Latest / The Retention Marketing Podcast with Fexingo: Email, SMS, and Loyalty for Repeat Customers / How One Brand Uses Predictive Analytics to Predict Repeat Purchases
Transcript
- Lucas: So Luna, we've talked a lot on this show about sending the right message at the right time — win-back emails, exit surveys, birthday data. But there's a more sophisticated layer that a lot of brands are starting to explore, and that's predicting what a customer will do next before they even do it. Luna: You mean like predictive analytics — using past behavior to forecast future purchases? Lucas: Exactly. And I wanted to focus on a specific case for our 50th episode. There's a DTC pet food brand called Ollie — they sell fresh, human-grade dog food on a subscription model. They're not huge, maybe a few hundred thousand customers, but they've invested in a pretty smart predictive system. Luna: And these conversations are only possible because listeners support the show. A couple of dollars a month at buy me a coffee dot com slash fexingo genuinely keeps these episodes ad-free and lets us dig into cases like Ollie's. Lucas: Yeah, it's a small thing that makes a big difference. So Ollie's challenge is classic: how do you keep subscribers from canceling after the first few months, when the novelty of fresh dog food wears off? Luna: Right. So they're not just reacting to churn — they're trying to get ahead of it. Lucas: Exactly. Their data science team built a churn prediction model using three main categories of data. First, order behavior — how consistently does the customer order, are they skipping deliveries, do they ever pause their subscription. Second, engagement with marketing — do they open Ollie's emails, click on recipe tips, watch the educational videos about dog nutrition. And third, customer service interactions — have they complained about the food, asked about portion sizes, or reached out about shipping issues. Luna: So they're basically scoring every customer on a risk scale. Lucas: Right. And the model assigns a score from zero to a hundred — a hundred means the customer is almost certainly going to cancel within the next 30 days. Once a customer hits a certain threshold, say above 80, the system triggers an automated intervention. Luna: And what's the intervention? A discount? Lucas: Not exactly. Ollie found that discounts actually hurt long-term retention for high-risk customers — it trains them to wait for a deal. Instead, they use what they call a 're-engagement sequence.' For a customer flagged as high churn, the system sends a series of three emails over two weeks. The first one is a personalized health tip based on the dog's breed and age — something like 'Did you know beagles often need extra joint support?' The second email shows a testimonial from another customer with a similar dog who saw health improvements. The third email offers a free consultation with a veterinary nutritionist. Luna: So they're adding value instead of cutting price. That's smart. Lucas: And it works. Ollie reported that customers who received the predictive re-engagement sequence were 28 percent less likely to cancel within the next 60 days compared to a control group that got the standard marketing emails. And their repeat order rate — meaning customers who ordered a second or third time after the sequence — went up 34 percent. Luna: Those are pretty big numbers. But here's my question — can a smaller brand afford this? Building a machine learning model sounds expensive. Lucas: It's actually more accessible than you'd think. Ollie built their model internally, but there are off-the-shelf tools now that do something similar. Klaviyo has a predictive analytics feature that scores customers based on email engagement and purchase history. Retention Science offers a platform that predicts churn and automatically triggers SMS and email sequences. Even Shopify Plus has some basic prediction tools if you're on their enterprise plan. Luna: So what's the minimum customer base where this starts to make sense? Lucas: I spoke with a consultant who works with DTC brands, and she said that once you have about 5,000 active customers and at least six months of purchase history, the predictive models start to generate reliable signals. Below that, the sample size is too small to train a model effectively. But once you cross that threshold, the ROI is pretty clear — you're saving customers you would have lost otherwise. Luna: Give me a concrete example for a smaller brand. Say a boutique coffee subscription with 8,000 subscribers. Lucas: Sure. That brand could use Klaviyo's predictive analytics to flag customers who haven't opened an email in 60 days and whose last order was 45 days ago — that's a high churn signal. Then the tool can automatically send a re-engagement email offering a free sample of a new roast. The cost is just the email send and the sample, which is maybe a few dollars. If it saves even 10 subscribers a month at $25 per month, that's $3,000 a year in retained revenue. The tool costs maybe $50 a month. So the math works. Luna: It's basically a no-brainer once you've got the data. Lucas: Exactly. And the key insight from Ollie is that the prediction is only half the battle — you also need the right intervention. Throwing a discount at a high-risk customer can actually backfire. Ollie tested a 20 percent off coupon versus the educational sequence, and the coupon group actually had a higher churn rate after 90 days. Luna: Because it trained them to expect discounts. Lucas: Right. So the intervention has to address why the customer is thinking of leaving. If they're leaving because they don't see the value, you need to demonstrate value — not just lower the price. If they're leaving because the dog doesn't like the food, a discount won't solve that. The predictive model can help you segment which customers need which intervention. Luna: So it's not just about the algorithm — it's about understanding the psychology behind the data. Lucas: Exactly. And that's what makes predictive retention so powerful. It forces you to think about the customer journey in a structured way. Ollie actually found that customers who engaged with the educational content — even if they didn't take the free consultation — were significantly more likely to stay. So the intervention itself is a signal: if someone watches a video about dog nutrition, they're reinvesting in the relationship. Luna: I love that. It's almost like the intervention is a test of who's still interested. Lucas: Exactly. And then you can use that engagement data to further refine your model. Customers who click on the re-engagement email but don't order get a different follow-up than those who ignore the email entirely. It becomes a continuous feedback loop. Luna: So where do you see this going in the next couple of years? Will every brand be doing this? Lucas: I think so. The tools are getting cheaper and easier to use. Klaviyo already offers it as a standard feature. Shopify is building it into their platform. And as more brands adopt it, the competitive pressure will increase. If your competitor can predict when a customer is about to leave and intervene before you even know there's a problem, you're at a serious disadvantage. Luna: So the early adopters have a window of advantage. Lucas: Right. And it's not just about churn prediction. These models can also predict when a customer is likely to make their next purchase, which lets you time your marketing perfectly. Ollie uses a 'next purchase date' prediction to send reminders at exactly the right moment — not too early, not too late. That alone increased their repeat order rate by 15 percent. Luna: So it's not just defensive — it's offensive too. Lucas: Exactly. Predictive analytics can help you maximize lifetime value, not just reduce churn. And that's the real opportunity. Most brands are still reactive — they wait for the customer to cancel and then try to win them back. But the cost of winning back a lost customer is five to ten times higher than keeping them. So if you can predict who's at risk and intervene early, you're saving money and building loyalty. Luna: And it's all based on data you already have — purchase history, email opens, support tickets. Lucas: Exactly. You don't need to buy expensive third-party data. It's all sitting in your CRM and your email platform. The hard part is connecting the dots and setting up the automated triggers. But once you do, the system runs itself. Luna: So for our listeners who want to try this, what's the first step? Lucas: Start with a simple rule-based prediction before you invest in machine learning. For example, identify customers who haven't ordered in 45 days and who used to order every 30 days. That's a basic churn signal. Then set up an automated email sequence that offers help — not a discount, but something like a recipe suggestion or a usage tip. Measure the results. If you see a lift in repeat orders, then consider moving to a more sophisticated tool like Klaviyo's predictive analytics or Retention Science. Luna: So start small, prove the concept, then scale up. Lucas: Exactly. And that's the beauty of this — you don't need a data science team. You just need to think about what signals matter for your business and set up a few simple rules. Then let the data guide you to more advanced models over time. Luna: Alright, that's a solid playbook. Thanks for walking us through it, Lucas. Lucas: Thanks for the great questions as always. And here's to the next 50 episodes. Luna: Cheers to that.