Listen on
Consumer packaged goods (CPG) companies spend millions on market data, yet often struggle to convert it into actionable insights due to the sheer scale of information and internal capability gaps. This challenge directly impacts business planning, product strategy, and market responsiveness. In this episode of Eventual Consistency, Sam Kahn, VP of AI and Data Science at TICKR, explains how modern generative AI is breaking through these long-standing limitations. With over a decade of experience in machine learning and natural language processing, Sam details how advanced AI solutions empower CPGs to transform messy data into precise forecasts, dynamic product categorizations, and proactive strategic decisions.
The conversation reveals how AI-driven tools perform tasks previously impossible for humans at scale, from cleaning ambiguous OCR data to searching vast external data sources for predictive signals. Listeners will learn how these capabilities translate into tangible benefits, such as a 16% reduction in forecasting error and faster time-to-value for complex data science problems, helping CPG leaders move beyond historical reporting to genuine predictive advantage.
Key Takeaways
Generative AI enables proactive forecasting by integrating vast external data at scale.
Traditional forecasting often relies on internal metrics. AI, specifically large language models, can scan and select external data from sources like the Federal Reserve Economic Database or news archives, identifying relevant covariates that a human could never process across thousands of products. This ‘generative predictor search’ has demonstrably reduced forecasting error by up to 16% in CPG operations, moving businesses from reactive to predictive planning.
LLMs can fix and dynamically categorize inconsistent product data, unlocking downstream analysis.
CPG companies receive product data from various vendors, often in messy formats or with inconsistent categorizations that are useless for internal business analysis. Generative AI models can infer true product names from corrupted or ambiguous data, such as lossy OCR scans, and re-categorize products dynamically. This critical data cleansing makes downstream applications like accurate forecasting and sales analysis feasible.
AI accelerates time-to-value for complex data problems, moving from months to weeks.
Deploying traditional data science solutions often involves significant lead times. For established generative AI applications within the CPG space, Sam Kahn states that solutions can begin providing predictions within a week, assuming no novel R&D is required. This rapid deployment capability means businesses can quickly see the value of their data investments and automate critical processes, freeing up internal teams.
Predicting future consumer sentiment with AI offers a strategic advantage for CPGs.
Beyond sales forecasting, generative AI can predict consumer survey responses based on historical data and current events. This experimental capability allows CPGs to anticipate shifts in consumer sentiment related to products or market conditions, enabling them to proactively adjust marketing, supply chain, or product development strategies to mitigate risks or capitalize on emerging trends.
Related: CorrDyn helps CPG companies define their AI strategy and engineering robust data pipelines that drive real business value. Learn more about our work in consumer packaged goods or explore 7 principles for LLM business cases.
Full Transcript
Jason: Welcome to Data BS, a show dedicated to tackling the big questions impacting the world of data and ML AI without any of the BS. My name is James Winegar. Each week I sit down with guests from across the data ecosystem to unpack how they’re shaping their businesses or the businesses of others through their real-world application of data engineering, ML AI, infrastructure, analytics, and more. No fluff, unfiltered, but slightly edited to remove noise. This is Data BS. Let’s get into it.
James Winegar: Welcome to Data BS. We’re here with Sam Kahn. So Sam, you want to tell us about you and what you do?
Sam Kahn: Thanks for having me, James. My name is Sam Kahn. I’m VP of AI and data science at TICKR. I’m also an AI research fellow at UC Santa Cruz. My background has about 13 years in machine learning and what people now call artificial intelligence. My journey started as a physics major at UC Santa Cruz. Then I got an internship at NASA working on recurrent neural networks for energy forecasting. That led me to a graduate degree at UC Berkeley. After UC Berkeley, I worked at Salesforce.com for a little bit on their customer intelligence team as a data scientist. Then I helped found a startup called Market.Space, which was a natural language processing startup, and that led me to where I am today. TICKR acquired that startup and I’ve worked at TICKR for eight years. We specialize in delivering AI and data science solutions for the consumer packaged goods space, which you might hear me refer to as CPG, and the retail space as well. You might not know what a CPG is. A CPG is a company like Lays potato chips or Coca-Cola or one you might have heard of is called Procter & Gamble. These are companies that usually take some item, put it in a package, and sell it at the store. We build data science and AI solutions to help those consumer packaged goods companies make their business operations more efficient, save time, and save money. Really the promise of AI and data science actuated.
James Winegar: So what’s TICKR’s core value proposition then? How does it differentiate itself from its competitors?
Sam Kahn: When I think of our competitive space, I honestly can’t think of anyone who has our level of expertise in consumer packaged goods through our business development team and our data science team. More specifically to what I do, we have a very deep level of AI and data science expertise and we’ve focused it into the consumer packaged goods and retail space. We’ve learned over time what the real pain points these companies are experiencing because we’re talking to them, we’re deploying real solutions for them. As far as I know, we’re the only company in CPG and retail focusing on AI and data science in this way.
James Winegar: When I usually think about companies providing data solutions to CPGs, it’s usually more as a vendor. Data vendors. And then you’re also partnered with those data vendors, right?
Sam Kahn: Yeah, the challenge we faced initially is we didn’t have data. How do you do data science without data? Part of our strategy over the past several years has been to partner with data providers like NielsenIQ or Numerator or Datasembly. Their business model is to sell data to these consumer packaged goods companies. This data often is simply sales data, like the retailer might be collecting sales transactions at the cash register. A lot of these data providers have deals with the retailers to collect that data. But then they sell that data to the CPGs for millions of dollars. Sometimes many millions of dollars. Some of these CPGs do have data science teams like Coca-Cola. They have a very well-known data science team, or Walmart. But a lot of them do not. They buy this data then they don’t know what to do with it. We partner with the data providers or the CPGs to build repeatable AI and data science solutions to help them get more out of their data. These solutions could be AI-driven business planning, hierarchical product categorization, forecasting, but now we’re not just doing forecasting. Our forecasting is what we would call AI-assisted. We have language models improving our forecasting models and making modeling decisions at scale as a human would, but deploying it at scale.
James Winegar: You’ve worked in the NLP space for a very long time. I think as long as I’ve known you you’ve been doing NLP. At least 10 years. There’s always been this promise with NLP that we’re going to be able to take all this crazy text data and do more with it. Then suddenly ChatGPT comes out, you start talking about Gen AI and how Gen AI enables this new capability that we’ve been trying to work on for a long time but it got to a certain point of effectiveness.
Sam Kahn: Yeah.
James Winegar: My gut feeling is that these new capabilities actually provided TICKR new market opportunities that didn’t exist before because you can do this automated decision-making type process. First thing is what are the most common use cases for TICKR in the CPG space and retail? And can you talk about examples of previously unsolved problems that you can address?
Sam Kahn: Because we’re working with very large enterprises, a really common use case is forecasting. It’s a very classical data science use case. They want to look a year, maybe even further into the future and understand how their business is going to do. I’d say that’s probably the most common use case we hear, but now with generative AI and the power of both the commercial frontier models and the open source models, there’s a whole new space of problems we’re also solving. Those problems would fall into hierarchical product categorization, taking the data from these large data providers and improving it so you can use it for downstream data science applications. A lot of the data comes in super messy, maybe they’re using OCR to scan receipts. The language models can go in and fix that data. Some of the problems that these language models have unlocked boils down to essentially a level of reasoning and what might be called zero-shot capabilities at scale. These models can make really human-like decisions at scale so you can deploy little nuggets of reasoning in different places of your application that we couldn’t before. One solution we’ve deployed and we’re actively working on with a customer is something we call generative predictor search. A problem with forecasting and an ask from our clients is, we have all this data, it’s internal to our business. We’re tracking our units sold, our sales, our promotional price, all this stuff that we control. There’s all this data out there on the internet and not just on the internet, just like in the Federal Reserve economic database, in the news. Wouldn’t it be great if we could go get that data and integrate it into our forecasting models? But it’s really not scalable to do for hundreds of thousands of different products across thousands of different categories. These language models, if you train them to search the web or search the Federal Reserve economic database, they can go look through all this data and make decisions as a human would on what they think could be predictive for our models.
Sam Kahn: And you might ask why don’t you just test every single time series you can find on the web and put that in your model? Well, as you know, James, spurious correlations. You’re going to find that the price of wood in Saskatchewan, and maybe you could reason through some reason why this could be, but it’s very predictive of the sales of my Lays potato chips. You could maybe find a causal mechanism there, but it’s likely not the case and it’s likely spurious. The language models can look through that data and they know that is likely a spurious correlation, so they help us do high-recall search and then bring in a few potential covariates to help our forecasting models.
James Winegar: So you’re using the LLM or the Gen AI capability to do a thing that a human would have normally had to have gone through, but people are not good at looking at hundreds of thousands of things over and over again.
Sam Kahn: Exactly. To be honest, a human would have to do it before but most of the time a human wouldn’t do it because it’s just not possible.
James Winegar: It’s not feasible.
Sam Kahn: Not feasible. The amount of time it would take to do this across every product category and things like that, which I think is also important to talk about, hierarchical product categorization. Why is that a hard problem for these CPGs? I’ve worked in CPG, so I’ve seen it, but- Essentially product categorization is hard for the CPGs like I was saying earlier, they buy this data from the data providers and the data providers will categorize it into their own categories. But those categories 99% of the time are totally useless to the CPGs because they bucket their data in ways that make sense for their business. It comes down to them not actually knowing how to recategorize this data at scale. It really just comes down to extreme multi-class classification. Using the language models as either few-shot learners or fine-tuning them and then doing few-shot learning. It really comes down to they just don’t know how to do it.
James Winegar: Let me see if I can reframe it in different words, you tell me whether or not you agree with me. These models have taken what would have required an expert CPG person to go across more work than they could ever possibly get done in their lifetime and provide the ability to do this classification of different toilet papers or toiletries, but maybe that doesn’t make sense because of how they do their manufacturing. Procter & Gamble’s got, I don’t know, hundreds of thousands of products that they sell. Categories and brands, etc. Being able to do that across their entire portfolio and across their different retailer spaces, or consumer spaces, helps them understand what is even going on.
Sam Kahn: Yeah, it helps them understand what’s going on, it helps them group their products so they can look at sales metrics in the ways they want. Also very important, the language models, because of their ability to look at a few examples and predict very well from those examples, their ability to meta-learn, they can change their definition of product categories very easily. Before this would require a full retraining of a BERT model. But they can change their product category definitions and with our AI systems they can then send those new categories to our system and then you can project the data or the products to a whole new set of classes. That was not possible before. And that is very exciting.
James Winegar: So it’s dynamic too. It’s not just the process of doing the categorization, it’s being able to evolve the categorization as things change in the market, which is happening all the time.
Sam Kahn: Exactly.
James Winegar: You’ve talked about it already, but data is messy, especially for these CPGs because it’s very distributed across all the different retailers, their direct-to-consumer, the e-commerce type platforms. There’s tons of stuff that’s going on and you got receipts and they get OCRed, we got structured data maybe coming from certain places. With Walmart for instance you kind of get an EDI feed or something like that, but I don’t know, Trader Joe’s might not, they probably do, but some place is going to just do orders and then you have this procurement process that gets managed and you have to figure out did it actually sell. Coupons are a very complicated thing in the space, how do coupons get utilized, where should you even deploy a coupon, why should you deploy a coupon, all these types of things. From TICKR’s viewpoint, my understanding is that you’re doing a lot of automated data cleansing because there’s really no practical way to do it across all these different data providers that you’ve partnered with. Do you want to talk me through what’s going on with the data cleaning side?
Sam Kahn: I think the easiest example would be the one you mentioned, the lossiness in OCR basically. Somebody buys something at a store, they get a receipt, and a lot of these data providers, their business model is to essentially pay people to take a picture of their receipt, OCR it, and they collect that data and then the CPGs buy that data from them so they can see what people are buying in the store. But there’s this process of OCRing the data where either the OCR is bad or some of the retailers don’t really want that to happen, so they make the names of the products or the descriptions of the products or maybe even the UPCs, they make them ambiguous. For downstream applications like forecasting or other data science applications, this missing data that gets collected through the OCR process really affects the data science models. We felt like this frontier of generative AI could look through this data and maybe clean it up and we’d get better downstream forecasting results. That’s what we ended up doing. Where did that intuition come from? It basically came from the fact that as a human we could do it, we could clean it, so if a human can do it but it’s not scalable, a language model could probably do it. Essentially we have a few different types of language models looking at this data and inferring what the true product name is. How does it infer that? We ground it with essentially an external data source of real products and then it can reason over that data. It’s basically a RAG pipeline. A new product comes in, you look for potential similar products, and then you have the language model reason either the actual non-corrupted data or it will tell you it can’t figure it out. Because we also don’t want false positives. That would corrupt the data as well.
James Winegar: Which means you also have a methodology to bring the feedback loop in, which is super important for these Gen AI use cases. I don’t have a high amount of confidence in what I just gave and so I’m going to create a system in place within the TICKR AI platform to get that human feedback loop in.
Sam Kahn: Yeah, it depends. If you have training data, and you have data you can benchmark the model against, if the system you build performs well on those training metrics, say the accuracy of UPC recovery, or the number of UPCs recovered. If you have enough training data to measure the performance of your system, you don’t always need a human in the loop. We do use a human in the loop for some of these applications, but for the most part we calibrate the models to only predict when they’re— they’ll never be 100% right, but when we believe that they’ll be 99% right. What we find is it still helps the downstream data science application so you can get more accurate business planning, product categorization, because if you have corrupted data, you don’t know what the product is, you can’t categorize it.
James Winegar: Which then breaks all your forecasting use cases.
Sam Kahn: Yeah, which breaks the forecasting use cases or it breaks the categorization use case we’re talking about earlier. Because the model won’t know what the product is.
James Winegar: Let’s get back to search, because you talked about the wood in Saskatchewan and then Lays potato chips and things like that. How does the TICKR platform prioritize those external data points that it’s looking at?
Sam Kahn: Let’s say we want to forecast Lays potato chips. We will have a bunch of internal metrics. This usually comes from the CPG like sales, maybe the promo price, the price, the base dollars, ACV, feature, display, all these metrics that CPGs track in stores. We don’t need a language model to search over that data. The space of covariates is pretty small. Anywhere between 20 and 100. But the space of external data you could maybe argue is infinite or close to it. That’s where the language model comes in, searches for that data, and then returns it. Basically it returns data it thinks could be predictive of Lays potato chips. But we’re not just going to put that data in the model and just fire it off and go. We did a lot of research to see if this actually worked and helped our models. The number off the top of my head is we got a 16% reduction in error for our forecast models by doing an external covariate search with a language model. That in itself speaks for itself. You can get a 16% better forecast, you’ll be able to plan better for your business. That 16% is driven by statistical models. We actually test the forecasting models on out-of-sample data, we see if it actually improves, and then we’ll deploy it. There’s cases where external data will not help a model. Sometimes you don’t have that much data, you need to be very careful about how you do forecast, you don’t want to overfit to your validation set, things like that. And you’re doing this at scale so you have to be very careful.
James Winegar: What’s one of the ways you manage being careful?
Sam Kahn: You could almost think of it as a machine learning classifier. We bring our decision threshold to a point where we’re sure that anything we put into the model we’re correct about. We err on the side of being careful. Everything we do is tested. We would never deploy a model without testing how it does on out-of-sample performance. We’re very careful not to do too many comparisons either. As we all know, you can get spurious correlations. Even with the language model doing that high recall search and then whittling things down for us so we can integrate them into our model, we still need to be careful. That is done through more traditional data science methods like out-of-sample performance.
James Winegar: So you have just traditional basic statistics that we’ve been doing for a long time, but you’re pairing it together with Gen AI capabilities, NLP capabilities. We got a bunch of time series data, a bunch of text data basically, or very unstructured or very messy, and you’re bringing that all together. We’ve talked about covariate search or product search. We’ve talked about different categorization and we’ve talked about forecasting. Are there any other use cases that you’re really bringing together?
Sam Kahn: A few use cases we’ve been working on with clients are automating the process of survey collections, response coding, things like that. Pretty low-hanging stuff for language models just to speed up that survey and the process of running surveys, collecting data. We’ve actually done some really interesting research lately, one of our AI scientists, on predicting the responses of surveys with the language models. We collect a lot of surveys and then we ground the language models in historical survey data and then we test whether it can actually predict next year’s survey responses, and we can see that we can actually get a pretty high level of accuracy there. This is one of our experimental research projects, we’re still exploring it, but it is pretty exciting because these models have a very strong level of expressive power and a very exciting ability to essentially mimic humans. There’s still a long way to go, but this Cambrian explosion of use cases is pretty amazing right now.
James Winegar: The survey one’s crazy to me almost. It’s almost like you’re doing a regression on the survey, but is there sentiment coming in from PR sources and things like that that you’re also bringing into that?
Sam Kahn: The survey, we went online and we found a bunch of surveys, I believe from some type of social work database. We had the hypothesis that maybe if we collected enough longitudinal surveys, we could then hold out a few years of data at the end and see if the language model could infer how these survey responses changed grounded on maybe current events, stuff like that. And it does pretty good. It gets pretty close to answering the way the humans would. Of course, this opens up a space of concerns around bias and concerns around propagating harm and stuff like that. This purely is a research project for us because we do have a lot of clients that live in the consumer survey space. Consumer packaged goods companies run consumer surveys to understand how people view their products. Wouldn’t it be great for them to be able to run a few surveys and then deploy to a language model and then predict what the broader sample or population would say?
James Winegar: Another thing from a business decision-making side, you could say, hey, the forecast is telling us because of whatever recent events that we’re going to have a downward trend in sentiment for whatever this product category is. So we can get ahead of that. Now, toiletries is probably not a good example because it’s a necessity, but you can be like, okay, toiletries is a thing that we need to deal with because a hurricane’s coming through. We know that people are going to be like, where’s all the toilet paper? Because they’re going to buy all the toilet paper. That leads to a bunch of negative sentiment for Procter & Gamble and all the different CPGs that sell toilet paper. If they could just get ahead of that in terms of, hey, we shipped 500 pallets extra toilet paper to Florida before Milton hit to make sure that people had supplies available during the recovery phase. Some interesting stuff to think about.
Sam Kahn: You can definitely protect yourself. You could think of a lot of use cases, maybe a bad news cycle. How is consumer sentiment going to change from this? There might be some news cycles you might perceive will really affect your business that won’t, and there might be some that will. Having a language model take an objective look, grounded in real data, not just off hallucinating whatever it wants, is an exciting proposition for us.
James Winegar: How do you do against the categorical variables? This is more just me talking about why I hate accuracy. Because accuracy across what— which buckets? You could also tie it into your hierarchical classification. Are you optimizing against your hierarchical classification as well?
Sam Kahn: This data was coming straight from the data provider. We weren’t doing any rebucketing of that data. But no, I’m fully aligned on accuracy being a bad metric. I don’t ever really use accuracy, but it’s digestible. Precision, recall, F1, those are the metrics I usually use.
James Winegar: Colloquial accuracy. How’d we do? Relative to what we care about.
Sam Kahn: Pretty good. The context of the prediction problem always matters too.
James Winegar: I think that’s important too, what is your benchmark? What are you comparing against? For a lot of these problems that we’ve talked about today, there was never really a person who could do it because there’s just too many things to go through. Your baseline is basically, does somebody catch it on accident because you have good operational processes? Usually the answer to that question’s going to be no, just because there’s a bajillion variables.
Sam Kahn: For me and for us at TICKR, most of the clients we work with have nothing in place. Or the way they forecast is they look at last year’s values and they just roll them forward with 20%-
James Winegar: Exactly.
Sam Kahn: Sometimes that works, a lot of times it really doesn’t.
James Winegar: CPGs have an interesting thing too because the more established CPGs are protecting their market versus growing their market. Procter & Gamble, Coke, etc., they’re about maintaining their market share more than anything. More profit usually comes from different campaigns or new SKU kind of thing, but from a market share perspective they’re not moving that much because Coke versus Pepsi, people have their opinion about what they like and that’s pretty much staying the same, except for new markets like Coke and Pepsi internationally did a bunch of stuff 50 years ago or whatever. For those big CPGs, market protection or brand management is the primary thing they do, but then you got these up-and-comers and they can get 20% growth just through the marketing basically, whereas the big CPGs got to do different things to have that sustained growth that’s expected for publicly traded companies. Another thing, I’ve worked with some CPGs in the past and talked to some people who were really interesting with their business acumen about the problem. They can see that there’s brands that are going to take off, do an acquisition, and then put the force of the major CPGs marketing capabilities behind it and really just dominate within that new sector very rapidly because they have all the processes to support that marketing capability, whereas a small company, unless they’ve done it before as a group, they have to learn those lessons, and so the machine can help take these brands way farther. There’s a lot of things there with justifying the value of an acquisition as well. Has TICKR done anything to help with evaluation of acquisitions at all or not really?
Sam Kahn: Evaluating how effective, or whether a company actually got the value they thought they were getting out of an acquisition? You’re suggesting that when a company gets bought you then figure out a way-
James Winegar: Or during due diligence. During the due diligence can you do some forward capacity planning forecast based on-
Sam Kahn: I’m sure there is, especially using some type of language models and assistant to go collect data and look at historical patterns, stuff like that. I’m sure there’s definitely a path forward for use cases like that. In this world of generative AI, I’m very bullish, I think almost everything will be possible in a lot of business use cases and processes will be automated in the next 5 to 10 years. That sounds like a bit of a moonshot, but I actually think it is possible.
James Winegar: I think the extremely repetitive and manual processes that are white-collar in a sense, or at the computer, are going to be the automated. It’s where there’s a lot of branching, or where people don’t even do it because the amount of branching is so large, like the hierarchical product categorization problem that we talked about. Because people have been needing to do that for a long time but they just couldn’t. We got a use case where it’s millions and millions of job descriptions. People are not going to go through that and extract out the salaries and the locations and things like that. You can try, but you’re not going to do it.
Sam Kahn: Totally.
James Winegar: There’s two sides of that, things that nobody would even do before because economically you couldn’t make it make sense. And then there’s the super repetitive, I think of a lot of data entry type of stuff.
Sam Kahn: A lot of data entry type stuff, a lot of things you might be doing in an Excel file. I do think a lot of data science is going to get automated. I think the language models are going to be able to reason as well as somebody with a PhD in economics or somebody who’s very specialized in causal inference, things like that. We’re not there yet, but they will be able to do that. I do think there’s a space of problems that will be solvable but aren’t solvable right now because there’s no data, because people haven’t thought about solving these problems. The one we’re hearing about a lot, it’s a little more high level, is agentic paradigms, agents. All people are talking about agents for solving very complex reasoning tasks but there’s no data to really benchmark agents yet. I know Scale AI is working on some, I’m sure there are some others. There’s the really easy things to automate, the things that there’s not that much data on yet or no data, and then there’s the other space that probably won’t be automated.
James Winegar: I’m less bullish on the agentic approach because there’s interfaces, anytime you got interfaces you got complexity, the more complexity the more problems.
Sam Kahn: I don’t think we’re there yet, but just like with other aspects of language models, like chain-of-thought reasoning or in-context learning, or you could maybe even consider RAG as an emergent paradigm, even though a lot of these have been around since 2020. I do think there is emergent behavior that has been shown from agents, but it’s all in really constrained cases. When you deploy them, it doesn’t work.
James Winegar: For CPG organizations evaluating TICKR, what are the pain points it solves? In other words, what technical or operational challenges does TICKR address that these organizations struggle with?
Sam Kahn: I’ll bring it back to something I said earlier, it really comes down to them being unable to save the money and time that they want with their current teams. We come in and we automate processes for them, we improve their forecasting and business planning and any other data science solutions we can build on top of their data to make them more efficient, more scalable, and save time and be able to focus their efforts where they want to focus their muscle, basically.
James Winegar: I would also argue time. Not just saving time but also time-to-value or time-to-market for some of the things. Maybe with an evaluation there’s a reason they’re talking to TICKR. The fact that you’re able to stand up a solution because you’re already integrated with the data providers in a month to three months, one quarter deliverable or within a quarter, is a pretty strong statement.
Sam Kahn: For a lot of our generative AI solutions, because we’ve trained a lot of the models already, they come to us with a problem. As long as it doesn’t require any additional R&D, it’s like a week for us to start flowing data into their system, giving them predictions. That speaks for itself. They start seeing the value instantly and then-
James Winegar: So from a platform play, you’re able to get stuff running with less than a month basically, unless there’s an R&D effort that’s required because the problem statement is new. Cool. All right. Where can people learn more about you and TICKR?
Sam Kahn: Just go to our website, tickr.com. We also have a blog where we talk about the research we do. It’s not exhaustive, but it’s some of the things we do on the side while working for clients and experiments that we make. If you’re interested in potentially a career at TICKR, just go to our careers page. We’re always hiring AI scientists, data scientists, engineers.
James Winegar: All right Sam, thanks for coming on. Hope you have a good one.
Sam Kahn: Thanks James. See you later.
Jason: That’s it for this episode of Data BS. If you enjoyed this episode make sure you subscribe wherever you listen to your podcasts and not miss the next one. This episode was sponsored by CorrDyn, a data consultancy that helps organizations unlock the power of their data. If you have a data challenge, we can help. Visit corrdyn.com, C-O-R-R-D-Y-N dot com to learn more. See you next time.
