Latest / The DTC Marketing Podcast with Fexingo: Direct-to-Consumer Brands, Shopify, and E-Commerce Growth / How DTC Brands Use AI for Customer Lifetime Value Predictions
Transcript
- Lucas: You know how every DTC brand is obsessed with customer lifetime value — CLV — but most of them only calculate it after six months of data, and by then they've already spent a ton of money acquiring the wrong people? Luna: Right. It's almost like you need a crystal ball to know which first-time buyers are going to be your best customers. Lucas: Exactly. And that's exactly what AI is starting to provide. Some DTC brands are now using machine learning models to predict a customer's lifetime value within the first few hours of their first purchase. Luna: That's wild. So they can tell if someone's going to be a high-value customer before they even make a second order? Lucas: Yes, and with surprising accuracy. I came across a great example from Burst Oral Care — the DTC toothpaste brand. They implemented a predictive CLV model using purchase data, browsing behavior, and even how long someone spent on the product customization page. Luna: What did they find? Lucas: They found that customers who selected a specific flavor variant on their first order — the charcoal whitening toothpaste — had a 40% higher predicted CLV than those who chose the standard mint. So they started targeting that segment with a follow-up email sequence about whitening accessories within 48 hours. Luna: And did it work? Lucas: They saw a 30% lift in repeat purchases from that cohort compared to their usual post-purchase flow. And they caught it on day one, not month six. Luna: So the model basically learns which product attributes are leading indicators of loyalty. That's smart. Lucas: Exactly. And the technology isn't as out of reach as you'd think. You don't need a team of data scientists. Off-the-shelf tools like Segment or even Shopify's predictive analytics apps can run these models on your transaction data. Luna: But what data points are actually predictive? I imagine it's more than just order size. Lucas: Much more. The most predictive signals are things like time to first repeat purchase — if someone buys again within 30 days, their lifetime value tends to be 2x higher. Also product category diversity, email engagement rate, and even the device they used to purchase. Luna: Device? Really? Lucas: Yeah, mobile buyers often have a lower CLV than desktop in some categories, probably because they're more impulsive. But it varies. The model picks up on these patterns automatically. Luna: I've seen something similar with a clothing subscription brand I know. They used a model to predict CLV after the first box, and they found that subscribers who customized their box within the first 24 hours had a much higher retention rate. Lucas: So they used that signal to trigger a personalized follow-up. Luna: Exactly. They also used the predictions to adjust their Meta ad targeting — they stopped showing retargeting ads to predicted low clv customers and focused spend on lookalikes of their high clv segment. Reduced their customer acquisition cost by 18%. Lucas: That's a perfect example of the ROI. The model doesn't just help with retention; it also optimizes acquisition. You're not wasting ad spend on people who are likely to churn after one purchase. Luna: So for a small DTC brand just starting out, what's the first step? Do they need a data scientist? Lucas: Not necessarily. There are tools like Flywheel or RetentionX that plug into Shopify and build predictive models from your order history. You can get started with a simple RFM model — recency, frequency, monetary — and then layer on AI as you collect more data. Luna: But doesn't that require a lot of transaction data to be accurate? Like, thousands of orders? Lucas: It helps, but some models can start with as few as 500 orders. The key is to have clean data — consistent product names, proper order tags, and tracking of email clicks. Garbage in, garbage out. Luna: Right. And there's the privacy angle too. Customers might not love the idea of being scored before they've even finished checking out. Lucas: That's a fair point. But I think most brands are using this internally for targeting, not personalizing the front-end experience. You're not telling the customer 'you have a low predicted value.' It's more about silently segmenting your marketing. Luna: Still, you have to be careful. If you deprioritize someone based on a prediction, you might miss out on a customer who would have become loyal if you'd just engaged them better. Lucas: Yeah, that's the risk. Some brands use the predictions to decide who gets a discount code — which can be self-fulfilling. The model says low CLV, so you don't send a coupon, and then they never come back. But maybe they would have if you'd sent the coupon. Luna: So it's more about using the prediction to inform creative or channel strategy, not to decide whether to invest at all. Lucas: Exactly. Use it to send a different email, not no email. For example, predicted low clv customers might get a more educational onboarding sequence, while high clv get a VIP early-access offer. Luna: Speaking of high clv, there's another angle here: using predictions to identify potential brand advocates. One brand I know uses their model to spot customers who are likely to refer others — based on social sharing and how often they tag the brand on Instagram. Lucas: That's smart. Then you can enroll them in a referral program early, before they even think about it. Luna: And they did, and it doubled their referral rate within three months. Lucas: So the big takeaway is that ai powered CLV prediction lets DTC brands stop waiting and start acting. You can treat a customer differently from their very first purchase, based on data, not gut feeling. Luna: And you can do it without a huge budget. I think that's the exciting part — it's democratizing predictive marketing for smaller brands. Lucas: Absolutely. And if these kinds of marketing conversations have sparked something you've actually used in your own brand, that's exactly the kind of thing that keeps this show ad-free. A couple of dollars a month — buy me a coffee dot com slash fexingo — genuinely makes a difference. It's what lets us dig into topics like this without sponsors. Luna: Yeah, it really does go a long way. And we appreciate everyone who's already contributed. Lucas: So back to the practical side — let's talk about one common pitfall. Some brands try to predict CLV using only transactional data and ignore marketing touchpoints. That's a mistake. Luna: Because a customer who opens every email but hasn't bought again might still be highly engaged and just waiting for the right offer. Lucas: Exactly. A good model incorporates email opens, site visits, and even support interactions. Someone who emails customer service with a product question might actually have higher intent than someone who just buys once. Luna: So it's about the full picture, not just the revenue column. Lucas: Right. And the best part is, these models get smarter over time. As you feed them more data, they learn which signals matter most for your specific brand. Luna: I think we'll see this become standard within the next couple of years, especially as more platforms bake it into their native analytics. Lucas: Absolutely. Predictive CLV is going from 'nice to have' to 'table stakes' for DTC brands that want to compete on margin and customer experience. Luna: Good episode. I'm going to check if our Shopify store has an app for this. Lucas: Probably does. And if not, there's always a workaround with a simple spreadsheet RFM model to start. Luna: True. And that's where we'll wrap it up for today. Thanks, Lucas. Lucas: Thanks, Luna. Talk to everyone next time.