Latest / The CMO Podcast with Fexingo: Marketing Leadership, Budgets, and Executive Strategy / How CMOs Turn Customer Data into Personalization at Scale
Transcript
- Lucas: So, Luna, I've been digging into a fascinating case from a major UK retailer that's been using machine learning to personalize their loyalty emails. And the headline number is a twenty percent lift in repeat purchases just from better targeting. Luna: Twenty percent is significant. But I'm guessing the story is more about how they got there than just the result. Lucas: Exactly. It's about moving from segment-of-one in theory to actually doing it across millions of customers. And the interesting part is they started small — with just one email campaign — and scaled from there. Luna: Before we dive deeper, I want to mention something that's been on my mind. We're able to have these conversations because a small group of listeners chip in monthly through buy me a coffee dot com slash fexingo. That's what keeps this ad-free and independent. Lucas: Yeah, and it genuinely matters. It's not a huge number, but every contribution helps us keep going. And if these marketing conversations have sparked something you've actually used, that's the kind of support that makes a difference. Luna: Exactly. But back to personalization at scale — what was the starting point for this retailer? Lucas: They had a classic problem. Their loyalty program had millions of members, but the emails they sent were what you'd call batch and blast. Everyone got the same offer, whether they were a weekly shopper or a once-a-year Christmas buyer. So the first step was consolidating all their customer data into a single view. Luna: That sounds like a customer data platform, right? But I imagine that's easier said than done. Lucas: Totally. They had data scattered across their e-commerce platform, their loyalty app, even in-store point of sale systems. It took about nine months to get everything clean and unified. And the key insight was that data quality mattered more than the model they used. Luna: Which makes sense. Garbage in, garbage out. But once they had that clean data, how did they actually decide what to personalize? Lucas: They started with a simple machine learning model that predicted each customer's next most likely purchase. Not based on what they bought last, but on patterns across thousands of similar customers. For example, if someone bought a winter coat last January and a scarf in February, the model might predict they're due for a new coat in November, and offer a targeted discount. Luna: So it's not just 'you bought this, so here's more of that.' It's anticipating the next need. That's a different mindset. Lucas: Exactly. And the twenty percent lift came from moving from a generic send to this predictive model. But there's a cautionary tale too. They initially tried to do everything at once — personalizing subject lines, images, send times, offers — and it was a mess. So they pulled back and focused on just the offer and the product recommendation. Luna: That's a good lesson in prioritization. What did they learn about the trade-off between personalization and privacy? Lucas: They were very careful. They used purchase history and browsing behavior, but they explicitly avoided using any sensitive data like health or financial information. And they gave customers a clear way to see what data was being used and to opt out. Luna: I think that's the ethical line every CMO has to walk. Personalization can feel creepy if it's too specific. How did they avoid that? Lucas: They tested relentlessly. For example, they tried a campaign that referenced a customer's last purchase in the subject line, like 'Your recent order is ready for a refill.' That saw a high click-through rate but also a higher unsubscribe rate. So they dialed it back and used a softer approach. Luna: So the lesson is to measure not just engagement but also the long-term relationship. That's something a lot of brands miss. Lucas: And that's why I think this case is so instructive. They scoped it as a series of experiments, not a big-bang launch. They started with one email stream, proved the ROI, and then expanded to other channels like push notifications and on-site recommendations. Luna: How long did that take? I imagine the internal buy-in process is a whole other hurdle. Lucas: It took about eighteen months from start to scale. And the biggest hurdle wasn't the technology — it was the marketing team's mindset. They were used to writing one email and sending it to everyone. Now they had to think in terms of hundreds of variations, and trust the algorithm. Luna: Trusting the algorithm is hard when your career depends on the open rates. But what about the data silos you mentioned? How did they get the whole company on board? Lucas: The CMO played a huge role there. She had to work with the CIO and the chief customer officer to align on a single source of truth. And she made a compelling business case: the projected lift in customer lifetime value was enough to justify the investment. Luna: So the lesson for CMOs is to think beyond the campaign and focus on building a long-term personalization engine. But what would you say to a CMO who's just starting out? Lucas: Start with one use case that has a clear revenue link. For this retailer, it was email because they could measure it easily. But for a B2B company, it might be personalized account-based marketing. And don't try to build a perfect model upfront — iterate and learn. Luna: That's pragmatic. And one thing I've noticed is that personalization often gets confused with just using a customer's first name in an email. This is a whole different level. Lucas: Right. Using a first name is table stakes. True personalization is about being relevant to the customer's current context. And with the rise of AI, the possibilities are expanding fast. Luna: What about the role of human creativity in all this? Algorithms can suggest, but the final message still needs to sound human. Lucas: That's a great point. The retailer had a team of copywriters review the ai generated subject lines and offers. They found that the model was good at picking products, but the human touch was needed for the tone. So they created a hybrid approach: AI for the logic, humans for the voice. Luna: So it's not about replacing marketers, but augmenting them. That's a reassuring message for our audience. Lucas: Absolutely. And I think the next frontier is personalizing the entire customer journey, not just the touchpoints at the bottom of the funnel. This retailer is already testing personalized homepages and even personalized pricing — though that last one is tricky. Luna: Personalized pricing feels like a minefield, but I'm curious. How are they handling the fairness angle? Lucas: They're being careful. They only offer personalized discounts, not different base prices, and they cap the discount range to avoid alienating customers. It's a fine line, but they're testing it in a small segment. Luna: It's a bold move. And it shows how far personalization has come. But I keep thinking about the smaller companies that don't have the data science teams to build these models. What's their path? Lucas: They can use off-the-shelf tools that have built-in machine learning, like a customer data platform with predictive scoring. Many of these tools are now accessible to mid-sized businesses. The key is to start with clean data and a simple use case. Luna: And I'd add that you don't need to predict everything. Sometimes a simple rule like 'if a customer hasn't bought in 90 days, send a win-back offer' can be a huge win without any sophisticated model. Lucas: That's a perfect example. The smart CMO knows when to use an algorithm and when to use a simple rule. And that judgment comes from experience and testing. Luna: So, to wrap up, the key takeaway is that personalization at scale is achievable, but it requires a solid data foundation, a clear use case, and a willingness to test and learn. Lucas: And it's not just a marketing project — it's a company-wide initiative. The CMO has to be the champion, breaking down silos and proving the value. If they do that, the potential is enormous. Luna: I'm already thinking about how I can apply this to my own marketing projects. Maybe it's time to start small with a simple predictive model. Lucas: That's the spirit. And that's a wrap for today. We'll see you next time.