Beyond Generative AI: Embracing a More Strategic Future | Syrup | PI Apparel

I'm James Terekhov, everyone. I'm CEO and co-founder of a company called Syrup. Syrup's an AI company. We empower merchandising and planning teams with inventory recommendations. This will not be a talk about Syrup. If you want to learn about Syrup, come find us at our booth on the third floor. Myself and my team would love to speak to you about your planning, buying, allocation, replenishment, rebalancing, topics. But today that's not the focus. We're going to be speaking about more strategic questions and concretely around how do we cut through the noise in GenAI and how can we embrace AI for a more strategic future. And you've got the, the QR code for the Q&A if you want to scan that and add questions as we go through the session, please feel free to do that. We'll also put it up at the end if you want to ask questions, then we can also do that. But let's get started. We hosted a session at PI last year where the question was, how is AI going to change the role of merchandise planners? And we were joined by planning leaders from Abercrombie Fitch and Bonobos and Reformation who all saw the power of AI to be able to change the game. And then we also opened up the floor to questions from the audience. There were probably about 35 or 40 planners And this is what we heard. This is what they told us they were most excited about. It was 4 things. The first was how can AI help forecasting and demand planning and getting more accurate and more granular forecasts? The second was around preseason planning. How can AI help with the merchandise financial planning process, with the AP planning process, with getting the buy right up front? The third piece was around in-season optimization. How can AI help the replenishment process, the in-season chase or reorder process, and the rebalancing process? And then the fourth and final bucket was around pricing and markdowns. How can AI help that process? And if you look at these four buckets, they're actually not generative in nature. You might use generative tools to help solve them, But more than anything, they're predictive use cases, right? Or use cases of predictive AI. So let's speak about the 2024 AI zeitgeist for a moment. I think there's really 4 big emerging trends that we're seeing. One is that consumer trends are moving faster than ever before, and they're more volatile than they've ever been. The second is that demand signals come from seemingly everywhere, right? Whether that's the TikTok trend or an event or influencers speaking about a certain thing online, demand signals are everywhere. The third is that while not everyone is pursuing a fast fashion model, the old model of 2 to 4 collections a season, very long lead times, that seems to be a thing that's going to be largely of the past in the near-term future. And then the fourth bucket is that channel complexity is really modeling the assortment planning process and the role of the store in that post-COVID world and the role of the boutique that's still being defined. Now, I think all of this can be overwhelming and the reality is it is overwhelming, right? There is simply too much data out there for us mere humans to be able to handle. And that is overwhelming. And while that might be overwhelming for us, for AI it's a goldmine. And you know, we think about data in this goldmine setup. And kind of, if you look at this, we have these three levels of the data goldmine. Close to the surface is your easily accessible data, right? Real-time sales and your inventory levels, product attributes. As we go midway down that goldmine, you can find things like, historical performance and ad spend and web analytics and weather. And then as we dig even further down, you can now find things like store analytics and social media and foot traffic. In the pre-AI world, what I would have told you is just focus on the top, right? If you're using a simple linear regression, focus on getting the top of top data, right? Because if we just add social media noise to a linear regression model to forecast retail sales, it ain't going to do much. That's actually very, very different in the AI world. What happens today is that AI models can pick up a trend on TikTok, can then link that to how click-through rates on your website are performing, link that to e-commerce transactions, and then link that to retail sales, creating this amazing web of linking all of this data. But if that's what's possible today, I think this, this opens up the question of saying, how far have we come, right? And as we look at 2024 in review, where are we, right? If it's possible, how close have we gotten? And I think the answer to that is that in AI, the big spotlight continues to be the flashy LLM models, right? The generative models. But if we ask ourselves, have we really made progress on these predictive use cases, use cases, I think, or it seems to me that the answer is unfortunately no. No, we have not. Which then leads to the question of why, right? Why, if the promise is so big and the opportunity is so big and there's so much excitement about this, why haven't we made real strategic progress? And what I would like to present to you today is an attempt to answer that question. With what we call an AI readiness model. Like, what does it take for organizations to be ready to really be able to leverage AI strategically? And this readiness model has 4 key areas, and we'll go through each of them over the next, over the next 5 or 10 minutes. The first is data availability, right? What data do you have access to that AI models can crunch? The second is around computational ability, right? How, how sophisticated are the engines to actually make sense of that data goldmine? Then the third is around strategic scalability, right? How can you tell the model what your strategy is and what you're actually optimizing for? And then the fourth and last one is around change management, right? So people, processes, organizational capabilities. Let's go through those four with the question of what does it take to be able to embrace AI strategically? And if we start speaking about data availability, this is both about data availability as well as data accuracy, right? Data is the goal, one of the rocket fuel for AI models. So what does it take to have the data to be able to run AI on top of it? Let's start with the good news. I think all of you have the most important data within your four walls already. Whether it's in an ERP system or an OMS system or a POS system or a PLM system, the core data foundation that you need to make, to run AI on top of, you already have that. However, it might not be historically clean. You know, we all know that suddenly a color coding changes from being blue to different types of blue, or a product that you've been running for a few seasons gets a little change and then there's no relationship to the old product. One of our European customers, their ERP was set up in such a way that it only had 8 season codes. So that means that every season they would chop off the top data. That's like throwing data from your gold mine out of the window every single season. What my recommendation here is to you is work with your internal tech teams or your internal data science teams if you have them, or even your vendors on bringing what we think of as a unified data model, right? How can you bring together all the different sources of data that you have? To give you another real customer example, we work with a large multi-billion-dollar footwear player, and what they do is they launch their styles online 2 weeks before they launch them in store. Now, this data lives on 2 different systems that report to 2 different parts of the organization, and retail and e-com isn't unified by them. Brick-and-mortar team that we were working with had real struggles getting access to that e-com data. The moment that we actually got access to that data by unifying those two, you can imagine the massive impact that having 2 weeks of sales data before something hits stores has on allocation and replenishment performance, right? So this is just another very tangible example of the benefits of unifying data. So that's data availability. I think it's the number one question that we get at Syrup is, is our data ready? Is our data good enough so that we can run AI on top of it? If you have an ERP or if you have a data lake or if you have a modern POS system, I think the answer is most likely yes. Yes, you have enough data to start your AI journey or at least the crawl part of the AI journey. We have had cases where we have seen data and we have told customers or potential customers, actually, I think you should do a data cleanup project before you run AI. But that's the exception. It's definitely not the norm. So my conclusion on this, I think as an industry, we are at a point where we are ready for phase 1 of AI, and that shouldn't hold you back. Don't have imperfect data. Data hold you back to start of that journey, but also start thinking about what can you do today, what can you do in the next few months to make ever more data available to your AI models. That's data availability. Let's talk about computational ability. By that we mean what ability do you need to be able to handle these vast data sources? The headline here is going to be spreadsheets aren't going to be enough, right? And linear regression models are not going to be enough. And I'm going to show you why that's the case. Let's take this example of this t-shirt. It comes in 4 sizes, size S to size XL, and it comes in 5 colors. So suddenly this one t-shirt is actually 20 SKUs, right? If you were to do this with denim, where we have 3-dimensional sizes of waist length, and fit. These 4 sizes could actually easily be 4 T sizes that could come in up to 10 different washes. Suddenly this one style of denim is actually 400 products, right? The combinatorics start going, going real big here, real fast. Now on top of that, this t-shirt doesn't just exist in isolation, right? It only exists as part of an assortment, which might sound a little philosophical, but it has real tactical implications, right? This one t-shirt will have a substitute relationship with some other t-shirts in your assortment, with shirts, polo shirts, and have a complement relationship with sneakers or trousers or jackets, right? So, every product in your assortment will affect other products and will be affected by many other products in your assortment as well. And then, on top of that, your different channels are going to be performing very, very differently across your assortment, right? And whether that's across different store types, whether that's your flagship stores and your pop-ups and your outlets, or whether that's across e-com versus retail, or whether that's within e-com, right, your dot-com site, your different marketplaces, they will behave very, very differently, which, you know, adds a whole other vector to this problem. Then on top of that, the rules of the game are constantly changing. How you fulfill demand and how you meet demand across your network will really change. If we talk about this peak season, for example, 2024 is very different from 2023. Thanksgiving is later this year, which means that Black Friday and Cyber Monday are later this year, which means that we actually have 5 fewer days between Cyber Monday and Christmas in, in 2024. On top of that, the 12 days before Christmas, or the C12 as we call them, behave very differently to the pre-Christmas period, and they are very, very important, as you all know, for, for retail. This year, the C12 only has one weekend with where ground shipping is going to reach customers before the holiday period. That's going to have a real difference on how channel performance is gonna perform. So all of this to say there's a lot of complexity in this, right? And so what's really needed to be able to make sense of this is a technology that can tackle big data, right? And AI is the perfect answer to that actually. It's beautifully equipped to handle these big data problems and to have the computational strength to make sense of that. All right. So, that's computational ability. Whether you're building AI models in-house or whether you're working with a vendor on them, it doesn't actually matter. My recommendation is the same. Make sure that you have the computational ability, that your partners have the computational ability to make sense of this data and that they're on a journey to add ever more computational ability as these things will only get ever more complex with time. Now let's talk about strategic scalability. And this might be the most interesting aspect of the AI readiness model. It might also be the most problematic. The question that I would pose to yourselves here is the following. Where are your business rules, your business metrics, and your constraints documented that, guide how you're making your inventory decisions. More often than not, we find this to be just an institutional knowledge, right? And whether that's just in the heads and minds of people who know how to do what they're doing, or whether that's a document that's somewhere, or whether it's within the black box of an old-school system that no one really knows how that works anymore, Those are all things that we see all the time. Practically speaking, the rules end up just being hardcoded in spreadsheet cells anyways. I'm seeing some giggling and nodding here. At the end of the day, the formulas in these cells contain some of your most important strategic questions. Whether that's on your target metrics, like what's your target margin, what are your optimal weeks of supply, or whether that's on your business rules, what are your minimum display requirements or visual merch requirements or your store SKU eligibility, or as your constraints, what are lead times, what are minimum shipments. These cells are covering the core strategy of your businesses. What I'm proposing is bringing the these strategies to the forefront of your AI models. It's really important that you don't just have the computationally most sophisticated models, but you can also strategically tell the model, whether that's knobs and dials or whether that's a configuration page of what are you optimizing for. Doing that in a scalable way is critical to be able to embrace AI strategically. All right, so that's strategic scalability. Now let's talk about the fourth and last component of the AI readiness model, which is the people. And that might actually be— if strategic scalability to me is the most interesting piece, the people is probably the most important piece. And I think when you hear change management, that might sound quite scary, right? I hear change management, I immediately think of large consulting projects and ERP transformations and multi-year journeys that are going to require you to learn a whole new language or a whole new system. I don't think that's it in the age of AI, right? That is a thing of the past and of old-school software. Here's what I mean concretely. Planners have really, really hard jobs, right? We all know the Sunday scaries, the Monday meetings, the business review where the planning team can't get things right one more time. I've actually never been a planner, but I've spent hours and hours and hours with planning teams in a consulting capacity. I did not envy the planner's jobs. There are so many tasks that you do just to keep the ship afloat. Here is where AI comes in. I believe that the core differentiator for AI for your people is that for your staff is that it can take a lot of these mundane tasks and have the intelligence run that for you so that over time you can spend more time on the creative and strategic questions and move from a task mindset to a strategic mindset. And I think the core mindset that's required for this is not that you're going to wait for this large change management program or this large transformation, but this is something that you can start within very small bits and pieces, right? And embrace the idea that AI can start taking little bits and pieces and you can start seeing the impact and the value of it. So that's what I mean by change management. I was actually having lunch on Friday with one of our customers who've done incredibly, incredibly well on their AI journey. It was lunch in Ohio and I asked her, what do you think was key for you to actually succeed? Succeed on that AI journey. And what she said was it was a mindset, right? That it was, she said something along the lines of, hey, we didn't expect everything to be perfect from day one, but what we did expect is our people to have an openness to try, to iterate, and to get used to something new. And that's what was the big unlock for them. So that's my recommendation here, right? Start, start small. First of all, don't be afraid to start and be open and be open to embracing this. Then the question is, why do we do all of this? What's the promised land? In my mind, it's the potential to create super planners. These super planners are not the AI. The AI is just the tool. The super planners are the people who have the ability to get infinite intelligence at their fingertips to be able to drive outsized business results and to make the planning job honestly a lot more interesting and a lot more rewarding. So what to do next? We talked through this, this 4-piece AI readiness model on data availability, computational ability, strategic scalability, and change management. This is going to be a journey. I don't think you need to wait until everything is in place to get started. And you're also never going to be really done. You're always going to keep iterating on this. I think that's number one. Number two, I'd suggest focusing on the cognitive problem here. This is not an ERP transformation. It doesn't actually matter what your ERP is, where the data lives, what your ERP does or doesn't do. It's really about how can AI help you solve the really difficult cognitive problems. Find people in your IT leadership who understand this. And then lastly, I'd say find use cases that are easy, right, that are easy to start and to iterate on. There's no point in starting with a broad preseason collection planning process if you can't even get the replenishment or the rebalancing right, right? So start with something that's easy, that's measurable, that's tangible, that you can show real results in. And so before we go to questions, I have two parting thoughts for you. One is an aspirational parting thought, the other is tactical. On the aspirational one, let me start there. I did my MBA at the Harvard Business School, and you know, I came into the MBA program, I was really excited to learn about retail. There's very, very few case studies taught about retail in the Harvard Business School curriculum today. Very interesting. Interestingly, if you go back to the '90s, retail was the number one industry that case studies were written on. That's partially because retail is the OG big data industry. Then something happened over the last two decades where our collective attention was grabbed by, I don't know, it was the rise of e-commerce or something else, but we kind of lost that. I think AI presents this huge opportunity, real generational opportunity for retail to go back to become a core data industry, and that planning is actually the place where that's going to take place. That's my aspirational parting thought for you. My tactical parting thought is, don't be afraid to start. The benefits of starting today will way, way outweigh any of the costs, whether it's time investment or money, of delaying this process. The winners of the AI age are going to be the ones that start early, that start iterating, that start testing, and that start really getting used to it. So don't be afraid to start. And that was that. So I think we have a few minutes for questions, right? Like 4 or 5 minutes. If you want to add questions on the QR code. There we go. Wonderful. Yeah. So maybe we'll just go through those if you want to add any more. Let's go it. Let's do it. We'll start with the first. What AI products should I choose that rapidly prove value? It's a great question. I would recommend starting small. And starting tangible. So whether that's carving out a brand within your organization or a region within your organization or a team that you think is particularly excited to start and try out something, start there. And then our recommendation is always to start with an in-season use case, right? Because you don't have to wait 18 months to see the results, right? If you're doing replenishments or rebalancing things, you can see the impact within a matter of weeks, right? So if it works, incredible, you have like something really cool to take to your C-suite. If it doesn't work, you can just discard it real quick, right? So that's, that's always our recommendation is start small, start tangible, because the real kind of eye-opening moments that we've seen at least have been when we've been able to show results in a matter of a month or two. And taking that to the C-suite and say, hey, we're gonna put this technology, our in-stock rates are up our full price sell-through is up 6%. That's really, really powerful. So I think anything you can do to do that is what I would recommend. And then second question, are there companies already using AI for planning? Definitely, right? Many. I think different versions of it. You know, you have like the absolute forerunners, you know, Amazons and Sheins and Temu of the world, they are effectively data and AI companies and they run their entire processes on AI. Other players, many of the companies that we work with, I guess by the nature of working with us, are using AI in their processes. I think as we've heard over the course of even just over the day or over the day-to-day, there's different different iterations, different phases of it. So I think this is not a, oh, we'll be the first in doing this. It's also not we are the last in doing it. We saw the show of hands earlier on Excel. I think this is, obviously this is the time to do this. How do I start an AI or productive-culottes? It's another great question. You don't need a data science team. I would say of the companies that we work with, probably 80% do not have a data science team. There are benefits to having a data science team. They can help prepare data, they can help prepare insights, but by no means do you need one. I think what you do need is someone in IT who can make sure that data exchange changes and data integrity and all of that, that that's set up properly. But the beauty of a lot of these AI applications is that you don't need expert engineers or data scientists or analysts to manipulate formulas. It's a bit like the ChatGPT of the world. You don't need to be a programmer to use ChatGPT. You don't need to be a data scientist to use cutting-edge AI tools for inventory either. So I would also say that's definitely not an impediment to start. If you have one, that's great. And I think you can work in synergy, but by no means, by no means that should stop you. Cool. Thank you.