Latest / The DTC Marketing Podcast with Fexingo: Direct-to-Consumer Brands, Shopify, and E-Commerce Growth / How DTC Brands Use Predictive Analytics to Forecast Demand
Transcript
- Lucas: There's a skincare DTC brand that, eighteen months ago, was holding sixty days of inventory on a product that only sold through in fourteen. And they fixed it with a spreadsheet and a regression, not a billion-dollar SAP deployment. Luna: A spreadsheet? That's a bold claim for a predictive analytics episode. Lucas: Well, the spreadsheet was the interface, but underneath it they had a simple machine learning model that weighed seasonality, ad spend, and a little bit of social sentiment. The point is, they didn't need a data science team. They needed the right question. Luna: And I guess the question is, how much inventory do we actually need to buy? That's the million-dollar question for any DTC founder. Lucas: Exactly. And if you're listening and these kinds of operational questions get you thinking, you know, the show you're listening to is ad-free, and that's only possible because some listeners toss a few bucks our way. If these marketing conversations have sparked something you've actually used in your own brand, you can support the show at buy me a coffee dot com slash fexingo. Luna: Yeah, it's a small way to keep the lights on. And we appreciate every bit. Lucas: Alright, back to that skincare brand. They were sitting on a product that had a shelf life, literally. And the cost of overstock wasn't just storage, it was write-offs. So they built a forecast that updated every week. Luna: What variables did they start with? I'm guessing just last year's sales and a hunch. Lucas: They started with twelve months of daily sales, ad spend by channel, and a few seasonality flags. Then they threw in Google Trends data for the ingredient in the product. Sounds weird, but it gave them a leading signal. Luna: So they were basically predicting demand from search interest. That's clever. Did it actually work? Lucas: It cut their dead stock by thirty percent in one quarter. But the more interesting thing is what it didn't do. It didn't predict the viral TikTok video that drove a two-hundred-percent spike in orders. No model catches that. Luna: So we're back to the old tension between data and gut. But I think a lot of DTC brands are overcorrecting. They either ignore analytics entirely or they worship the model. Lucas: That's exactly the tightrope. And there's a real cost to over-relying on historical data. If you're a brand that launched a year ago, your history is basically a pandemic and a supply chain crisis. The past isn't a great guide. Luna: Right, so what's the alternative? How do you forecast when you have no history and no clear trend? Lucas: You look at proxy data. Cohort behavior from similar brands, category growth rates, even macroeconomic signals like consumer confidence. One DTC furniture brand I talked to uses housing starts as a leading indicator. It's not perfect, but it's better than guessing. Luna: Housing starts for furniture, that makes sense. I wonder if they also factor in return rates. Because if you're forecasting net demand, returns are a big swing factor. Lucas: They do now. They actually built a separate model just for returns, because their return rate varies by product and by season. That's the kind of nuance that separates a useful forecast from a toy. Luna: So what's the minimum viable product for a small brand? Do they need Python and a data engineer, or can they start simpler? Lucas: Start with a Google Sheet and a naive forecast. That's just, next month equals last month, maybe adjusted for seasonality. You'd be surprised how often that beats a complicated model. The key is to measure your forecast error. If you can't measure it, you can't improve it. Luna: Yeah, we did that for a client once. Naive forecast, and they were shocked at how close it was. But then they hit a product launch and it all fell apart. Lucas: That's the classic failure mode. Models are terrible at discontinuity. Launching a new product, entering a new channel, that's when you need judgment. The model can't help you, and pretending it can is dangerous. Luna: So what's the role of the founder in all this? Are they the final check on the machine? Lucas: They have to be. The best approach I've seen is what one CEO calls 'forecast as a conversation.' The model proposes, the human disposes. You look at the numbers, you apply your context, and you make the call. Luna: I like that. It keeps the human in the loop. But what about the ethics of all this data collection? To predict demand, you're tracking a lot of customer behavior. Where do you draw the line? Lucas: That's a great question, and it's one that a lot of brands are wrestling with right now. You can forecast demand without tracking individuals. Aggregate trends, anonymized cohorts, that's enough for most forecasting. The moment you start predicting individual behavior, you're in murky territory. Luna: And the privacy regulations are only getting stricter. So maybe the smart play is to build demand models on aggregated data and leave the personalization to the marketing team. Lucas: Exactly. And that's the thing, the marketing team and the supply chain team often live in different worlds. But predictive analytics is where they collide. The same data that predicts a spike in demand also tells you which creative to push. Luna: So how does a DTC brand actually get started? I imagine there's a lot of hype and a lot of shiny tools promising AI magic. Lucas: There is. But I'd say the first step is to pick one SKU, the one that gives you the most headaches, and build a simple forecast for it. Track your error for a month. Then you'll know if you need more sophistication. Luna: That's a practical first step. And it doesn't require a data science degree. I like that. Lucas: And once you have that baseline, you can start layering in external data. Google Trends, economic indicators, even weather data for some categories. The weather is a huge predictor for fashion and outdoor gear. Luna: Oh, that's a good point. I remember a brand that sold rain jackets and they didn't check the forecast. They had a terrible season. It was almost embarrassing. Lucas: Right, and that's not even predictive analytics, that's just checking the weather. But the point is, the data is out there, you just have to know where to look. Luna: So, if I'm a DTC founder and I want to get better at forecasting, what's the one book or resource you'd recommend? Lucas: I'd actually recommend reading about the history of forecasting. The story of how weather forecasting became a science is a great lesson in humility and incremental progress. And it applies directly to demand forecasting. Luna: That's a smart answer. It's not about the tool, it's about the mindset. Lucas: Exactly. And the mindset is, you're always wrong, but you can be less wrong. The goal isn't perfection, it's reducing uncertainty just enough to make better decisions. Luna: And that's something any brand can start doing today, with a spreadsheet and a little thought. Lucas: Yeah. And the brands that do it well will have a real edge, especially as we head into a period where consumer behavior is shifting faster than usual. Being able to adapt your inventory quickly is a huge advantage. Luna: So, what's the one thing you want listeners to take away from this episode? Lucas: That predictive analytics isn't about predicting the future. It's about quantifying your uncertainty. And that's something any DTC brand can start doing, no matter their size.