Пълният препис на Predictably Good - Conversation with Brendan Ring and James Doyle

00:02:28.560 What did you notice? 00:02:28.639 --> 00:02:30.240 Was the trade floor as busy as ever? 00:02:30.560 --> 00:02:32.000 The trade floor was pretty busy. 00:02:32.800 --> 00:02:37.840 I had been to IFT for a couple of years, so it was great to be back again this year. 00:02:38.000 --> 00:02:39.680 Lots of familiar faces there. 00:02:40.639 --> 00:02:45.599 I did hear a couple of people say that the trade floor wasn't as big actually as other years. 00:02:46.560 --> 00:02:55.840 So there's definitely a lot of a move towards technology rather than it kind of often competes with IAFP, which is very much around food safety. 00:03:18.800 --> 00:03:24.159 So yeah, definitely a strong move to digital and um, you know, and obviously innovation. 00:03:24.319 --> 00:03:34.800 Obviously, it's it they're look the you know it maybe it sells the legs into having the more high-tech uh types of companies that come along to this because they're showcasing their wares and people come to see the latest tech there as well. 00:03:34.879 --> 00:03:39.439 But uh um, but yeah, no, a lot a lot of a lot of cool data in AI as well. 00:03:39.520 --> 00:03:39.759 Yeah. 00:03:40.000 --> 00:03:46.159 Yeah, I always enjoy the trends that you see on display, the different flavors and ingredients that people are promoting each year. 00:03:46.240 --> 00:03:52.159 Did you notice any new trends and flavors or ingredients that are uh types of ingredients? 00:03:52.400 --> 00:04:01.120 Yeah, like I suppose over the years, you know, you would have seen a lot of um maybe you know, moving away from say animal-related uh products and that kind of stuff. 00:04:01.280 --> 00:04:03.360 That seemed to be less um in this case. 00:04:03.520 --> 00:04:09.840 I think it is more like um definitely clean label, um, you know, a lot of natural ingredients. 00:04:09.919 --> 00:04:13.439 Um, you know, definitely that was a very kind of popular theme. 00:04:13.520 --> 00:04:21.519 You know, most of the more advanced high-tech companies definitely had a a strong uh green or clean presence, I would say. 00:04:21.680 --> 00:04:21.839 Right. 00:04:22.000 --> 00:04:23.920 That was kind of the biggie from the ingredients side. 00:04:24.240 --> 00:04:30.399 Even the ingredient providers had some kind of digital offering or backup to their offer, their services. 00:04:30.480 --> 00:04:31.279 Is that what you're saying? 00:04:31.519 --> 00:04:42.639 Yeah, like it's I suppose, you know, I guess when companies are looking to buy ingredients, you know, to web in it that that's the first place you're going. 00:04:42.720 --> 00:04:54.560 So if you don't have the right tools for them to go and search for this and get their appropriate specs and all of that, um it's you know, you're not going to find the information that you need in a timely manner. 00:04:54.720 --> 00:05:10.480 Uh, so a lot of the the larger companies, you know, the the way they're able to display their wares as such, um, and you know, doing

that digitally as much as um in in the real world, uh, was you know, definitely the the tech was really up there, you know. 00:05:10.560 --> 00:05:20.079 Um If they had their own tool there as well in terms of MPD, and um there's you know there's a lot of new tools around that, and we can we can talk about that as well on our own work in that space. 00:05:20.160 --> 00:05:38.560 So definitely MPD, sorry, new product development, um, and helping companies to develop new product and looking at those those trends and um you know how how do you search that you know really rich data source that's out there um and use that every company is trying to be more sophisticated with data, yeah. 00:05:38.720 --> 00:05:47.199 What how they do business all at the end of the day, they're selling an ingredient and a very slit, sometimes a very uh novel ingredient, yeah. 00:05:47.519 --> 00:05:49.519 With different different capabilities. 00:05:49.680 --> 00:06:03.279 So I understand uh you mentioned AI, we'll come back to that, but you were over there and you pitched um a new NPD data trending capability with a group of potential buyers or investors. 00:06:03.360 --> 00:06:04.000 How did that go? 00:06:04.240 --> 00:06:12.639 So, yeah, so that was um an Enterprise Ireland organized event, um, kind of like Dragon's End or Shark Tank, maybe more familiar in the in the US. 00:06:12.800 --> 00:06:15.199 Um, there was a great panel of speakers on it. 00:06:15.279 --> 00:06:25.920 Um, we had um yeah, we full range of uh three Irish companies essentially there that were uh talking about their different AI products and that. 00:06:26.000 --> 00:06:29.600 So we we were delighted to have the opportunity to talk about some of the stuff that we're doing. 00:06:29.759 --> 00:06:37.759 So essentially our on the MPD stuff, it's it's uh we're using AI to pull in a lot of data, um, mostly pretty messy data. 00:06:37.839 --> 00:07:08.079 Um so using AI to actually find the right data, and then using AI as elements of helping to clean up that data, um, but you know, in a lot of cases writing code to do that in a very kind of robust and traceable way, and then being able to visualize that data and and understand and spot trends, and you know, because you have you have such a vast amount of data, you need to be able to group and traditional technology, uh like Python code or whatever struggles with that kind of aggregation unless it's very, very sophisticated. 00:07:08.160 --> 00:07:10.160 So, but AI can do that really, really well. 00:07:10.319 --> 00:07:18.560 So we're using all of those tools together, and now um industry, um, they sort of were talking to companies, they are talking to others, um, we're talking to others. 00:07:18.720 --> 00:07:24.480 So, what they really love about it is being able to have traceable data, so proof. 00:07:24.639 --> 00:07:33.839 So, rather than just saying, Oh, here's an interesting trend, here's an interesting trend, because this, you know, this is the volume of data that uh people are looking for here. 00:07:34.000 --> 00:07:38.000 This is how this has been, you know, and these are in real counts and real values. 00:07:38.240 --> 00:07:41.120 So that brings so much credibility. 00:07:41.279 --> 00:08:02.399 It and obviously, when we have the team that we have here in terms of being able to understand the science of the data and the nuances of what that flavor, what those compounds mean, what those ingredients and how those ingredients are used to produce different products, when you put that validation around it, now you have something that you can you can trust to make those decisions on. 00:08:02.480 --> 00:08:03.759 And that that really resonated with the thing. 00:08:06.959 --> 00:08:10.000 I suppose the future that they hope are going to be popular, right? 00:08:10.240 --> 00:08:10.560 Exactly. 00:08:10.879 --> 00:08:11.199 Okay, right. 00:08:11.360 --> 00:08:11.759 Yeah, yeah. 00:08:12.800 --> 00:08:13.360 Excellent. 00:08:13.519 --> 00:08:15.120 So data and AI. 00:08:15.279 --> 00:08:22.079 Were there any other like was it AI literally everywhere across the three different events from Food Safety Summit to IFT? 00:08:22.319 --> 00:08:23.439 And AFTO as well. 00:08:23.519 --> 00:08:27.920 So AFTO, just for people to understand it, so it's the Association of Food and Drug Officials. 00:08:28.000 --> 00:08:32.240 So essentially it's it's a they had their 130th, which is amazing. 00:08:32.399 --> 00:08:36.320 Um 130th annual education conference, they call it. 00:08:36.559 --> 00:08:47.919 Um, so this is essentially mostly regulators, probably 80, uh maybe 60 percent regulators in the in the audience, and then industry and consultants and various others. 00:08:48.159 --> 00:09:07.759 Um, so um yeah, I'll talk about one case study there, but just a on on the Food Safety Summit, um, which is a quite a large conference in Chicago, um, the keynote, um, which I thought was really interesting, um, turned into an AI workshop. 00:09:08.000 --> 00:09:14.960 Um, so this is with maybe I did maybe 4,000 people in their main room, the main conference room. 00:09:15.200 --> 00:09:19.360 And so there's a bit of, you know, and a bit of talk, and then you know how AI is is coming. 00:09:19.440 --> 00:09:28.480 And, you know, um really I suppose what people see is like the rate of innovation that can now occur because what AI can enable. 00:09:28.639 --> 00:09:46.480 So, you know, sm you know, it's so much easier now for even individual business units to to develop new and novel uh things that maybe you know you might need a massive software team or a data science team to do before you can do so much more now um with you know with the tools that are available. 00:09:46.639 --> 00:10:03.519 So, yeah, so that workshop or that kind of keynote was like being led by who were handed out a uh sheet um and a couple of challenges on it, and it was the objective was use AI to solve these, and there are food

safety related um challenges. 00:10:03.679 --> 00:10:08.799 You know, how would you write uh um an SOP or you know, all these different questions that you could answer? 00:10:08.960 --> 00:10:16.240 And then people stood up and talked about their experiences of it, and then other people stood up and talked about the challenges that they have using AI. 00:10:16.320 --> 00:10:27.279 And obviously, the you know, a lot of questions around how can you trust the data and how can you trust results, and you know, there's a lot of discussion around how to ensure what you're doing can be validated. 00:10:27.360 --> 00:10:29.120 So that was that was really a big topic. 00:10:29.440 --> 00:10:33.519 You know, of course, there's people you know saying, Oh, it's going to come and take our jobs. 00:10:33.679 --> 00:10:39.679 Um, you know, for sure the strong message that that came across is like it'll take your job if you don't embrace it. 00:10:39.919 --> 00:10:49.120 Um so you know, really just like yeah, you know, a little bit of FOMO, AI FOMO, as we as we say here, um, how do you uh uh you know embrace it? 00:10:49.440 --> 00:10:50.080 Embrace it. 00:10:50.320 --> 00:10:59.360 And you know, uh some companies have a lot of kind of restrictive policies in terms of allowing them to use it, and that that you know, that was a strong theme at AFTO. 00:10:59.440 --> 00:11:11.519 Obviously, it's a regulatory environment, and you know, if it's not done the right way, it can for sure you can leak data, and that's you know, that's uh that's their kind of suckable offenses, so you don't want to uh and get yourself down that road. 00:11:11.759 --> 00:11:36.879 Um one interesting case study, if if I have a moment to talk about it, um, is uh at AFTO they had um there are uh a couple of other speakers from um uh one in particular is from I think it was the city of Boulder, and they were talking about an example of how they're using they've built an AI system and nothing to do with food or anything, but I thought it was really interesting nonetheless. 00:11:37.039 --> 00:11:57.120 Um so the police force there wear cameras, they wear body cameras, and that's analyzing what's happening in a situation, um, and it's giving them real-time feedback as so based on the regulatory, based on the situation, um, and based on what's uh what's their next best action to take. 00:11:57.279 --> 00:11:57.440 Right. 00:11:57.600 --> 00:11:59.440 And it's doing that in real time. 00:11:59.759 --> 00:12:04.879 Um, so really, really impressive and really high pressure situations and really obviously central data. 00:12:04.960 --> 00:12:08.799 So you can imagine the number of hurdles that they have to get over in order for this. 00:12:08.879 --> 00:12:18.879 Now, obviously, the human makes the decision at the end of the day, but like as a as you know, on the police force, it's complex to know what you're allowed to do. 00:12:18.960 --> 00:12:21.759 You know, you and there's no time to go looking

up the manual. 00:12:22.159 --> 00:12:24.639 You need to know your stuff, you need to know or you need help. 00:12:24.720 --> 00:12:41.600 Yeah, so you know, you could like if it's getting proven out in that case study, you could see how that might work in food safety and people who aren't working in different operational jobs, yeah, you know, yeah, advising them on stop to stop there, redo that. 00:12:41.840 --> 00:12:58.399 So, like you you kind of wonder is the day of of um you know old school learning is that starting to fade away, and a little bit more of like guidance, you know, like got real-time guidance relative to the scenario, is more and more of that going to start coming in. 00:12:58.559 --> 00:13:03.039 Um, so yeah, that that's uh definitely seemed to be a hot and interesting area. 00:13:03.360 --> 00:13:04.159 All right, interesting. 00:13:04.240 --> 00:13:22.080 Was it well received by the audience or were they freaked out by the things that yeah, definitely it was a definitely a conservative um group of people, but they they also saw the credibility of who was doing this, and it's like, what if they can do it? 00:13:22.320 --> 00:13:23.039 Yeah, it's inspiring. 00:13:23.279 --> 00:13:37.120 You know, we need to be, and you know, some of the actually I I got planked afterwards because of what one of the um one of the questions that came up was like, you know, in in terms of like is the ways to use AI without, you know, kind of breaching anything or doing whatever. 00:13:37.279 --> 00:13:50.639 And one of the one of the things we suggested was like um say in a spreadsheet, sometimes the formulas you might need for a spreadsheet are like, you know, they're really complex, and it could take you like a year-long Excel course to learn some of these things. 00:13:51.039 --> 00:14:11.600 Put what you want into a chat and ask it to give you the Excel formula, and that on its own, you know, you've never put any data into chat or you haven't, you know, exposed anything, but like the the complexity of a formula that you can build in in um some of the the tools is really exceptional. 00:14:11.679 --> 00:14:19.440 And they're like, oh my god, that has been you know, that that even just that advice alone for some people was was uh hugely useful, hugely helpful. 00:14:19.600 --> 00:14:20.399 Yeah, yeah, exactly. 00:14:20.480 --> 00:14:22.159 And people definitely have an appetite. 00:14:22.399 --> 00:14:32.879 Yeah, and I know we're using it to prototype all sorts of systems that would be very time consuming to build, and you can literally prompt them into existence in hours, right? 00:14:32.960 --> 00:14:34.000 We were talking about that earlier. 00:14:34.240 --> 00:14:34.720 Yeah. 00:14:34.879 --> 00:14:37.360 Um, so let's change track a little bit. 00:14:37.440 --> 00:14:51.600 So you probably met a bunch of customers uh at and different uh types of organizations, and and some of the themes we're seeing across all the campuses are that they do possess a lot of data, scientific, supplier data, operational data. 00:14:51.759 --> 00:15:02.080 Um, but from those conversations, what what what were the barriers or were there barriers they were just talking about that was preventing them getting it into insights or decisions? 00:15:02.480 --> 00:15:18.080 The big one really, I think, and I suppose um as we've seen, we've kind of presented at different conferences, and as we've been talking to customers and all of that, the economics of accessing messy data didn't add up before. 00:15:18.159 --> 00:15:18.399 Yeah. 00:15:18.639 --> 00:15:19.360 Now they do. 00:15:19.519 --> 00:15:19.679 Yeah. 00:15:19.919 --> 00:15:22.320 Um, that's been the really dramatic change. 00:15:22.399 --> 00:15:36.159 You know, you talked about you know companies and lots of different data and some companies use a lot of like PowerPoint or some use even spreadsheets, but spreadsheets can get very messy very quickly, um, even though it's somewhat structured, but you know, when there's free text and all of that. 00:15:36.399 --> 00:15:41.519 Um and you know, obviously documents and SOPs and and all of that type of stuff. 00:15:41.600 --> 00:15:50.240 Like they the regular data that you might get pulled off a factory floor, that's being like, you know, that's kind of 10 years old now, being able to pull in data from sensors and all that kind of stuff. 00:15:50.320 --> 00:15:52.080 That that's really trivial now, you know. 00:15:52.159 --> 00:15:56.879 So, like, yeah, you can have high volume there and even high velocity. 00:15:57.279 --> 00:16:06.879 Um, the one that was always the Achilles heel of data science is variety, you know, when you consider those three Vs, volume, velocity, and variety. 00:16:07.360 --> 00:16:18.879 So with the advent of and and it's not just using like LLMs to process that data, it's using even LLMs to write the code to process that data. 00:16:19.279 --> 00:16:22.080 How do you build that pipeline of data engineering, maybe? 00:16:22.399 --> 00:16:22.799 Exactly. 00:16:22.879 --> 00:16:29.919 Yeah, and then being able to scale that like the cost of computing has like dropped so much, and the cost of storage has dropped so much. 00:16:30.080 --> 00:16:48.320 So, you know, what previously may maybe have been stored now when actually some of the projects we've done for some of our customers, like uh 2000 MSDSs, and so rather now than having like each one of them maybe 60 pages long, um, so having all of those long you know documents, you could imagine if you had stacked all of those pages, that'd be pretty long. 00:16:48.480 --> 00:17:00.399 So we've turned that now into tables, I think maybe you know, with hundreds of columns, with a row per, or actually with sometimes multiple, multiple rows per, but computers and all that so easily. 00:17:00.559 --> 00:17:16.799 Um and now you can visualize and get insight from that and can ask questions and filter things and you know process it in such in

a way that was just impossible before, and you know, as purely in the last two years that it came economically viable to do that. 00:17:16.880 --> 00:17:21.119 So that's really what imagery is the barriers are disappearing in that form, yeah. 00:17:21.440 --> 00:17:22.559 Barriers are disappearing there. 00:17:22.960 --> 00:17:24.960 Yeah, okay. 00:17:25.119 --> 00:17:30.799 And um, what are the customers actually asking Crem Global to do now or to deliver now? 00:17:31.039 --> 00:17:35.119 Are they asking for AI early warning to help integrate data? 00:17:35.359 --> 00:17:38.160 Yeah, or specific business problems. 00:17:38.799 --> 00:17:41.839 So all of that data and insight, that's all very fine. 00:17:41.920 --> 00:17:43.200 What are you going to do about it, really? 00:17:43.279 --> 00:17:44.079 So that that's it. 00:17:44.160 --> 00:17:48.960 So that's you know, really where the whole kind of value proposition steps up. 00:17:49.119 --> 00:17:58.400 So when you want to start getting, you know, when you can tabulize tabulate that data, now you can actually start to build models and build, you know, engines to learn. 00:17:58.480 --> 00:18:15.440 You know, we're talking about this earlier in terms of like, and you can timestamp that data and like blind yourself to what that would have looked like in the future, and then allow the models to predict what might happen in the future, and then you can move forward, you know, step by step in time and start to see what did actually happen. 00:18:15.519 --> 00:18:17.759 And you know, when you have all that retrospective data. 00:18:18.000 --> 00:18:29.759 So that's really the exciting part, and that's what customers are coming to talk to us about now, in terms of like, now that we've all of this data and we have our data house in order, what are how can we get into more predictive and preventative? 00:18:30.000 --> 00:18:36.319 So whether it be on the food safety side, whether it be on the you know, what's the next trends coming from a food food perspective? 00:18:36.559 --> 00:19:00.480 Um, you know, obviously work across you know, cosmetics and with work with regulators and work, you know, you know, quite quite a broad range of um of customers and use cases, but you know, the recurring theme is always, you know, now that we can you know get our data in order and then start using that for like you know, really predictive preventative. 00:19:00.880 --> 00:19:14.160 It's like it's kind of exciting times then because a lot of those hurdles that existed now have been you know removed, and now people are really looking to make the most of the data predictive, whether it's for safety or for new business opportunities. 00:19:14.319 --> 00:19:15.519 Yeah, so that's exciting. 00:19:15.680 --> 00:19:25.039 So it sounds like you know that things have moved forward in the last 12 months substantially, really, in terms of the community's awareness and maturity. 00:19:25.279 --> 00:19:46.160 And the it has moved

well has moved a lot in the last 12 months, but uh even what's more remarkable is the rate of change as I think the world has never seen anything like this, so yeah, that's uh so we tend to attend these conferences every year so we can kind of judge the changes here, but you're saying you're seeing the rate of change accelerate. 00:19:46.319 --> 00:19:51.279 Yeah, and uh IFT is a great place to steal that energy and the buzz. 00:19:51.440 --> 00:19:54.400 And safety summit is excellent for food safety. 00:19:54.640 --> 00:20:01.839 Um, and I know in um IAFP, which I'm gonna talk about with James, people were talking a lot about data culture for food safety. 00:20:01.920 --> 00:20:11.440 So now that the tools are there and the analytics are there and the capabilities are there, what kind of culture can you develop to actually make sure that you're able to use that property? 00:20:11.599 --> 00:20:13.759 So I will get into that with James. 00:20:13.920 --> 00:20:19.279 Um final question then, looking ahead, looking ahead another 12 months, then or even 24 months. 00:20:19.440 --> 00:20:23.680 Yeah, how do we uh what will change um in materially? 00:20:24.000 --> 00:20:30.559 What do you think will materially change on how companies manage risk or seek out insights for business opportunities? 00:20:30.880 --> 00:20:36.880 Uh I think um obviously AI is central, it's it's for sure, unless there's some something crazy happened. 00:20:37.200 --> 00:20:45.119 Um, but I think the biggest limitation right now with AI is you know not being able to trust the output. 00:20:45.200 --> 00:20:45.359 Yeah. 00:20:45.599 --> 00:20:49.440 So I think that's where most effort is gone into in terms of being able to validate. 00:20:49.599 --> 00:21:08.640 Like you can even see the leaps that has been that has been made, you know, in the last probably might know 10 months, you know, uh when you ask questions, there you know, you get a lot of like references and citations, and you know, it's you can you even get links now, so you can click on that and go and have a look, you know. 00:21:08.720 --> 00:21:11.279 So even that, you know, those small things on its own. 00:21:11.440 --> 00:21:31.279 Um, you know, obviously we use it, you know, in a lot of cases to write code, and then that code can be validated, and that's a lot more kind of reversible to be able to and traceable things like code, test it, it's doing what it's supposed to do, yeah, under various and it'll it'll come up with all these edge case scenarios that as a human you could never easily take up. 00:21:31.599 --> 00:21:37.920 Yeah, so um so can you to help test your systems and build trust in your systems, yeah. 00:21:38.079 --> 00:21:38.960 So yeah, yeah. 00:21:39.119 --> 00:21:40.880 So that's a good place to leave it. 00:21:40.960 --> 00:21:43.200 So uh thanks so much, Brendan, for coming in. 00:21:43.440 --> 00:21:44.240 And uh pleasure. 00:21:44.559 --> 00:21:53.359 Looking forward to like revisiting uh Food Safety Summit, um the AFTO annual education event and Ift again next year, I presume. 00:21:53.519 --> 00:21:54.400 Yeah, very good. 00:21:54.559 --> 00:21:55.599 Good, thank you. 00:22:00.799 --> 00:22:01.680 Welcome back. 00:22:01.920 --> 00:22:05.279 Brendan has given us the commercial and customer perspective. 00:22:05.440 --> 00:22:20.799 I now want to examine the same shift from a scientific and regulatory side with James, who's attended IAFP with me, as well as the Government Chemical Conference in the UK and the FIN annual meeting and the ACS fall meeting. 00:22:20.880 --> 00:22:22.559 So and you travel as ever James. 00:22:22.640 --> 00:22:23.119 It's been busy. 00:22:23.200 --> 00:22:23.599 It's been good. 00:22:23.839 --> 00:22:25.200 Yeah, it's been great actually. 00:22:25.440 --> 00:22:27.920 I really enjoyed the IFP meeting. 00:22:28.000 --> 00:22:31.359 I always do the really good scientific conversations. 00:22:31.680 --> 00:22:39.440 You know, um what's what struck you from IFP this year in general and across the different scientific uh sessions and things? 00:22:40.960 --> 00:22:52.799 What was wonderful to hear, and I've seen this sort of evolution of IFP over time, is that there's a real shift towards people accepting or putting data at the heart to decision making. 00:22:52.880 --> 00:23:09.759 And so there's this sort of demand we see both in new methodologies, lab data, time I think is a key one, key factor, as in I need to decrease the time to decision making, and be that with rapid testing or predictive modeling. 00:23:10.079 --> 00:23:21.119 Um I think there is a huge shift where that is just accepted as a necessary tool and a critical tool in moving forward for food safety. 00:23:21.440 --> 00:23:28.480 Yeah, I think the whole data and AI has been discussed for a few years now at IFE and it's starting to mature that AI tools have got better. 00:23:28.640 --> 00:23:32.480 Obviously, people realize that AI is helping us organize our data better. 00:23:32.640 --> 00:23:34.559 So we have more data now. 00:23:34.960 --> 00:23:40.799 So what I noticed, and I actually was walking in my own section around data culture for food safety. 00:23:40.880 --> 00:23:47.039 And even in the keynote, they were talking about what were the what are the biggest risks to protecting global food supply? 00:23:47.200 --> 00:23:53.759 And the top three risks they identified were actually mostly human factors, and they accounted for more than 50% of the votes. 00:23:54.000 --> 00:24:02.880 And those are things like capacity, awareness, training, lack of commitment, the second one, and then the third one political climate, funding, government influence. 00:24:03.039 --> 00:24:13.680 So I found it fascinating in the world of AI, where we're dealing with AI and data, that's these human factors are still the highest of challenges for the industry. 00:24:14.000 --> 00:24:15.519 Yeah, and and so two things. 00:24:15.599 --> 00:24:17.039 So I I

totally agree with you. 00:24:17.119 --> 00:24:26.640 And what was very interesting to see in the in the symposia within IFP is that it wasn't uh AI wasn't carved out specifically as its own session. 00:24:27.039 --> 00:24:29.920 Because in the past it was nearly introducing a new technique. 00:24:30.000 --> 00:24:32.640 Yeah, it was just part and parcel of so many sessions. 00:24:33.440 --> 00:24:33.680 Yeah. 00:24:33.839 --> 00:24:39.599 Um so so that's maybe one point I thought, but and it's becoming so practical. 00:24:39.759 --> 00:24:46.640 And but as you mentioned with those sort of three constraints, uh is is it trust of the tool? 00:24:46.799 --> 00:24:47.119 Yeah. 00:24:47.279 --> 00:25:02.559 And uh so yes, the culture needed, like, and that's what I was trying to get into when my talk was how do you develop that right culture to trust the data, trust the models, and make decisions under uncertainty, where it's a very consequential like cost involved in in making those decisions. 00:25:02.720 --> 00:25:16.000 Um so I think like it's great, as you say, to see AI be more pervasive across all sessions, and then the sessions are more subtle around well, this is about data culture on AI, this is about data structures for AI, etc. 00:25:16.240 --> 00:25:18.880 So it's like you know, bringing it to the next stage. 00:25:19.119 --> 00:25:27.440 Yeah, and at no point has people people need from informed decision making. 00:25:28.319 --> 00:25:32.000 And it is never binary, you never have all the affirmation. 00:25:32.079 --> 00:25:36.960 And so we so you talked about embracing uncertainty in in a talk months ago. 00:25:37.119 --> 00:25:42.640 Um, I think it's a wonderful phrase, uh, but it's always been embraced. 00:25:42.799 --> 00:25:48.319 This is just another source of information to help better inform the risk. 00:25:48.400 --> 00:25:54.319 And as I said, often it's the time you need to be making rapid decisions on emerging risks. 00:25:54.559 --> 00:25:57.200 Um and that's what I I thought as well. 00:25:57.759 --> 00:26:03.440 You know, we're seeing a lot of people understand the food safety as a systems problem. 00:26:03.599 --> 00:26:03.839 Yeah. 00:26:04.000 --> 00:26:07.680 So it's I'm I'm not looking at one pathogen anymore. 00:26:07.839 --> 00:26:15.440 So I need multiple sources of information, and and and you know, things like good data governance, data culture, I think are key to it. 00:26:15.759 --> 00:26:22.319 So we're we're people are all very keen to detect incidents and even to predict predict and prevent incidents now. 00:26:22.400 --> 00:26:26.400 And there's a lot of talk and even demand for that, as you're saying, they want to move to that capability. 00:26:26.559 --> 00:26:32.720 Did you see any evidence that systems are moving towards the capability to actually predict and prevent? 00:26:33.119 --> 00:26:44.640 I I think um there is there definitely is a shift where you're getting closer and closer to a food safety operating model which brings all of these in. 00:26:45.359 --> 00:26:58.000 Um I I think people are still hitting the challenges of of good data governance, data management, of having data structured in a way that it can be utilized by all these wonderful machine learning and AI tools. 00:26:58.319 --> 00:27:07.200 Um my takeaway is that in the near term, it it's it AI is a decision support rather than a decision replacement. 00:27:07.519 --> 00:27:08.400 Of course, yeah. 00:27:08.640 --> 00:27:11.440 Um, and and that's what I've I saw. 00:27:11.680 --> 00:27:16.240 Numerous talks where people are bringing that in day to day as a decision support tool. 00:27:16.880 --> 00:27:34.400 And we talk about phrases like uh, especially with maybe staff turnover capacity issues, resource issues, that it helps enable decision making rather than training a new person up to get out to speed where they can break better decision um or better informed decisions. 00:27:34.799 --> 00:27:47.119 Um so I think that's where we see I see see a lot of people looking at um uh early signal extraction to inform decisions and support them. 00:27:47.359 --> 00:27:53.119 Validation is another key word that we saw a huge amount of people working on validating it goes back to your the idea of trust. 00:27:53.359 --> 00:27:53.759 Yeah. 00:27:54.079 --> 00:28:04.880 And it's it's interesting with AI and how powerful it's become, but that still that human has to make a decision at the end of the day or decide if the evidence is strong enough for the decision. 00:28:05.200 --> 00:28:08.880 So it's it's fascinating in all industries now. 00:28:08.960 --> 00:28:18.559 I've been talking to people in lots of industries around AI and they they're using it as a companion and a helper, but at the end of the day, we need to make the decision on what to do with that, and we're responsible. 00:28:18.640 --> 00:28:22.319 So it's it's it's really great to see that maturing and people understanding that. 00:28:22.400 --> 00:28:23.359 So great. 00:28:23.599 --> 00:28:38.559 Um, so which developments in measurement detection or data analytics uh do you think are producing better scientific evidence now from the conferences you've been to across chemistry and food safety, etc.? 00:28:39.200 --> 00:28:44.480 Um, what what what are you seeing that are improving, methods that are improving that help us with this evidence? 00:28:45.200 --> 00:28:48.000 So methods, lab detection. 00:28:48.079 --> 00:28:49.920 So there's a lot of emerging risk. 00:28:50.000 --> 00:28:57.599 We've you know, unfortunately, every year we talk about uh another big food recall, global food recalls and national food recalls. 00:28:57.920 --> 00:29:05.680 Um and so so there are always new methodologies coming on, low, always lower limits of detection. 00:29:05.839 --> 00:29:18.799 And I I do think technology like AI was probably the real big disruptive um uh influencer, a wee bit like um all genome sequencing and finally

understanding the power of of a technique like that. 00:29:19.039 --> 00:29:22.880 But I go back to that point around time, I think a huge amount about rapid testing. 00:29:23.039 --> 00:29:31.599 Yeah, rapid testing and uh trying to stay ahead of that curve to to move before you you get that incident, before you get that recall. 00:29:31.680 --> 00:29:42.720 And unfortunately, you know, WA Joe brought out the 2026 stats 866 million illnesses, 1.52 million deaths from foodborne illnesses annually. 00:29:42.960 --> 00:29:48.640 Um so I think that's what I'm seeing is a huge amount of focus on shortening the time to decision making. 00:29:48.799 --> 00:29:49.519 Yes, exactly. 00:29:49.599 --> 00:29:55.759 And these technologies helping with that rapid detection, rapid data analytics, decision support tools, dashboards. 00:29:56.000 --> 00:30:07.519 So the scientific sessions are always great at IFP, but actually, some of the value, real value comes from those coffee talks and chats over coffee at the at the trade center and at other uh receptions. 00:30:07.680 --> 00:30:08.480 What would you like? 00:30:08.559 --> 00:30:18.880 What was the kind of word on the street in terms of like some of the things that weren't being said maybe on stage, but you were learning from more one-to-one discussions? 00:30:19.119 --> 00:30:20.720 Yeah, and it's true, it's true. 00:30:20.799 --> 00:30:26.319 And yeah, I often we we may have speaking roles or running a symposium, which is wonderful too. 00:30:26.400 --> 00:30:28.000 I'd recommend anyone to get involved in that. 00:30:28.079 --> 00:30:30.559 That deadline's coming up actually um uh soon. 00:30:30.640 --> 00:30:37.920 So um uh both for IFP US and Europe, it'd be Copenhagen in 2020 setting. 00:30:38.400 --> 00:30:44.720 Um, so yeah, I'd recommend anyone to get involved in those sort of receptions. 00:30:44.799 --> 00:30:47.599 We go to uh session, which is wonderful actually. 00:30:47.680 --> 00:30:54.480 I think it's become a bit of an um uh uh IFP um sort of an event now, which is lovely. 00:30:54.640 --> 00:31:04.319 And so we you still you can hear the challenges that people are facing, both maybe on the regulatory side with with constraints on what they can accept. 00:31:04.480 --> 00:31:11.680 Uh you know, especially with new technology, it's very difficult to accept all different new technologies that come on board. 00:31:11.759 --> 00:31:19.920 There's due diligence that needs to take place, and then you hear the challenges that that people are having um to try say ahead of the curve. 00:31:20.079 --> 00:31:36.799 So we've had wonderful conversations, especially with the same some of the same people over the years where years ago they were just getting their data house in order, now they have, and now they're driving to the you know, a machine learning algorithm isn't giving us the insight we want. 00:31:36.960 --> 00:31:38.960 What are the missing pieces of a jigsaw? 00:31:39.119 --> 00:31:58.160 And so often there's really interesting um sort of debates into how could we, and it's something we do quite often, what other flavors of data can we bring to enrich their data source that could be, as I said, maybe some of the missing pieces of a jigsaw that would elevate a non-obvious signal. 00:31:58.480 --> 00:31:58.799 Exactly. 00:31:58.880 --> 00:32:05.039 There's a lot of public data out there, and sometimes people are looking at what's happening in their facility, but look, what's happening outside the facility? 00:32:05.119 --> 00:32:19.039 What's uh coming down this the line in terms of all this information we can get from RASelf and uh alerts and other on other food safety data sets that are in the public domain, but integrating all that as geology and especially doing it in a scientifically valid way. 00:32:19.279 --> 00:32:20.000 Yeah, yeah, yeah, yeah. 00:32:20.079 --> 00:32:20.559 Yeah, exactly. 00:32:20.720 --> 00:32:42.640 And though not publicly available, I think during your session um uh one of the audience members gave something that would be obvious to us to use, but I don't think it was obvious to some of the people attending, where they were they realized that one of the heightened risks of food safety, the the data set came not from food safety data, but from HOR. 00:32:42.720 --> 00:32:43.680 I don't know if you remember that question. 00:32:44.000 --> 00:32:44.400 Yeah, right. 00:32:44.480 --> 00:32:47.759 So I I I don't know who it was, but let's call him John. 00:32:47.920 --> 00:32:48.160 Yes. 00:32:48.400 --> 00:32:52.880 When he went on on vacation, yeah, their food safety um risk spiked. 00:32:52.960 --> 00:33:02.880 And so actually, when they joined HOR data to their food safety data, they got a far better picture and realized actually the the risk mainly goes back to your human factors. 00:33:03.279 --> 00:33:10.559 Um the risk was when this guy, John, went on vacation and suddenly the standards were were dropping. 00:33:10.880 --> 00:33:18.799 There's a couple of great anecdotes uh shared uh around data and joining the dots and learning from all of the data at the end of one of our sessions. 00:33:18.880 --> 00:33:23.440 I do remember that uh it was great to get that industry's perspective, and that was great. 00:33:23.519 --> 00:33:35.920 Like, um, so did you see any other good collaborations between industry scientists, regulators that are helping like break down those boundaries to solve food safety problems? 00:33:36.240 --> 00:33:37.279 Uh yes, yeah. 00:33:37.359 --> 00:34:12.960 So even you know, we take um, let's say a government chemists uh conference in the UK, always striving to the various working groups looking at at emerging risk and technology that can support it, data that can support it, and uh highlighting the fact we were there co-presenting with Finn, the Food Industry Intelligence Network around fraud and food integrity and authenticity, and the that that how to I guess share data, how to

access that data, understanding that at times this is confidential data or deemed confidential. 00:34:13.119 --> 00:34:24.320 Um so uh a big drive I've seen is that people understand that if your data starved, your insight starved. 00:34:24.719 --> 00:34:28.480 And so they're looking for data in all over the place. 00:34:28.559 --> 00:34:33.840 Yeah, and and then on top of that is harmonization and standards, yeah, so that we start to speak a common language. 00:34:34.000 --> 00:34:34.239 Yeah. 00:34:34.880 --> 00:34:43.840 So this is where you know food safety for me, then yes, it becomes a systems problem from supply chain, but a systems problem from a from a data point of view. 00:34:44.400 --> 00:34:51.440 So adding in harmonization of terminology and standards unlocks a huge amount of power. 00:34:51.519 --> 00:34:55.920 And for us, data is an asset, yeah, and everyone has it, they just need to harness it. 00:34:56.239 --> 00:35:02.400 It's a lot, it's a lot of work in harmonizing, and we do a lot of that to try to connect data sets that were never designed to work together. 00:35:02.480 --> 00:35:10.559 So these these initiatives, and I know those one at IFB talking about data dictionaries and harmonization for food safety, and we'll be supporting that going forward. 00:35:10.639 --> 00:35:17.280 Uh, that will save a lot of that admin work that was allow us to move towards predictive analytics more quickly in in these projects. 00:35:17.519 --> 00:35:18.480 Yeah, yeah, exactly. 00:35:18.559 --> 00:35:33.519 And I think we'll see that we're also seeing it in in some publications now where people are publishing their raw data, especially these machine learning models that publish the algorithm, that publish the data that was, I guess, machine learn ready, was it was made machine learnable. 00:35:34.159 --> 00:35:45.360 Um and I think you'll also see that in the future that data harmonization and standards also have them in a in a format that's ready for to access AI tools. 00:35:45.679 --> 00:35:46.639 Brilliant, yeah. 00:35:46.960 --> 00:35:50.639 So you've been coming to IFP for a number of years, as have I. 00:35:50.960 --> 00:35:58.719 So did the 2026 meeting feel different from where maybe people were five years ago? 00:35:59.519 --> 00:36:02.480 Uh I yes, yeah, I think so. 00:36:02.559 --> 00:36:05.119 And I probably alluded to it um a couple of times. 00:36:05.199 --> 00:36:11.519 So one is that things like AI is is embedded in all the talks. 00:36:11.679 --> 00:36:21.360 It's not this sort of new shiny, I wouldn't say what is this, yeah, yeah, and and maybe also deemed or viewed a little bit skeptical, or let's wait and see. 00:36:21.840 --> 00:36:28.480 Um, which I can understand um when you when big decisions are on the line, you can't adopt everything straight away. 00:36:28.719 --> 00:36:41.360 Um I I think five years ago a lot of the conversation was about generating data and having better tools and kit and methodologies and lower limits of detection. 00:36:41.440 --> 00:36:47.599 And um now I think the the acknowledgement is there isn't a data storage. 00:36:47.679 --> 00:36:58.559 I think as you mentioned a few minutes ago, um it's it's it's that maturity shift in how can we now utilize tools to better harness the data that we're generating. 00:36:58.639 --> 00:36:58.880 Yeah. 00:36:59.440 --> 00:37:08.239 Um uh and so we've seen a lot of conversation that wouldn't have happened five years ago around uh data governance, data culture, it's your session. 00:37:08.320 --> 00:37:13.920 You know, I I now maybe I'm wrong, and maybe there was a data culture session five years ago. 00:37:14.000 --> 00:37:21.119 But certainly it was food safety culture was built, but now it's moved to date more focus on data culture, I believe, in the last year or two. 00:37:21.360 --> 00:37:23.840 And I think that's going to be interesting next year. 00:37:24.000 --> 00:37:26.960 I don't know, but ask about next year there's all the seemingly well nothing. 00:37:27.199 --> 00:37:30.400 But I I and I see a lot of leadership buy-in. 00:37:30.639 --> 00:37:37.920 I I see you know, again, these sort of informal coffee chats where people saying, Oh, it's a great idea, but I just I can't get any leadership buy-in. 00:37:38.159 --> 00:37:40.159 Yeah, you do need budget to do this. 00:37:40.800 --> 00:37:50.559 Um, and sometimes it is um it's not the exciting work if if what you're doing is getting a data management system in place. 00:37:51.360 --> 00:37:54.960 But by doing that, you unlock so much for the future. 00:37:55.119 --> 00:37:59.599 Um, maybe in the very, very short term, there isn't immediate gains. 00:38:00.159 --> 00:38:12.239 But by doing that, and we talk to people who five years ago were starting that journey, and now we're talking to them, and they're and have been well before five years, unlocking benefits and insights to stay help the curve. 00:38:12.480 --> 00:38:13.039 Absolutely. 00:38:13.199 --> 00:38:16.400 And I I was um talking with as I say about data culture. 00:38:16.480 --> 00:38:27.760 I was trying to encourage every food safety person in the room to think of themselves as a as a leader in their organization in terms of driving that culture around data for food safety um and how to do it. 00:38:27.840 --> 00:38:34.320 So I think that that's really like that's doesn't cost money and just think you have to think like a you know, data scientists all the time. 00:38:34.559 --> 00:38:46.239 Yeah, but looking ahead then, um, what do you think 2027 will bring for food safety, AI, and risk intelligence across the scope of all of the the conferences you attend, perhaps? 00:38:46.719 --> 00:38:54.239 Um so so I think in in so there are probably some teams, food safety, fraud. 00:38:54.559 --> 00:39:02.719 We have the the Holy Food Project that's an EU funded project that will soon wrap up around early risk detection, yeah, and weak signal detection. 00:39:02.880 -->

00:39:08.400 So I I could see, and then that's often very early stage, low TRL stuff. 00:39:08.559 --> 00:39:11.840 So pace that will rapidly be coming through. 00:39:12.239 --> 00:39:34.320 Um so I think you'll see um I guess the the the moving closer of food safety, integrity, resilience within the supply chain, all moving forward, as I said, maybe with data and so forth, that at the core, I think a move between to prioritization, not above detection, but prioritizing your resources and where you look. 00:39:34.639 --> 00:39:35.119 Exactly. 00:39:35.519 --> 00:39:50.159 And and so being as smart as possible, but intelligence, data driven, to see where I should be looking when, to what extent, and what's not good enough, but what ensures that I mitigated that risk. 00:39:50.320 --> 00:39:50.880 Yeah, yeah. 00:39:51.039 --> 00:39:53.440 Um and so and I think a lot more sharing. 00:39:54.559 --> 00:39:57.199 IFP is a uh genuinely wonderful place. 00:39:57.440 --> 00:40:03.519 I've been going to conferences for far too long on various disciplines, and people are very open, very keen to share. 00:40:03.599 --> 00:40:13.519 When we do our, you know, any symposia that we host, when we're preparing it, we talk to people about is there any actionable intelligence people can take away? 00:40:13.840 --> 00:40:16.400 Well then conversations are less, yeah, right. 00:40:16.559 --> 00:40:32.079 What we want is that people could implement something as a result of a symposium or or or um uh presentation, and so um I think that's what we're seeing is that people are willing to talk about best practice, things they've implemented, yeah. 00:40:32.239 --> 00:40:50.239 Um, and I think you'll you'll be hearing more about the implementation of AI, maybe moving from that decision support to decision making, decision enabling, even though yeah, I always like to see what people are doing, and it also inspires others that if they can do it, we can do it, you know. 00:40:50.320 --> 00:40:53.599 So it's really good to see people sharing their success stories. 00:40:53.840 --> 00:40:55.360 So thanks, James. 00:40:55.519 --> 00:40:56.800 It was a great conversation. 00:40:56.960 --> 00:41:09.360 Um, what comes true from both conversations with Brendan and James is that the center, the central challenge, I believe, now is no longer simply collecting more data or even adopting AI. 00:41:09.519 --> 00:41:20.320 It's converting that knowledge of scientific knowledge and industry data into actual trusted, actionable decisions that can actually prevent problems. 00:41:20.480 --> 00:41:23.840 I think we're not there yet, but we are we are we're on that journey. 00:41:24.000 --> 00:41:30.719 Now we have all the tools, we're starting to get all the data, but we really need to get those decisions right to prevent problems. 00:41:30.800 --> 00:41:34.559 And that's the whole goal of these uh initiatives in the food safety. 00:41:34.800 --> 00:41:44.239 And from our contributions and interactions at IFP, I'm more convinced than ever that data-driven food safety starts with culture, not dashboards. 00:41:44.400 --> 00:41:50.639 As I mentioned, the human factors are becoming the limitations rather than the technology. 00:41:50.880 --> 00:42:00.079 That might be a surprise for people coming from a data scientist, but in the age of AI, data scientists, data science and AI has become more accessible. 00:42:00.159 --> 00:42:08.000 Yet the culture needed to manage data, trust the data, and make consequential decisions based on the data signals has to mature. 00:42:08.639 --> 00:42:11.519 Thanks for listening to Predictably Blood Good. 00:42:11.760 --> 00:42:14.320 I I hope you enjoyed the conversation. 00:42:14.559 --> 00:42:25.039 Follow the podcast or find us on YouTube uh so you don't miss future episodes and share it with anyone interested in AI, data, and predictive intelligence and help of shaping better decisions. 00:42:25.280 --> 00:42:29.840 I'm Cronen McNamara, and I'll see you next time on Predictably Good.