 Here we go. Good morning and good afternoon. Good evening. Depending on which part of the world you're based, this is going to be an extremely interesting webinar, online panel. We will be talking a lot about laboratory data, testing results, connecting different data pieces together and extracting value. I'm giving a couple of minutes to our participants to join. While I will be introducing the panel, we will zoom in the ways in which we can unlock the value that is hidden in laboratory data. But also discuss a little bit about what we mean with this seminar and how we can put data in practice to inform risk-preventing prevention. Let's start. It's all about testing, food safety testing, but not only food safety testing, testing that is generating insights related to what we find in food, in food samples, how safe is the food that we are consuming and that we are using in products, as well as additional aspects of the food samples that we test in terms of allergens, the presence of ingredients or other aspects in the food that we look at that shouldn't be there. And it's a huge business. And it's a business that is growing. While it is growing, it is also generating lots and lots and lots of data. We had this conversation a couple of months ago in one of the workshops that we had on the topic of using data to train AI models and everyone was complaining to me because they said, yes, AI is good, but we really struggled with putting the data together, finding a way, first of all, to digitize properly what we generate to testing, harmonize it, make sure that the different data pieces, they are talking about the same thing and combining them so that we can extract value from a large variety and a large volume of testing data. And this is becoming even more practical. If we look at where these insights may add value. I love this example. It's a quote from a colleague that we had this conversation, it's two or three years ago and he said, I really think that we are testing in blind. We invest so much money in food safety and fraud testing and I still feel that we are testing for things in blind. We are not sure where should we be focusing on, how should we prioritize what we are looking for. And it really made an impression on me investing such an amount of money in testing and then still struggling with extracting the value and taking real decisions. And then there is also the other dimension of having access to all the data that is being generated. Another interesting conversation that I had with a colleague describing the model that they are following that is combining internal laboratories that are generating lots and lots of testing results with laboratories from a third party that are storing the results in their own systems and he was telling me that when I'm asking this third party to provide me with access to our data, the data that we have paid for, they want to charge us. They ask for additional money so that we can understand what is the value there, what are the testing data telling us. So it can become quite a struggle putting everything in place and then using it in real decisions and that's why this panel is here. When people were registering for the webinar, many of them completed the registration questionnaire that asked them to rank these challenges in terms of importance. As you can see, many of the people in the audience they feel lost and sure in ways in which they can best combine all the data that is being generated and then integrating them in internal systems. This ranked as the number one challenge and then number two is whether they are testing for the right hazards or are we looking for something that is of high priority right now? Is there any other emerging risk that we are overlooking? And finally another challenge that has to do with the data sitting in plenty of systems, very often for parties and being very difficult to access in years. This is why this panel is here and this is why I've invited this wonderful group of panelists and I will ask you to introduce yourselves briefly and then tell us a little bit about why you're here and why did you find this panel interesting? Joel, would you like to start? Sure, absolutely. Well, thanks for having me first. That's the first thing I'd like to leave. But my name is Joel Masso. I work as Associate VP of Science Programs at Western Growers. We have a number of data type programs related to the food safety of fresh produce. Western Growers, for those who aren't aware, we're a trade association. About 50% of fresh produce in the United States originates from members who are part of our trade association. So it's effectively all the specialty crops, lettuce, citrus, nuts, those sorts of things that originate from the Western States of the United States, Western Growers represents. And from my background, I'm a quantitative microbial risk assessment person from graduate school as well as spent many years at a global testing company thinking about and helping clients use their data or accumulate data to be able to try to get to what is that holy grail of food safety management, of using your data to predict and prevent as opposed to just react to. So I'm excited to be part of this panel and to contribute. What we've learned at Western Growers with our data sharing programs, my background in microbial testing. Wonderful. Alec, how about you? Yeah, likewise. I want to thank you for having me on this panel here. So I'm currently serving as a global manager for digital quality and food safety at ADM. So it's part of my, my day to day is looking at our digital transformation and automation opportunities within the business, but also what is our analytics strategy how can we drive continuous improvement within the company of quality and food safety as well as quality control lens. So really last 15 years, my background has been in lens implementations as well as quality and food safety digital system implementations. And then what's the analytic strategy that those can drive from either mostly from a statistical quality control perspective. So really been a passion of mine for the last 15 years and really what I've lived over the course of my career. So I'm very excited to be here over the course of my career. It's been such a large breath of different products, different businesses that I've had the privilege to work in from agriculture to food and beverage to ready to eat type products. So I'm excited to be here. Wonderful. And last but not least, Yanis. Nikos, thank you very much for the intro for the very nice intro. It's a pleasure and an honor to be with the panelists that we have together today with us, with Alex and Joel. And I would like to thank them that they are here. My background is in the computer science, so I will speak more from the technology point of view. So however, I felt in love with the food safety domain and the challenges that this domain has very early in my career. So in the last for more than 10 years now, I'm working, helping the companies to digitize the risk, monitoring the risk assessment and the risk prevention workflows. And I'm very happy to be here because I see the struggle on how we can manage and how we can combine all the testing results. And I see also the opportunity of how much this can help in improving the risk prevention workforce. It's wonderful to have you, the three of you here. I think that you are the right people for this conversation and the mix that we have around the table is very interesting. Let me provide a little bit of an intro. I was trying to pull together a couple of slides to explain to our audience even the non-expert ones to explain them why this is important and why do we talk about such a large variety and a large volume in food safety or quality testing. And my favorite exercise is looking for real actual data pieces. And my even more favorite example are the certificates of analysis the typical outcome from one of the laboratories. I'm going back many, many years and I remember I recall the certificates of analysis that my dad was breaking at home because he was working on very, very relevant related topics. They looked old fashioned to me pieces of papers with numbers. I always thought computers can do it better. And here we are today looking again at old fashioned pieces of paper or information and we are having this conversation about whether and how computers can do it better. And when we talk about testing food safety plus testing we're looking about in essence understanding contamination contamination factors and the degree of contamination in different aspects and looking for different types of contaminants and hazards. So we are talking about a world that is quite diverse and broad and depending on the area in which we focus we are dealing with different types of data being generated. So here with a very brief research on the types of data that I see being generated for example for chemical contamination I saw the traditional pieces of paper or PDF files that I was showing you before with different ways of expressing what they found in terms of chemical contaminants presence and samples different versions of this data being generated up to a spreadsheet that can be automatically produced by Linz software like this table that I have here as an example very very very similar if we look at biological or microbial contamination although there we're looking at different parameters different types of contaminants different methods are being used and in many cases understanding if the contaminant that we're looking for the organism that we're looking for is the one that we are expecting or not even employees more advanced methods like PCR testing lots and lots of additional data being generated and then if we start looking at ways in which we can for example look at the authenticity of the sample if it contains what we expected to contain or what it declares that it should be containing then things can become even more sophisticated and the data types that are being generated can be even more complex thankfully computers have been employed thankfully plenty of software systems have been developed and deployed but this means that we have many of them it's a whole ecosystem different limbs, platforms or database is storing this information so this is a world that is becoming quite complex a large variety of very heterogeneous data and with many many systems in which they are stored so Yanis I will ask for your help to navigate us a little bit into how we can make this world a bit more unified or harmonized before I do this I want to get a feeling from Joel and Alex on what does testing mean in their own lives their day to day practices and in their own worlds Alex I will start with you what does testing mean for you sure so I see test results coming from multiple facets in my experience some of them come from the lab which is traditional testing I think we all see but I see a lot of test results also coming from the shop floor whether that could be micro testing that we do on the shop floor could be in line testing we have operators do could also be in line instrumentation I really need to gather data on a daily basis that data I see really supporting both our material sourcing if we think about raw material agricultural materials coming in from a grading perspective or bringing in receipts of new raw materials but I also see that from a manufacturing operations and really supporting that process I think about all of these items really from a daily basis I see thousands and thousands of test results that could be harmonized but typically it's high variability in data different formats different languages timeliness of the data is different it could be real time it could be a weeks delay depending on where it's at so that really influences the consistency in the data and some of the challenges that I see on a daily basis to understand also some of the data is structured it's documents it's bench sheets that are manually recorded at our plants so how do we get that data in order to really drive that value and unlock that value we're all looking at from either a risk management perspective or looking at our lab and facility operations local fund structure data here you say but also data that is coming with different time latencies so maybe they count as a delay that's very interesting Joel what about you yeah well thanks I'm going to approach it from the current role that I'm in as well as the data projects I'm associated with so I referenced earlier that Western growers represents a group of growers and one group specifically has kind of entered into this new paradigm of testing and how to use data and we really have a lot of outbreaks and issues associated with them when it comes to food safety and the industry continually just chips out like good agricultural practices how to implement new methods and the hard piece is always to figure out ways to assess our efforts doing anything and where potentially are the risks coming from so as we do outdoor agriculture where we have limited ability to predict or prevent the impact of the industry on the field like they're in the proverbial kind of like wild goose chase so a few years ago the industry banded together and recognized that all the efforts people were doing on microbial testing specifically so for pathogenic E. coli and salmonella and water testing so indicators there's an enormous amount of data that's going on and so there's got to be a better way we can learn something from all of these data points and transition from yay it passed my specification ship the product to what did all of this money and effort mean so western growers created with our partner from global a data sharing platform that allows the industry to anonymously submit their data and their efforts how and where that data is generated that also leads to lots of questions when you start looking at data from all these different streams is what does it really mean so this is a relatively new project it really is that new paradigm that everyone talks about of like how do we use data more productively what does microbial testing or just any kind of testing mean and aggregate as opposed to just reacting to it there's two programs that are on it there's two they're both related to leafy green production the western growers leafy green food safety data sharing program and then the leafy green marketing agreement remain test and learn program and so we're managing both of these programs for the industry they are unique and different they also are somewhat looking at the same problem and effectively you're trying to use this data to verify the context to maybe metadata or things associated with these data points that could teach us something about managing what can be a very high-risk product it is also leading to some really productive conversations that we need to be having as a kind of food safety society in respect to understanding origin of data and how you design a program I think we've been somewhat sloppy slash with testing in the sense that we do the test we get the result and move on when you start looking at data in aggregate the critical nature of like why you took the test what the test could theoretically and statistically mean and the limits of the test methods and how you got through it actually have pretty huge implications to interpreting what the results may tell you and that's what we're getting at right now which I think is it's great and there's the exercise to learn something as an industry I think that's really formative that we're learning in this project but I think probably more productive actually is the conversations that we're now having about like well we've always done a salmonella test but maybe they're all not equal or how could we standardize what we're doing such that when we are analyzing the data the first question that comes to mind isn't do we trust it and that it's really leading to a lot of really productive I would say more maturity than a lot of other produce segments as well as a lot more complexity so it's been it's an interesting exercise on many different levels so I'll stop there but that's my experience right now with testing data and culture I really like the fact that you you talk about how pooling together data from different stakeholders first of all requires a new language how can we describe the data among us it serves as a way to verify and inform the practices that we follow when we are producing but at the end of the day it puts on the table how we use testing percent this very sophisticated but also very expensive instrument amazing thank you guys but very very interesting to get a bit of a feedback from our audience as well in terms of what are they testing for so I will ask here for the help of the magic pool and I will give you a few seconds to respond to this what are you primarily testing for what kind of contamination are you afraid of what is most important for you guys in the audience lots of quality characteristics I see can you also see the results yes ok so I see lots of testing in terms of quality in terms of authenticity and then chemical biological contamination this is where I see responses coming in ok let's use this as input for our conversation you saw the result off we go Janis we need a common language and we need ways in which we can trust this huge volume variety velocity of data help us understand a little bit better how we can do this yeah I will try to do this in the next slide there are two parts in the first part I will try to describe which are the challenges and how we can overcome these challenges that we have when we try to combine the data in the second part of the few slides that I have I will show what can this unlock which business cases this can unlock and some of them are very relevant to the discussions that we have already so how many times you had the you needed to to get a trend for a risk parameter or you needed to identify the frequency of the parameter and you asked from other systems or from other databases to get an export of the data because you thought that it is easy to combine this data and answer the question that the critical question that you had in mind it happens to me many times I am very optimistic in that but still there are you cannot we cannot do that we cannot do that we cannot combine the different files that we are getting from the systems and although we have already the systems and we have already the data we cannot combine them we cannot harmonize them because that we have already mentioned some of them so first of all it's a very small part of the challenge but it's still an obstacle we have different formats of the data it can be Excel.pdf format database is API so it can be the data can be available in different formats we have both structured and unstructured information and unstructured information is a bit more challenging I will share an example as well in the next slide we have different structure information in terms of how which fields are used to describe the information the so called metadata I will also deep dive a bit about it we have different terms used to describe the same things so even if the foods there are so many food and vocabularies out there different systems are using different vocabularies and classifications and as Alex also mentioned there are different languages used so these are the main challenges that we see and although we think that the most important problem when it comes to harmonization and trying to combine the data is to harmonize the data itself the result, the actual number this is not the case here the most important issue is to harmonize the fields, the metadata that are used, the elements that are used to describe our data and we have very important elements there that are used for instance just take a very simple example shown in this example in the 1880 slide we found it at the old years of agronome when you need to describe the publication you need to have the idea of the specific publication the author, the title, the short description you need to have the date you need to have the catalog so it's very much the same thing for all the things for all the resources especially the testing resources they have so many parameters so it's about combining the metadata what is needed which is the solution when we we need to describe in a unified way the testing resources one solution is to define to specify a unified reference data model and this has two main steps the one step is to select the key attributes that we need to describe all the data the testing results and this may include the date of analysis the method that was used the type of the sample which parameters was analysed the actual results so all these are the columns that we see in a typical Excel tabular format but the second very important step is for each of these elements the metadata elements we need to define which are the values that are around are free text values around or they are controlled values and it is a specific list of values that we need to add there so these are the very important two steps to define a unified model that can help us to harmonise all the different testing resources testing the results so and how we can what we need to do is we need to do the mapping of the data that each testing source follows to the reference to this unified data model that I described so each data source follows its own data structure has its own metadata uses its own formats for units for dates for vocabularies for countries for hazards and for products and many other parameters we need to apply a mapping process that will ensure that each of these elements from the original record will be mapped and will be converted to the elements of the reference model and yes technology is very important but the question is if the technology could help in something like that yet there are semi-automated mapping tools that can significantly help and make easier this process so it is a difficult process the mapping it is a difficult process but it can be supported by technology and by even new technologies like AI as some specific example so let's start with the example of extracting the metadata and the values for each metadata element from the certificate of analysis the favorite files the favorite files so there it is even more challenging because there are totally different layouts of these certificates of analysis as you can see here there are elements that are used to describe the same things like for instance which was the parameter that was analyzed if it was aflatoxing or if it was a physical parameter and this and even the results and not only the values not only the metadata elements may be described and may be available in different languages for instance the corresponding the relevant elements and then we still have the challenge to map and transform everything to the reference model and even if we go to the easier case of do this kind of do this kind of harmonization for files that come from systems that we are using like the laboratory information example of data of a data export from a limbs A and from a limbs B you can see I'm highlighting here the column of the food that was analyzed and even the name of the element is described in a different way the values of course may be very similar in the final meaning like it's both in both cases it has to do with wheat or wheat flour the way the values that are used and the language that is used to describe this element are different so these are the challenges that we see also in practice when it comes to the harmonization so which are the what we can do and which are the typical steps that we can follow in a repeatable and reliable way so we can harmonize the different data testing results and the different resources so first of all we need to collect all the testing results in one place in one store this may be a data lake or any other technology that can be used and then after collecting and having all this in one store to transform the testing results from the original data model from the original data to the reference data model to the unified data model if we have different languages used in the testing then we can apply a step for the translation this can be assisted by machine translation or it can be it can be also post-processed by people to improve the quality and after doing that we need to map also the values from the vocabularies used in the original data models to the vocabularies that are used in the reference data models and after that in order to make them available we need to store and index this unified version of the lab tests and make them available in an easy way for the machines to get them and show them in dashboards and some good practices that I could share here from similar worlds that we have done so far is that the one thing to use the unified data model is a very important step it is also very important for the different vocabularies for all this different list of values that we have for different data elements to follow standard vocabularies and data standards and there are already standards to describe the food categories and the foods like the food x2 or the codex ZFSA classification there are also standard vocabularies used to describe for instance the chemical contaminants like the cast numbers and also it is important to use standard matrix systems to harmonize results because we need to transform the original units to the reference units so we can have something that can be compared in terms of values so these are some good practices and you talk a lot about the data layer and lots of work that has to go into the harmonization part but certainly a bit about where you see those put in practice in generating value so if we manage to harmonize all these different testing results that comes from internal or external data sources because we have both cases then we can have some very interesting use cases enabled so one of the use cases that I want to share with you the first one is that we can interact with risk assessment over aggregated external testing results in this case we can harmonize and aggregate millions of sample analysis results from sample analysis from many external data sources and there are such data sources already these are mainly results testing results surveillance results published by the authorities we can link them to search based on the metadata taxonomies that we have discussed so far and this will enable to have real-time analytics and interactive navigation per commodity hazard and geography so it is easy then to answer questions like which is the percentage of samples that were found to be above the regulatory limit for a contaminant like lead in health and spices or in specific material like cinnamon so this is the one use case the second use case is that we can combine both the external harmonized data with the internal with the company's internal harmonized testing data and we can have we can use the harmonized and aggregated millions of testing results to link them to search facet that can be used to search and have real-time analytics per commodity hazard and geography so in this case we can unlock even more value because we are taking advantage both the internal the company's internal testing data from the different testing sources but also we can see what the external lab test results are and which are the trends and emerging issues identified there so I see that this builds on the first use case and provides even more value and the third use case is focus on delivering fully the fully harmonized data streams of testing data of the combined and harmonized to internal dashboards and platforms so we have all this disconnected data testing data either these are files, databases, APIs we can collect them securely store them and harmonize all this disconnected testing data with the process that we mentioned earlier and we can make all the fully harmonized data stream available through machine interface like an API and the API can be used to integrate this harmonized data stream in internal dashboards that can have different facets as we mentioned also in the previous cases per geography, per per parameter it can provide aggregated statistics it can identify trends comparable statistics and this can again unlock a lot of value when it comes to risk prevention so these are the three use cases I would like very much to hear also our panelists what do they think about the use cases before going to that because I will just share which benefits we see working with companies and harmonizing the data and unlocking the value of the testing data we see the benefits which are applicable to all the three cases to be focused in three areas first of all the same time the efficiency we are putting a lot of manual manual effort for processing combining all these different files so saving this time is something very important and we can devote this time and focus more on the decision-making part the second thing is that we can use all this data harmonize data to identify early food safety risk trends in commodities and key commodities and with using this emerging issues, this emerging risk to inform the risk assessment and the prevention and the third benefit that we see something that Nicos mentioned at the beginning of this webinar is that we can reduce the blind testing because we can focus more our testing in the areas where we see that there is an increasing risk or an emerging risk so these are the benefits but you describe three different use cases external resources, testing results internal the combination of the two and then presenting them in a platform or serving them in a harmonized format in an internal system that will use them which one of the use cases is more relevant for panelists Alex, what do you think? What did you hear? I love the idea of creating that common definition or that common data model we have all these languages that we speak from a data process perspective but what is that single definition that we're all going to rally around to say this is how we define say a test result so we all have that common footing now that we can sit here and analyze and we can write analytics off of, to me that's key that's cool so I love that component of this piece here so to me a lot of these pieces are, I don't know the term on far fetch there's some here that are to me low hanging fruit and some are just more complex to execute if we utilize the first use case on what is a risk out there that we need to identify and rally around a lot of times we can look at that and pull that into a singular data model to what Yanis was talking about easy to define but harder to process off of so to me that's probably the biggest opportunity I see is starting to tap into those data sources and then start to drive our internal processes off of the next piece is really for me number three and starting to create that cohesive canonical model inside start harmonizing all of our internal data for me that's very doable in my experience it's going to be complex which is no issue but we can do it with common technology and what we have today process creates data and the data drives the process so it's as much as harmonizing our processes it is harmonizing that data so that we can work in a common way produce common data against those canonical models which is things that we've done in the past that I've seen be successful and then how can we to me then the second use case is how can we bring together the best of those two worlds we have the common data out there from regulatory perspective to bring that together into a cohesive analytics strategy and start to drive insights from there and what I've taken away from this and just listen to be honest what are some of the notes I've taken that's interesting so I hear you highlighting the value potentially arriving to a common language and making sure that the processes that are generating data and the systems that are generating data they follow these and they speak into this new language absolutely it's a handshake yeah data without process is dead we need both in order to create that it's not a process yeah it's going to be tough it will have data without process so what do you think yeah I think that Alex is pretty thorough in his explanation of what digestion of what Diana's kind of art presented I think when I look at what disparate users contributing to data there's overlays with environmental data that may have been collected for other people that were kind of at that culmination point of you had to start and so we were starting and now you're starting kind of assembling the cars you're driving down the road and you're realizing maybe this needs to be improved upon I think the emphasis on how to get consistency as well as identifying how to get consistency amongst different groups so while Alex may represent one company they probably have one goal they have consistent expectations or at least you'd hope right and cross all their different entities the program we're running is an industry and the outcome everyone's looking for is to reduce to get there and the testing programs designed for people's programs are all different and I think that's so one is to be intentional about the language that you use and what you ask from your laboratories but then it's also about the method information of how those tests are that data is generated and what transparency do we have on that and how do you get them realizing to some extent data the questions that we're now leading to having had that or how do we get more clarity and more information and more intentionality about testing collection and data collection I loved Alex's data is dead if you're just kind of collecting it randomly and that's really true is that if we don't understand the context of a data point it will tell you anything you want to do and that's the threat I have where I see we're kind of at this I'm lovingly calling it the messy middle of going from an industry that in large part received testing results as the finality of something my product's good I'm shipping it save it away moving on to a culture that's trying to use the data system or risk this is particularly appropriate when you look at outdoor bag production is you're outside you need to identify when risks are increasing is if you're in that middle phase you're still reacting like it's a final point but you're also aware of the need to start looking at data and aggregate the messy middle part is figuring out and that is identifying some of the weak points within our kind of food testing culture in the sense that like we historically as users of testing in the food industry ask for a salmonella test or a stack test or an O157 test what we haven't asked is all of the other data associated with what generated that data point what was the sampling methodology what was the incubation time we've been really happy with it was not detected we're leaving on the table a lot of information about our testing point we are leaving on the table a lot of information about the way the testing is performed and I hear you describing something that I like a lot that to do this we have to be intentional in the way that we treat what we generate in terms of data so that we incorporate more information than the one that we currently have and that at the end of the day the systems will support this transition in the middle space the process that Alex was describing Yeah exactly I mean I think it's the intentionality and the discipline to recognize what a data point means and then to the point of just creating data to create data that's not very valuable so how do we re kind of frame how we think about testing to be like I want a compliant test result too I want a test result that teaches me something that changes my behavior in the future that's a very different goal or expectation of a test and I think we're starting it's a discovery right perhaps this finite set of resources when it comes to testing and really what you want to do is learn from your test so that you can change your behavior okay but historically we've used tests to just confirm what we already knew and those aren't the same goals and how you design your program where you forward have to change because of that So I hear you both describing your journey and Yanis I know that you have lots of experience in how this journey looks like what do you hear from what does it resonate with you and where do you see some challenges ahead Yeah so I hear very interesting points and things that it's true that I have seen them every time that we are trying to solve this harmonization aggregation challenge so I will I very much what I hear is that I hear very much that we we need when it comes to harmonization aggregation we need to start from the design so we need to start by adopting when we design our systems our approach our workflows the data workflows we need to design them following and using data processes and also I would say data standards I would add also the data standards because the one thing is to have the data processes I fully agree data without processes is dead but also even if we have processes but when it comes to the vocabularies we are using different terms of variation of the terms we are not using data standards I will say only the example of the country if we are not using the ISO list for the countries and we are using our own list for the countries this is a small obstacle but it's still an obstacle that will create delays in unlocking the value so I I hear that I hear also very much that we are leaving a lot of information a lot of data when it comes on how the data is generated so the provenance the process the methods that are followed the conditions we need more elements to describe in a correct way in a fully in a complete way I would say the data to get the value because it's not only and it happens very very much when you are collecting data from available public available spaces because you have only a few of these data elements and it's not possible to extract meaningful conclusions lots of value that we have because we don't have all the information and all the process and this is so we need transparency there I agree very much what can current technologies do we can I think that I would agree very much with Alex we have we have technologies that can solve and can address many of these challenges and we have now even more mature AI technology that can help to especially in a semi-automated way to help us in the mapping which if you have thousands of terms of even hundreds of thousands of terms this may take a lot of time to do the mapping the technology there can help us we have the technology for the data processes I think that knowing very well this space for more than 10 years I still feel that we lack some of the data standards so for instance a data standard for all the potential hazards that we may have in food it's still missing that is adopted globally it's still missing or it's different between the US and in Europe in Asia for instance so this is still a challenge standards data standards are missing the common language and then this brings in mind the food safety standards but didn't achieve didn't arrive at the common language but at least they have a benchmarking mechanism so that they can translate from one standard to another music in my ears but we are we only have a few minutes left I will I will skip I have to skip the poll for the audience I'm really sorry guys but I give you the opportunity to send us a bit of feedback I have a pressing question here that I have to ask you guys we talked about time when we're testing data generated and how relevant they are in real time or in retrospective so how important is real time and insights and information for you what do you think Alex could you like to start for me the earlier I understand the environment and test results the more proactive I can be around the operations so it's always helpful to see something before it happens rather than reacting to it after it occurs so of course I need to understand it after it occurs so I can do the appropriate capital root cause analysis and correct but I always prefer to have that data and information upfront so that we can be proactive about our risk rather than reactive about our failures upfront means being able to be more proactive what about you you were describing an exercise that is quite that introduces quite a delay in getting all the data together and selecting insights how much does real time matter real time is what everyone wants so that they can do exactly what Alex just described I want to know in advance when risks are increasing such that I can modify my behavior and prevent it from happening that's what everyone is going towards so I'll provide maybe a different context as well to the value of retrospective data is that one of the things that is challenging especially within food systems is that you can't test every portion so you are making assumptions about prevalence you're making assumptions about what your testing program statistical limits are able to detect if you have the ability to go in retrospect and see what happened you can better inform data collection going forward so maybe my risk has gone down and I actually need more testing than I used to use or maybe it's gone up and I actually could use less or if I approach it from a different perspective so even though the Holy Grail that everyone is trying to get to is real time ability to be predictive and what not and that is most certainly the purpose of our program the learnings from retrospective analysis can't be overlooked as well even from the sense of that they can help define and characterize the risk of an area as well as better inform your data collection going forward so I'll put that in even retrospective is important and I have another interesting question and that's the only one that we can answer and I think Jan is our last priority on that it has to do with costs because harmonization for example mappings can be a very expensive exercise especially if people have to do this manually or if you need to hire someone external to do this for you how can we use technology to reduce this cost and make the process more efficient what do you think Jan's so I mentioned something about that so I will elaborate it a bit more yay it's true that the mapping for many many different terms used in many different data elements metadata elements can be very costly exercise our approach and what I see that can be of help to reduce and to make it more efficient is first of all a very good collaboration between the teams that know very well the data and the technology teams that can apply tools to facilitate this mapping then we have now the text mining tools that we have and the AI technology can help us very much to do in a very fast way and quite efficient way a first mapping first version of the mapping and then we still need the human validation to make sure that we have accurate mappings the automated mappings may have 70% accuracy or may have even 80% accuracy it depends on the original data and the original terms and how much of the terms and training data we can have but yes this is one of the things that we can do to facilitate this mapping so we can use tools to automate part of the process but these tools have to be applied hand in hand with the data experts the people that understand the data that's what I hear you say and the domain experts which they know the process is how the data was generated all these things are very important I know that you love this topic we could talk for hours but we have to wrap it up because we promise to keep this going for an hour more or less if you choose one highlight from our conversation one thing that you kept, you are keeping as a takeaway or that you learned today what will this one thing be Joelle, would you like to start? I think common language is really important as we proceed forward with data collection as an industry the common language Alex? yeah I think that was a key key common language and another piece I've kind of taken away from this is the data standards that kind of go along with that scalable repeatable common language means data standards not what I hear you say Giannis? the one thing that I will keep is the great opportunity that I see of unlocking the value of all this data even if there is we need to make the step of transforming it and following a common language this I think that it's worth it to do it because we can unlock a lot of value so there is a way to unlock the value we describe some of the ways today this is another ancient slide that I had to dig out from my archive because these are not new problems that we are talking about they are very very old problems yes we do need a common language but the books are going to keep on coming in different formats described in different ways so we have to deal with this reality of putting in place harmonization pieces to help us arrive at the maybe not a common language everyone to be speaking of a common language but at least use a common language when we are interpreting what we hear from many many people and that's my key takeaway it was a pleasure to have you with me thank you so much for joining this very interesting conversation I'm inviting our audience to take a look at one of the dashboards that we are developing that are putting together chemical contamination test results and see how they look like and then give us some feedback to help us improve them any last comments closing comments thank you for the opportunity Alex you likewise great conversation today thank you Janis thank you so much and again it was an honor to be with such a nice panel and to have a great conversation the honor was mine and it was a pleasure as well thank you and thank you to the audience that joined us today bye bye bye bye