive analytics when they are available in this kind of format. What we also did together with the client is that we quantified which is the business value, which is the return of investment, if you want to put in place this kind of technology what you can get back and we did does that we did that for the three scenarios for the risk monitoring for the risk assessment but also for the prevention. We analyzed all the assessments that he needed to do and he needs to do every very frequently to have a dynamic risk assessment approach. And we end up with savings that are larger than four millions. And this has to do with savings from the manual monitoring with savings from the manual assessments as we are showing to these diagrams. In this diagram, sorry, but the most important component of the saving is the preventing the prevention component, what we can save save from preventing incidents. And this is how we also analyze this for the three years. So we have a journey of the business value for these three dimensions. So there are many points in which we believe that AI can make a difference. At the end of the day it's very important to keep in mind that it is about improving the process, which is the transition that we should make very carefully. As it happens with any other digitalization project, it has to be, it has to carefully, we have to carefully consider how it will be introduced into the company. We need to keep in mind that we don't want to distract the way that the things are being done, but we want to improve the work. We want to optimize the work for the risk prevention. We need a very good and planned approach so that the right people can be allocated and the routines be put in place so this process will be improved. This is why we are always proposed in for this kind of technologies we are always proposed an exploratory place so that the people can understand the potential and how the solution can help them to become more proactive. So it's always good to have such effects. I think that's where I come back in, Janice. Thank you for that detailed explanation. Excellent overview and showing the deep dive of the technology and how you helped your client there. That's really interesting. So I guess, you know, there's a few comments and questions in the chat, but also I have some questions for you, so a quick question. So how safe are these predictions? Everyone's asking about that, you know, the confidence level, how confident can we be in the predictions for the future? Can we really trust it? So is there any way, I know you've already given an example there, but do you have any examples of how you can explain that a little bit further? Yeah, I would be happy to do so and indeed this is something that everyone, even ourselves, we are asking when we see this kind of approach. How safe they are, which is the confidence level. So I would say that it's important to know what is feasible with the prediction method that you are working with and the data that we are using. In our case, we are trying to answer the question of if incidents and hazards will increase within the next few months. So it was about forecasting method. And such a forecasting approach can have a model for a forecasting model can have a very good accuracy. We can trust it, but we need to know that this model are very, very good when you have periodic patterns, when you want to identify periodic patterns, or when you have what we call seasonality, when you want to identify increasing or decreasing trend, or you want to identify anomalies in the data. For instance, to identify the extreme case in the incidents, in the time series of all the incidents like the case of ETO that I showed. I was expecting that this will be very good in classification and classifying things into the classification problems is not realistic. So it's not possible. What we are doing to also prove how which is the accuracy of our model is that we are validating the models, both from the accuracy, but also from the use cases point of view. So for the accuracy, we are using some specific methods that estimate the error between the actual and the predicted value values for a reference period, and we are using several periods to estimate this accuracy. It's not only for a specific period. We know that the model in different periods perform performance well, and if it does not perform well, we are showing this to the people, to the users. So they know that the accuracy there is low. And we are also validating the models for historical issues, such as the Fibronil case, the ETO case, or even other not so well known cases but things that were very important in the industry. So if you know the capabilities of the prediction method, and the accuracy for the specific data. Yes, the answer is that you can trust the model. Okay, thank you. I think we need to move on because we're close to the time so yeah so trying to reflect and wrap up from what we saw there, you know there's lots of information around the analytics and how we can move forward. I think one thing I just reflect from listening to Yanis there was you know there's hints out there already, but probably lots of us are too busy to see them so we don't have time to focus on it because we don't have got routine process and system for capturing them. And we need to also then think about that for the future and try to use technology more to help us but I think whether you use the agronome solution the food archive platform or not. You know if it was me again and I was back in my old roles. You need to use a similar process to what's described here by Nikos and Yanis. It's a well thought out process. Carefully calculated steps and I like it because I've seen several big food manufacturers using this kind of process, and usually from like an introductory pilot and a step like that is a good way to get started on the deployment. I think, you know, excuse the picture there. We don't really want to look at myself on the screen there too long but some of the pictures are from, you know, global brands and big companies need to understand the benefits and use that upside potential for food industry and how they can really use it to help their teams to deploy the technology and really enhancing your risk management framework and identifying probably sometimes internally the people, the routines, and how you can improve those decisions in support from the tools you know it's not just one thing or the other the tools are informing you as a company as excellent professionals to make those decisions. But the deployment of the tools can be hugely beneficial and drive productivity savings for time, money, and avoiding recalls you know that's ultimately what we want to do we're trying to avoid these issues. Helping us to be better and help to mitigate and prevent crisis is like I unfortunately went through a few times during my career. You know for the last probably five or six years we've been talking about big data and technology in global industry trade associations gfsi etc. And we've been trying to promote and me particularly promoting the use and deployment of technology to support reducing audits helping to get better at risk assessment. But some people are more interested and passionate and some are not, but I think we need to keep pushing that and driving the adoption because, you know, talking to some of the water, a lot of the big players at gfsi type board level and big multi nationals. They're trying to use that now many people adopting these models now. This is the future this is what you need to do to reduce risk to improve your food manufacturing processes. And the next slide please I think we can move on now to final discussions I guess so I think we're about at time so I'd like to really thank me because and Janice and all the participants and any questions we've not missed yet maybe we can try and cover something now in the last couple of minutes or so. You know if you want to use AI power food predictions and risk assessments. What are you and Janice going to do and how can how can you help them explain that a little bit more. Yeah, so we want to give a flavor of how the food risk predictions look like to everyone that was this webinar and thank you so much for your attention. So, by signing up. You will receive a weekly update that includes emerging Chris trends and predictions powered by the guy. So please use the link to sign up. Thank you so much. And it's quite simple you just indicate three to five ingredient categories or product categories that you want to receive predictions for and then an email will come with specific steps that you can follow to activate this free period of receiving the predictions. So do we have time for a couple of questions. I think you can accept to you you're in charge ultimately I'm just moderating your session for you to. I think you had a there was a question I heard from my list around the 20 year anniversary of a criminal I don't think he mentioned that to me I don't if you want to answer that because I would ask the help of young is did we see. Okay, really might after 20 years of historical data, did we see something like this coming up as a, as a risk in our models. Yes, we see, we see that there is still an issue with acrylic amide, but there is a lot of knowledge for for managing acrylic amide in the in the industry because this is something that comes up during the process and so. I believe that now from now on we will not have so many cases. And so this is not something that the model says that will emerge during the next years. But it always depends on how we are managing this in the, in the industry, because it's highly linked to the process itself. I don't know if you is any more questions that we want to ask or I guess we're about we're slightly over time so I'll leave it up to you to. We've got plenty of good information. Yeah, I think people are happy on the chat seconds you see I think people ask a lot of the questions and comments around the accuracy and the confidence level stuff that. Janice has just responded to so I think we're good. Yeah, and most of the questions I see that they have been replied so far, but we are open to hear others as well. The good point is as we said to leave with is to sign up for the free item attached sending your ingredients and sign up for the freebie. If you have specific questions that we can look at together and so we can dive into the live model and see what do we see in terms of prediction and why does the model predict something these are things that we can explore together in a scenario that you may have in mind. Absolutely. Thank you everybody for your time and I think that concludes the webinar for today. Thanks everyone for watching and it's been interesting for us and we're glad to share any more information. Please use the links attached. Thank you. Thank you, Neil. Thank you, Neil. Thank you all. Bye bye.