AI in Food Safety: Enhancing Compliance, Consumer Insights, and Product Innovation withMark Haas

[00:00:02] I get asked a lot, what are customers doing with AI? How do they measure it?

[00:00:07] Well, I had a chance to talk to Mark Haas. He's the CEO of the Helmsman Group, and they focus on the consumer packaged goods space.

[00:00:14] Now he comes at it as a consumer packaged goods expert, subject matter expert, and we learn how they're using natural language processing to get better outcomes and better things to be played on this bonus episode of the Business of Tech.

[00:00:32] Well, Mark, thanks for joining me today.

[00:00:34] Yeah, thank you for having me, Dave.

[00:00:36] Now, I'm super interested to dive into this, but you came with some of this really interesting premise of injecting AI into food regulation and food safety.

[00:00:47] And I feel like I need to start with, tell my audience a little bit about the backstory of what you've created and how you got here so that we can dive into some of that thought of how it works.

[00:00:57] Sure. I've been in the consumer packaged goods industry for 30 years.

[00:01:02] And we've had a development agency for the last 13.

[00:01:06] Four years ago, to create more efficiency in our agency, we set upon building a data warehouse.

[00:01:16] We looked at off-the-shelf solutions.

[00:01:18] None really fit the bill in terms of the sophistication that we were looking for.

[00:01:24] And it's sort of a kismet moment.

[00:01:28] We were renting, subletting one of our offices in our facility in Emeryville, which is where we started this business.

[00:01:35] And it just so happened that the tenant that was renting space from us came by a common space one day while I was mapping out the architecture of a database.

[00:01:47] He asked me, is that a database?

[00:01:49] And I said, yes.

[00:01:50] And the conversation grew.

[00:01:52] And it turns out that he is a NLP expert.

[00:01:57] And we joined forces and built this platform that we now call Regulate.

[00:02:03] Gotcha.

[00:02:03] And tell me a little bit about what Regulate does.

[00:02:07] Right.

[00:02:07] So initially, it was just to archive information.

[00:02:10] It was a static database to warehouse different supporting pieces of information for food and beverage.

[00:02:16] In food and beverage, it's a highly regulated industry.

[00:02:20] We have, I would say, often more than 100 pages of documentation per each ingredient.

[00:02:25] And there are regulations, of course, with how you produce different products.

[00:02:31] And when we...

[00:02:35] Let me pause for a second here.

[00:02:39] When we...

[00:02:41] The initial project was simply just a data warehouse and to hold this information.

[00:02:46] When I learned more about Ishan, my business partner, our CTO's background, I dove pretty deep.

[00:02:56] It was pulling some things from my college years.

[00:02:59] I was involved in the fractal project at the University of Washington in the 80s and early 90s.

[00:03:05] And I was...

[00:03:06] I did code in C++ back then.

[00:03:09] And so I had, you know, some solid knowledge.

[00:03:12] And so I dove pretty deep with him about database architecture and current programming languages and his area of expertise.

[00:03:20] And didn't really understand natural language processing at the time.

[00:03:25] This was 2019.

[00:03:25] And so I really immersed myself in what the potential utility of NLP is.

[00:03:32] And it occurred to me, as a domain expert, the complexity of regulations and how we would go about effectively vectorizing that data to interpret it for purpose-built outcomes.

[00:03:44] It was something that nobody was really looking at.

[00:03:46] This is before OpenAI really was putting anything into the market at the time.

[00:03:51] And so I felt that there was something here that we needed to explore and dive a little deeper on.

[00:03:57] And the truth is, now with where AI ML is, specifically NLP and LLMs, the power is pretty remarkable of what we're able to accomplish now.

[00:04:08] We've had some amazing tools.

[00:04:10] Now, I'm dying now to know.

[00:04:12] So give me a great use case of what you're doing with it.

[00:04:15] The most recent module that we built, we call it Consumer Sentiment Analysis.

[00:04:20] And so we go to open API connections on social media channels, X, Reddit, and Amazon.

[00:04:28] And we're looking at what people are talking about.

[00:04:31] So we create a framework about a product.

[00:04:33] And so we might talk about a snack bar.

[00:04:35] And that snack bar, people will comment and discuss and share opinions and facts and whatever is on their mind.

[00:04:43] Reddit appears to be the best channel when we're dealing with open API connections.

[00:04:48] Although if you are working with a brand and you had curated channel and you have better analytics and better information, we can extract a lot more information from that.

[00:04:57] And parsing the data from commenters allows us to understand in a more organic fashion what large bodies of people truly think.

[00:05:07] You invite an audience member to be part of a focus group or something like that.

[00:05:13] They tend to have some sort of confirmation bias or others have confirmation bias.

[00:05:19] We're looking at real live conversations between people and what are the commonalities.

[00:05:25] And in food and beverage, we tend to focus on product and recipe a little too heavily at times.

[00:05:32] When we analyze the conversations with people that are anonymous to us, we get greater insights, price points, merchandising locations, packaging information, font, typeset, script,

[00:05:49] things that relate to the other aspects of the packaging that are really important for marketing and consumer engagement.

[00:05:55] And so we analyze this in three different ways.

[00:05:59] A couple of typical reports.

[00:06:01] We have a word cloud report that helps us understand volume and frequency and what's being discussed.

[00:06:06] We've created our own GPT prompt to help us understand the context and the syntax of the information.

[00:06:15] So we're not looking at the actual commenters.

[00:06:17] We're looking at the amalgamation, the output in a word cloud report.

[00:06:20] And so it's busy and there's a lot of information.

[00:06:24] And at times it seems incongruent.

[00:06:27] When we analyze it through our GPT prompt, we're getting very clear insights as to what consumers are really thinking, what their drive is.

[00:06:36] And we have other filters.

[00:06:39] We can look at it.

[00:06:40] I wish it was different or what is positive or negative.

[00:06:43] So we can change the perspective.

[00:06:45] We can change the lens with how we look and output the reports.

[00:06:49] We have attribution reports.

[00:06:51] So, you know, quality of ingredients or package or, you know, what do I want to see different?

[00:06:56] So those are basic histograms and sort of volume metrics.

[00:07:00] And then we have another report that we call a centrality analysis.

[00:07:05] And it's pretty unique.

[00:07:07] We can analyze the data and the interconnection of thoughts.

[00:07:13] This is really, I think, the advantage of deploying NLP for this purpose.

[00:07:18] We can dive in and fully understand and analyze the connection points, what a central concept is, why that's supporting, why that's necessary in a particular product.

[00:07:33] So if you're designing from scratch, you have an idea, you know, we call it a white space innovation.

[00:07:39] You don't know where to start.

[00:07:40] We can run this consumer sentiment analysis and look for the central nodes.

[00:07:46] And from that, what are the supporting nodes that really make that shine in consumers or consumers eyes so that when we formulate, we can incorporate those elements or attributes into the final product.

[00:07:59] Gotcha.

[00:08:00] And that includes, I mean, even things like flavor trends, right?

[00:08:02] Like you can actually get a sense of what people are interested in.

[00:08:05] Like, how does that change and how are you tracking it?

[00:08:08] Yeah.

[00:08:09] Yeah.

[00:08:09] Yes, we can do that with flavor trends.

[00:08:11] We haven't really modeled that.

[00:08:13] That's a, we think we're going to build another module that is built on the architecture of this consumer sentiment.

[00:08:19] The consumer sentiment in the CPG world is a pretty big nut to crack.

[00:08:25] Nobody really knows what consumers are thinking until after you go to market and you see transactions.

[00:08:30] What do people actually buy?

[00:08:31] And there's, it's a, it's a risk mitigation strategy to utilize our platform to leap ahead, to project, to foreshadow what probable outcomes are likely to be.

[00:08:44] In our internal testing over the last year, we've proven through the use of this data model, our module, that never before conceived fusion concepts are now hitting the market and resonating with consumers, fulfilling that particular need state.

[00:09:06] Right.

[00:09:07] So that they're getting what they're talking about, what they're asking for instead of what is been modeled through quantitative analysis.

[00:09:16] So the typical trend has been the way it's been done for the last 30 years in CPG is we go to syndicated data.

[00:09:25] We do a market size opportunity.

[00:09:27] We do a basic mathematical movement analysis.

[00:09:30] What's moving?

[00:09:31] How frequently is it moving?

[00:09:33] What is the velocity of the product?

[00:09:34] What's the total dollar value of a particular category or product?

[00:09:38] And then we try to find white space between what's moving, you know, where is there another future potential opportunity?

[00:09:46] And that's just somebody like me as a subject matter expert opining on what that could be.

[00:09:51] And then recruiting a focus group audience to validate that concept, which to me is nothing but confirmation bias.

[00:10:02] And so you're maybe successful, you know, if you're experienced, you'll hit a second base outcome.

[00:10:11] But, you know, home runs are pretty infrequent.

[00:10:14] And what we're striving for is, you know, home run at bat every single time.

[00:10:20] And so, I mean, I would think that you've got so much data at this point.

[00:10:24] What have you learned through this process about what works really well with sort of managing this much data to be effective and what doesn't?

[00:10:34] Well, I know my peer engineers out there, we're all in the data quest, right?

[00:10:38] We're all looking for more data.

[00:10:39] We learn something and then we need more data and we, you know, want to keep going.

[00:10:45] What we've learned is there are so many business applications for how you analyze the data and what problems you're solving.

[00:10:53] So if you look at traditional business problems, what is my budget going to be next year?

[00:10:59] Should I increase my price?

[00:11:00] Am I likely to incur inflation?

[00:11:03] You know, these are just basic business metrics.

[00:11:05] We can incorporate that.

[00:11:08] There's plenty of data that is in the public domain.

[00:11:13] You know, so we have futures indices.

[00:11:15] We have, you know, that could be ICE or whatever.

[00:11:18] All these different markets where we look at soybean, futures, corn, doesn't really matter.

[00:11:24] Most every ingredient is either derived from a commodity or a derivative of a commodity.

[00:11:29] And meaning that, you know, canola price, well, canola has a market now, but sunflower seed oil is a derivative of the soybean oil market because it's a seed oil and it's an alternative to.

[00:11:42] And so you can create some metrics to analyze what sunflower pricing might be.

[00:11:48] And if that is in your recipe deck, we can build a market basket analysis and project what's likely to occur based on costs, which, you know, translate ultimately to price.

[00:12:01] And businesses can make informed decisions.

[00:12:03] Do I need to have a conversation with my key accounts and, you know, talk about a price increase?

[00:12:08] Is it, you know, nominal enough that I can absorb some of this cost and I can have a strategic, you know, pricing competition with my competitor and not have a price conversation?

[00:12:20] So the degree of insights that we can glean and supply to business solutions is pretty powerful.

[00:12:29] We're learning something new every day.

[00:12:32] And we've been asked, can you do sustainability outputs?

[00:12:36] Yeah.

[00:12:37] And we just, it just, it just depends on how we design the architecture of the database, what we can pull from the data and ultimately what we can connect to.

[00:12:48] I like to connect to public domain sources.

[00:12:50] There's a lot of insights out there that is effectively free data that allow us to understand consumers, weather patterns.

[00:12:59] There's a lot of things that influence the purchase, purchase behavior of consumers.

[00:13:03] Gotcha.

[00:13:04] And there's a compliance element to this too, right?

[00:13:06] That you can actually use all of this data to help on the FDA compliance side.

[00:13:10] Tell me a little bit of how that works.

[00:13:12] So federal regulations, public domain data, everything's written, the code of federal regulations, CFR.

[00:13:18] So the FDA is based on 21 CFR 100, which is the food regulations.

[00:13:23] And so that's everything.

[00:13:24] What you can call a product, we call it standard of identity.

[00:13:27] What has to be put on a label, what the nutrition facts are, what the supports are, claims that you can or can't make.

[00:13:34] And all of that is in gigabytes of data.

[00:13:39] We can use our engines to parse that data, to query that data set, specific questions.

[00:13:47] Is this recipe in compliance?

[00:13:50] Is this nutrition facts label that we generated in compliance?

[00:13:53] Are there alternatives?

[00:13:56] Is there a different way of naming this?

[00:13:59] What's right or wrong with a proposed label, whatever that label may be?

[00:14:05] We have optical character recognition engine.

[00:14:08] We can look at anything.

[00:14:11] We can look at our own outputs and then compare that to the data set.

[00:14:14] And regulations are ambiguous.

[00:14:18] They're not super clear always by design.

[00:14:22] And so NLP is another powerful tool that allows us to understand probabilistic outcomes based on ambiguity.

[00:14:30] Gotcha.

[00:14:31] Okay.

[00:14:31] So I guess the final piece I want to get a little bit of insight is how are you measuring success of this?

[00:14:37] Is it profit improvement alone?

[00:14:41] Is it production?

[00:14:42] Give me some of the sense of the KPIs and the success metrics that you're looking for on these kinds of data efforts.

[00:14:47] The number one is, well, is I would say it's not the principal purpose, but it is a major superlative and a differentiator.

[00:14:59] It is the compliance and assuredness that everything is going according to plan.

[00:15:06] The business case, the primary deliverable for business case, the reason why people would subscribe to the product is go to market.

[00:15:13] So we can reduce the speed to market and improve profitability.

[00:15:17] And that by itself and some of the tests that we've done, you know, a product, for example, that has a $5 million launch target year one.

[00:15:25] We've been able to accelerate its launch timing by food products take years to put together typically for strategics.

[00:15:34] We reduce that time to market by 22 months, which translated to $13 million of unbound revenue based on the business's compounded annual growth rates.

[00:15:44] And so that's just one product.

[00:15:46] And, you know, there are, I would say, typically more than 20,000 new products brought to market every year.

[00:15:52] Well, right there is the exact definition of a business outcome.

[00:15:55] And that's what we talk about all the time in the show.

[00:15:57] Mark Haas is the CEO of the Helmsman Group, where he leverages a deep background in consumer packaged goods consulting to pioneer AI-driven solutions for the food and beverage industry.

[00:16:07] Under his leadership, the company integrates cutting-edge technology to streamline compliance, drive product innovation, and support brands in navigating FDA regulations.

[00:16:15] Mark, I've learned a ton today.

[00:16:17] Thanks for joining me.

[00:16:18] Thanks for having me, Dave.

[00:16:19] I really appreciate it.

[00:16:22] Are you ready to get your brand in front of the tech leaders shaping the future of managed services?

[00:16:28] Here at The Business of Tech, we offer flexible sponsorship opportunities to meet your needs, whether it's live show sponsorship, podcast advertising, event promotion, or custom webinars.

[00:16:39] From affordable exposure options to exclusive sponsorships, our offerings are designed to fit businesses and vendors of all sizes looking to make an impact.

[00:16:49] Prices start at just $500 per month, making our packages a fraction of typical event sponsorship costs.

[00:16:58] Be a part of the conversation that matters to IT service providers worldwide.

[00:17:04] Join us at MSP Radio and amplify your message where it counts.

[00:17:08] Visit MSPRadio.com slash engage today to explore all the ways we can help you grow.

[00:17:15] The Business of Tech is written and produced by me, Dave Sobel, under ethics guidelines, posted at businessof.tech.

[00:17:25] If you like the content, please make sure to hit that like button and follow or subscribe.

[00:17:30] It's free and easy and the best way to support the show and help us grow.

[00:17:35] You can also check out our Patreon, where you can join the Business of Tech community at patreon.com slash MSP Radio or buy our Why Do We Care merch at businessof.tech.

[00:17:48] Finally, if you're interested in advertising on this show, visit MSP Radio dot com slash engage.

[00:17:55] Once again, thanks for listening to me.

[00:17:57] I will talk to you again on our next episode of The Business of Tech.

[00:18:04] Part of the MSP Radio Network.