Under the Hood of Cresco Labs: AI, Tech Integration, and Custom Retail Solutions ft. Zach Marburger

Under the Hood of Cresco Labs: AI, Tech Integration, and Custom Retail Solutions ft. Zach Marburger

Zach Marburger: We've built a lot of those parts of the stack ourselves because we feel that it can offer us actual competitive differentiation. When you look at a lot of the head-to-head competitive metrics we measure to make sure that's true, we believe it's panning out. So for us it's really about identifying the problem and working backwards — is it a big enough problem to solve with technology? Then we develop a thesis from there and ultimately solve it with a technology product, or by working with a vendor externally. Bryan Fields: What's up guys, welcome back to another episode of The Dime. I'm Bryan Fields, with me as always is Kellan Finney, and this week we've got a very special guest, Zach Marburger, Chief Information Officer at Cresco Labs. Zach, thanks for taking the time. How are you doing today? Zach Marburger: I'm doing all right, guys. How are we doing? Bryan Fields: Excited to have you here. Kellan, how are you doing? Kellan Finney: I'm doing really well, really excited to talk to Zach, really excited to dive into the back end of what makes these companies tick and what separates them from each other. How are you doing, Bryan? Bryan Fields: I'm stoked. I think we're going to get really nerdy today into topics that I think Kellan and I both really enjoy — topics I think other people are going to find more interesting than they realize. At the end of the day we've got buckets of big MSOs, but there are differences between them all, and Cresco's got some real unique characteristics we're going to highlight today. But Zach, before we dive into the fun stuff — I know you've traveled to Denver before — if you had to have a current location, East or West Coast, where are you? Zach Marburger: I guess Chicago is east of the Mississippi, so we're going east there. Bryan Fields: I love it. So cool — can you give a little background about yourself and how you found your way to Cresco Labs? Zach Marburger: Yeah, through a roundabout way. I started off in Colorado like a bunch of folks did in cannabis in the early days, and I've been working in cannabis since 2013-2014. We had a software company at the time, and Charlie and the Cresco folks were starting up around that same time in Illinois. We met each other in Colorado around then through a consulting relationship, and one thing led to another — I was arguably the first or second guy in the door, depending on who you ask — and I've kind of been here ever since, doing mostly tech stuff and holding the role of CIO. But I think I've also been the janitor, the sales guy, the everything, since then. It's been quite the ride. Bryan Fields: What is your background? Zach Marburger: Largely I'm just an entrepreneur. I have a technical leaning — I'm a hack-job programmer — and I have a network of very strong, technically inclined professional friends who build really material stuff that I've always worked with. I was a child who grew up on the internet, that first generation, so tech has always been a piece of the heart. Previously I built two small tech companies and raised angel and venture funding for both. I've always just been a hacker, in the traditional sense. Bryan Fields: I love it. So take us through Cresco and some of the specialties your skill set brings to the table that people might not be so familiar with. Zach Marburger: I think maybe zooming out to the industry first, because that determines the sandbox we play in. In cannabis, as I'm sure you've talked about on previous episodes, we have to trace everything — there's this seed-to-sale traceability requirement that exists in slightly different ways from market to market. It adds a really burdensome opportunity for operators — entering data into systems, adding tags to plants, all these things that allow you to trace what you're producing and selling. At the root of your technology infrastructure, that's really where your heartbeat lies, where you're interfacing with systems like Metrc, BioTrack, New Leaf Data Systems, and some others. Sometimes they're a necessary evil, sometimes they're an amazing partner — depends on the circumstances within the state. But ultimately all the technology systems a cannabis business has, including Cresco's, have to tie back to that in some form — our cultivation environment making sure our plants are talking to the state, our manufacturing environment, or our retail. It's important to have a good understanding of that basis, because I think a lot of operators, ourselves included, are trying to figure out what this looks like in the future. Is it necessary in its current format? It can be very expensive and burdensome for operators to do things that are in the spirit of compliance — which we're fully behind — but could it be done a different way, quicker, better, faster, without all the considerations we go through today? For us, having a really good understanding of our obligations to those systems and rules allows us to build tools, systems, people, and processes around it so we can be slick, reduce operational costs, and increase efficiencies. That's really where we have some special sauce. Kellan Finney: Yeah, and track-and-trace can be really burdensome. A lot of companies have had it as a real scar in their business plan — takes a lot of man hours, a lot of time. There is a silver lining though — the macro concept that if you can keep track of everything throughout the supply chain, you should be able to create correlations and do business intelligence that optimizes your process as a whole. But currently, because these programs are maintained and implemented by the state, a lot of those high-end data analytic tools aren't embedded into the software. Is that the state of the industry right now? Zach Marburger: Yeah, within these track-and-trace programs they're a little disjointed, maybe in a more friendly term, where you need extra steps of integration and learning to make sure you can use, for example, a basic Tableau environment, or integrate a data lake like Snowflake and interface it with your seed-to-sale system. There are a lot of traditional tech things that take two, three, four, five more steps as a cannabis business. A lot of what our peers and we go through as public companies — going through more financial regulations — means you also have to have super robust controls in those systems from a financial audit standpoint. Our rules, regulations, and obligations increase every single day, and that detracts from building features that are great for customers or other more competitive things. That's a big, unspoken piece of consideration for larger companies in the space — they have to build a lot of compliance-driven controls and features to meet regulations, and that takes time. For us, the integration piece and ability to do analytics comes with interfacing with these systems, and that extends to our people. We have an internal team of about six or seven folks focused on data pipelines first, and prediction and modeling second — that's a key competency for us, along with really talented people who actually do the work. Bryan Fields: So let's take a step back, because I want to make sure everyone understands — a business like Cresco is vertically integrated, running maybe six businesses simultaneously in a single state: cultivation, processing, retail. When you're spinning up these businesses, they have their own tools, and you have to figure out how to connect them so they communicate and pass information back. Is it more need-based or pain-based when that first starts? Take us behind the scenes — what's the conversation with Charlie like when you're scheming out, hey, we want A to talk to B to solve this problem, versus, hey, this would really alleviate a problem we hadn't thought about? Zach Marburger: I think it really starts with a business problem and working backwards from that. Taking cultivation technology as an example — our folks who run cultivation are really technology-driven, as much into tech as into plants, and they come up with really crazy, awesome ideas that we have to fulfill. That means engaging with vendors like Trym, which has a sensor that we use in some of our facilities to capture environmental data at a frequency that lets us dial in plant conditions to drive yields. At that point it's important to know how this is ultimately going to scale — does it have an ROI, what's the schedule, and what value does it drive for customers, because at the end of the day the product's going to the customer. Does it make it better for the customer? That's a big internal question that helps drive product quality. Otherwise it's sometimes needs-based — maybe we encounter a problem in distribution, where now we have so many trucks on the road that every time we're late or take a wrong route it costs us far more money than it used to, so investing in technology to be more UPS-like becomes an interesting idea now that maybe three or four years ago wasn't even a consideration. It's also where we create differentiation. Something maybe unique about what we do internally — and this isn't a critique, just an observation — is that a lot of folks use off-the-shelf tools, like a Treez or a Dutchie point of sale, or an e-commerce system like Jane or Dispense. We've built a lot of those parts of the stack ourselves because we feel it can offer real competitive differentiation, and when we measure head-to-head competitive metrics, we believe it's panning out. For us it's about identifying the problem, working backwards, asking if it's big enough to solve with technology, developing a thesis, and then solving it with a technology product or working with an outside vendor. Bryan Fields: Is there a critical point where you're out looking for a technology solution, find something off the shelf, and think, this just doesn't do XYZ, so we're going to build it? Walk us through how you decide to buy versus build internally. Zach Marburger: It starts with sourcing first. We lean towards buying if we can — it's generally cheaper, easier to maintain, and quicker to start. It also has a commodity cost — if everybody's doing it and it's freely available in the market, you'd be silly not to buy, it may cost you more not to. If we go through that process and aren't happy with the quality — there might be three options but none do what we need — then we evaluate building. There's a concept I've seen people talk about on Twitter: compounding product interest over time. You built an ERP-like system that four years ago you wouldn't have built, but because you're so far along, building more now is much cheaper and easier, and you're adding IP value to grow its full capabilities — which is kind of cannabis 3.0, not your capacity and assets, but your capabilities. We didn't just start building this thing — we've been building it for going on eight years now, so we have a robust ecosystem where our buy-build evaluation may skew towards build because of how big that system has become, though not without its own challenges in maintenance. Kellan Finney: Does it make it easier to integrate, since you built the pipelines — you know exactly what bolts fit right here, so integration and correlation between departments becomes faster? Zach Marburger: No doubt. Traditional master data management becomes cohesive and manageable. When you buy or acquire companies — like our last acquisition, a small retailer in Pennsylvania, a market we already operate in — it's really easy to say, hey, two new stores, welcome to the family, you now use these systems, click these buttons, and it will drive this outcome. That produces a better employee experience and a predictable outcome — we're not relying on an external vendor to deliver what we want. But the bigger you get, the riskier it becomes. I'm the tech guy people call when something breaks — most of the time not somebody external — so there's a great responsibility to make sure it performs at a high clip, otherwise it doesn't go very well for us. Bryan Fields: We can get into the fun aspects, but I don't think people realize how hard it actually is to start. You have an idea, a concept — then you have to hire a software developer, figure out the programming language, where to host it, and your team's still handling everything Cresco does while trying to build this internally, which isn't easy when you're not a Silicon Valley startup with unlimited cash. What was that first product you built internally, and what was the goal you were trying to hit? Zach Marburger: Great question. Before Cresco, in gen one of cannabis, my partner — who runs software engineering today at Cresco — and I built essentially the first Jane. We scaled it to about 200 locations, and at the time the biggest locations in the country used it. We sold it to a micro-cap company right at the beginning of cannabis. We had built that thing before, so we knew what the data model looked like and how businesses looked at selling online and traceability data, and we thought we knew what the future would look like, so we started building it early on at Cresco, piece by piece, with a really small team — we don't have a large engineering team, we have a small team of 10x-ers, that's what we have. You have to create a development culture that endorses and rewards excellence — that's what we've been able to build. Over the last two years we also added product capability, because engineers by themselves can be dangerously good and dangerously idle if they don't know what to build, and our product team has really leveled us up in making sure we're building the right thing at the right time and pace. As for exactly what code, libraries, or technologies — that's an ongoing evaluation, largely a preference of our distinguished, lead engineers, who lean towards being a Node and Java shop, a JavaScript shop, but we evaluate case by case. We use a lot of technologies that a lot of people use, like AWS — it's not proprietary, it's really what you build. Going back to compounding product interest, if we'd had this conversation four years ago when COVID started and we were launching e-commerce, we didn't know if it was going to be amazing and we were stressed out. Now we're on the other side of that — it did work, and 80-plus percent of our customers on Sunnyside are through our website. Kellan Finney: So a lot of that impact and your KPIs within that success rate are measured on downstream impact, like increasing sales — similar to how Amazon handles products that don't instantly increase revenue. Do you handle your tech projects with similar KPIs to reward those team members? Zach Marburger: Absolutely, and giving credit to a large part of our effort that's not as sexy from a building perspective but absolutely necessary for us to operate as a functioning corporate entity — we do the non-sexy stuff too. We just implemented a huge procurement system that allows us to buy things in the most compliant, cash-controlled way, versus a less controlled purchasing environment. There's traditional corporate operations stuff we have to participate in and be excellent at if we want the balance sheet we desire. That stuff is at least half the effort as well. Bryan Fields: After you achieve that first real success internally, people start saying, hey, Zach's team can build a tool for that. Is that across the entire operation, or is it focused on retail, cultivation, processing? Walk us through the inner workings. Zach Marburger: On the wholesale side, I'll speak in percentages to keep it light — we probably do 50% of the business ourselves and the other 50% we elected to go with a vendor very strategically, and that's CanX. I don't mind naming them — they're a cannabis system in the space, fantastic, a really development-driven company in our view, a bunch of engineers who built a really good product that does what we need. On retail, we cover about 90% of the base ourselves, and the other 10% is Dutchie for us right now, and we have aspirations to merge those into one full-scale ERP — that's our intention and goal. On the corporate side we have a lot of systems just like any real corporation — our marketing team uses oodles of things, we have a sophisticated customer growth engine where people text, email, and do everything related to marketing to bring customers into the store. On our legal side we recently got some really slick technology for applications and otherwise. At my last glance we interface with about 75 SaaS systems in total, some more material than others. When people think we can build them something useful, that means our team's skills are in high demand, and that's the current and ongoing sentiment here. For us it's about prioritizing — that list is long, so make sure you do it in the order the business desires, and that's the art and science we go through daily. Bryan Fields: 100 percent. The reason I wanted to bring that up is I think there's a ton of tools everyone's familiar with on the retail side, but the internal operations — cultivation and processing — is part of the growth of the industry, where more adapted tools can lead to more efficient operations and better margins. If you connect cultivation and manufacturing systems, you get better visibility into how strains are yielding and how they're going through processing. Are there tools your team has built to give you a leg up there? Zach Marburger: We have, and it's largely driven by the ops team's desire to know that stuff — we're not involved in the business acumen or decisions they have to make. We have a couple of really slick, technically inclined folks on that side who say, hey, I want to understand performance across variants, sizes, strains, everything — the assumption being we drive positivity on the next inputs and outputs going through the model. We have great trackers and measurement capabilities in that arena that drive product innovation, and I think we have some pretty great products and share in a lot of the markets we're in, driven by folks who have a real desire to make great products and use technology to their advantage across the chain, including manufacturing. Kellan Finney: I'm so glad Bryan asked that this morning — I was thinking about our conversation while getting coffee at Starbucks, using the app, and thinking about how ordering that coffee updates inventory within their ERP, helping manage their entire supply chain of coffee beans. I was wondering if that's exactly where Cresco is going with your ERP — integrating retail all the way back to cultivation. If so, is there value in that real-time information from the retail location feeding the cultivation and manufacturing teams? Zach Marburger: At the end of the day you're really just managing cost and revenue as quickly as possible, so you never have more or less on hand than you need, at the operating margin you desire. For us, connecting those data streams as closely to real-time as humanly possible lets us know what we didn't know previously. In traditional systems like SAP or Oracle, that's what they do — they let you dial in exactly how much of an input you used against your latest estimate, know what you have on hand, execute purchase orders, and operate with certain EBITDA and margin expectations. That's where we're trying to get to in cannabis, because the ultimate data obligation is seed-to-sale — melding those worlds closely together. We do it a certain way, some competitors do it differently but intelligently, and some don't do it at all because they don't feel it's necessary for their scale. Our expectation is to get to a very traditional ERP-like lifecycle where you purchase intentionally with full-scale controls and understanding of turn and quality of inventory, feeding back into even more intelligent decisions next time. We're not there yet, but that's absolutely the goal, and we'll likely still tap outside vendors to complete the mission. Bryan Fields: Kellan made it seem effortless because it's Starbucks, but internally there need to be excessive tools communicating with each other, and some of these tools aren't as commonly found in cannabis as they are in oil and gas or food and beverage. I'm assuming you've had to bring those tools over to get the same level of information other manufacturers and cultivators use to make better decisions. I want to highlight the online-versus-in-store number — did I read that correctly, 85% of your retail purchases are online? Zach Marburger: I believe in our last investor presentation it was 80 to 85 percent, and it depends on the market. That's driven, in my view, by the fact that we really started in Colorado as a retailer — when Sunnyside and our key markets moved from medical to adult use, you see step-function drives in revenue and we opened up several stores. For us that's continued because of our investment into Sunnyside dot shop. We set the tone that throwing up a website with an iframe wasn't going to do it — you need to deliver a better customer experience, because people buy so many things online now that if you compare the experience to a big retailer or an upstart DTC company, a quality difference of significant proportions will make you struggle. If you think about a consumer's purchasing process, most of it happens online — you're on the website, looking at the product, maybe going to Leafly, maybe back to Google, bouncing around — and when you're in the store it's a very short period of time, though it should be great, and our staff does wonderful things with education and upselling, and the stores are beautiful. But if you look at time spent, the majority is online, so if that's not great at conversion, you've got an issue. We knew that early on and doubled down when COVID came, developing an operational back end we're proud of and confident in, so people don't have to wait forever in the store to get their orders. Our goal is to get people in and out quickly. It's largely due to behavioral and macro shifts, and the staff — if you've been into a Sunnyside in an adult-use market, the staff is impressive, a machine, really engaged and able to make it work, and that's what drives the numbers. Kellan Finney: For future consumers — like alcohol, cannabis is similar in that the majority of purchases come from chronic users who use every day and know what they want, versus first-time or educational experiences being cultivated in retail locations. Looking through the crystal ball, do you think most stores will eventually become fulfillment centers because the majority of transactions will happen with people who already know what they want? Zach Marburger: I think there's a really interesting geographical dependency — as long as states remain fragmented, legal and illegal, and neighboring border stores matter, we have border stores, some competitors have border stores, and those stores perform at a high level tied to their geographical location. That withstanding, I think a lot of foot traffic happens in high-density areas, but even then folks tend to behaviorally prefer online for a myriad of reasons. Our goal is to do both to the best of our ability. If you went to the store closest to our HQ in River North, you'd see a very non-cannabis-like experience — you almost feel like you're in an apparel shop, accessories galore, free-flowing, no waiting room, bright, not forced in any direction. If you walked in off the street you'd feel like you could still be served and accomplish what you came for, which is distinct from a lot of cannabis businesses. You still can't really see the product in some states, which is strange — you walk into Walgreens and you see what you're going to purchase, but in cannabis that piece may change over time, allowing a different interaction and normalization with the product. There's psychology involved with some of the compliance rules driving some of that. Market to market, border stores withstanding, I think those conditions push folks to lean towards online more often than not. Bryan Fields: I have to ask about AI tools — with the rise of ChatGPT, is your team bringing AI in-house? Have you found a segment internally that's most adaptable, one you're most excited about? Zach Marburger: I think long term AI will definitely be part of our stack. Today, some of the most common ways we use it are through co-pilot programming tools that speed up our engineers' workflows, and on the finance side we have existing tools enriched with AI capabilities rather than adopting a whole new platform — the same platform just gets AI capabilities added. That lets us do things like process invoices faster; when you're a really big company you have oodles of invoices and purchase orders, and it becomes a big job to manage that pile, so we've seen cost reduction and efficiency improvements in document processing across the board — that's low-hanging fruit, and something we've focused on first. I see future use cases being more intel-driven — why does a customer need to build a cart themselves if you know what they want? Could that cart be optimized to your inventory position and their preference simultaneously, or messaging and pricing strategy suited to people with the highest price elasticity? In the future AI will make those decisions faster, which is hard to do right now even with existing LLMs. The tools coming out are amazing, and our marketing folks use them for generative content too, but we're probably inning one and a half — too early to tell, but very excited. Bryan Fields: The reason I ask is your team seems to have the propensity to take on newer technologies and experiments where other companies might be more hesitant, because you need the right mindset and culture. I could see a scenario where a really big yield on a new plant strain triggers an insight that maybe you should change operational plans and grow more of that strain going forward — the kind of forward-thinking, riskier approach your team seems focused on. Zach Marburger: Yeah, I'm really excited about AI getting eyes, which just happened recently, and integrating it into camera systems. I've seen cool stuff on the security side that could reduce cost and increase visibility, and you can imagine what it could do for plants if it had eyes and could see everything all the time — you could engineer it to look for something specific and do a lot of interesting stuff. Cameras are really interesting for us and for cannabis right now because those companies are innovating so quickly with AI that we can use our existing infrastructure — no capex, just some opex into software — to get a step function in capabilities. There was a company a couple years ago integrating cameras to analyze plant health and chlorophyll or something specific, I think they were acquired, but that kind of optical image analysis for plant health and nutrients is definitely being applied to the industry right now — imagine integrating IR cameras simultaneously, and then AI controlling your HVAC — it'll be lights-out AI for everything at some point. Cautiously optimistic right now because it's still mostly chat interfaces, but the capabilities are going to be big. For us it's prioritization — we'd build a lot with it if we could, but we're still figuring out exactly what it is for us. Bryan Fields: Is there a particular issue your team is struggling with that you'd say, hey, if an entrepreneur built this solution it would make a big difference for us? Zach Marburger: Good question. Right now a couple people in the space are attempting to build what I'd call a SKU system. In the non-cannabis world, everything at a grocery store has a barcode and a SKU. In cannabis there's really no subscribed-to or required SKU system, so each company has its own identification system for products, and you get really fragmented data. All the data and analysis providers in our industry — Headset, BDSA, or catalog providers who provide great macro capabilities to operators like us — there's no SKU, so I don't know where to go to say, that's this exact product identified by these standards with these metrics. Data is tough, it's all kind of messy as a result. A couple of people are working on a SKU system, as simple as it sounds — I'd put Lucid Green in that bucket, they have a really interesting system. If someone solved that, it would help and support everybody. Also refining the customer experience — messaging and acquiring customers — is always a good place to play. There are real disadvantages cannabis companies have; it's really hard for us to text people, so a couple companies have come up with clever ways to put a wallet pass in somebody's phone so you can send push notifications instead of texts. Things like that, that let us reach customers compliantly and develop new customers, are a really underserved area, since we can't participate in Facebook ads and things like that. Bryan Fields: What is the most expensive lesson you've ever learned? Zach Marburger: Luckily we avoided one with CrowdStrike — we're not a CrowdStrike customer, so that didn't make the list. Long ago, in my first company, called Topple Track, we did copyright protection and management — we worked with several thousand record labels, basically like Google janitors, filing large legal takedown notices at scale whenever piracy results showed up. In the early days we were going through a physical-to-cloud transition and had a big meeting the next day with what became our biggest integrator, and the most trivial programming circumstance — maybe a misplaced comma — ended up causing a full-on reset, starting from scratch, because we were flying by the seat of our pants. That demo didn't go as planned and may have set us back a couple months. It wasn't that big a deal in the end, but I still remember the looks on people's faces — it wasn't what I intended. Quite the learning lesson about being prepared. Bryan Fields: What question do you wish more people asked you at Cresco or in cannabis? Maybe what you think is next? Zach Marburger: I think technology, in many instances, not just in cannabis or here at Cresco, is often brought along for the ride instead of being at the front of consideration when you're not a technology company — it's like, oh, tech will just solve it, you just put things together and it works. Sometimes that's not the case. So just asking, hey, what do you think about this, as early as possible, would serve any organization well, because most of the time it doesn't work right away, and second-guessing is a terrible way of making decisions. Bryan Fields: Prediction time, Zach — what percentage of cannabis companies do you think will follow Cresco and start integrating proprietary software stacks, and what's the biggest benefit most people don't recognize? Zach Marburger: I think a lot of our peer set is already doing this to some degree. On the retail side, I think at some point you may see over 50% of the market have some sort of custom shopping interface, not the boilerplate approach of today. What's unique about cannabis is we're not as impacted by the marketplace syndrome the way food delivery is — most customers don't go to marketplaces to order, they go directly to the dispensary because that's where the loyalty program and discounts are, with some cannibalization from services like Google Maps and Yelp. So investing in a custom retail environment is more common in our industry. On the wholesale and manufacturing side, I think it's less likely — we partner with folks like CanX to get the whole job done, and I think those supply chain companies will have predominant market share because it's such a challenging product. Kellan Finney: I agree with Zach on retail. I think infrastructure will follow more of an automotive approach, like Tesla making drivetrains for a lot of electric cars — I could see a company like Cresco coming out with an ERP system that works so well they diversify revenue and sell it to other MSOs while maintaining their own proprietary solution. But on the front end, I think it's a no-brainer for everyone to own their retail website and control that consumer experience, given the psychological games being played. What do you think, Bryan? Bryan Fields: I think having the mindset and willingness to build tools that don't currently exist, or adapt other tools, is a real separator for companies. People don't recognize how challenging it is to have six businesses wrapped into one, communicating with each other, in just a single state, and to make good decisions from the top you need clear data sets that give you actionable insights. The more you narrow down key KPIs in cultivation and processing, the more it solidifies margins all the way through retail. What your team's doing, Zach, is awesome, and it's fun to dive into the nuance — I can only imagine requests come in like, hey Zach, it'd be really cool if you built this tool, and you're like, yeah, I just need six months and five million dollars, and put it on the list. Bryan Fields: Zach, for listeners who want to get in touch and learn more, where can they find you? Zach Marburger: Best place is probably email, just my name at crescolabs dot com. Bryan Fields: We'll link it up in the show notes. Thanks for taking the time, this was a lot of fun. Zach Marburger: Thanks, guys, have a good one.