Latest / The Email Marketing Podcast with Fexingo: Newsletters, Open Rates, and Subscriber Growth / How Personalization Algorithms Actually Work in Email Marketing
Transcript
- Lucas: So there's this idea floating around that the more sophisticated your personalization algorithm is, the better your email results will be. And honestly, that's not how it works in most real-world cases. Luna: Yeah, I've seen brands pour months into building some custom recommendation engine, only to end up with something that feels a bit... off. Lucas: Exactly. So today I want to peel back the hood on how personalization algorithms actually function in email marketing, and why a simple rule like 'based on what you browsed last week' is often smarter than a neural network. Luna: Let's start with the basics then. When you say 'algorithm,' what are we actually talking about here? Lucas: At its core, it's a set of instructions that decides which content, product, or offer to show each subscriber. The old school way is rule-based: if someone clicks a category, send them more of that category. The newer way uses machine learning models that weigh dozens of signals — past purchases, time on site, even device type. Luna: And those models are trained on your historical data, right? They learn patterns that a human might miss. Lucas: Right. And the classic example is the retailer who sees that people who buy a certain brand of running shoe also tend to buy moisture-wicking socks. The algorithm spots that correlation and starts recommending socks to anyone who buys the shoe, even if they never browsed socks. Luna: That's where the magic happens. But I've also seen algorithms get it wrong in embarrassing ways. Lucas: Oh, absolutely. Like recommending products from a category the person just returned, or sending a discount for dog food to someone whose dog passed away last month. The model doesn't understand context, it just sees correlation. Luna: So how do you know if your personalization is actually working? Is it about open rates, click-through rates, conversions? Lucas: It depends on your goal. If it's a product recommendation email, then conversion is the number to watch. But if it's a re-engagement campaign, then opens might be the right metric. The trick is to define success before you start, not after. Luna: And then test. You'd run an A/B test between a personalized version and a generic one. Lucas: Exactly. And here's where a lot of marketers stumble. They don't have a clean split between the test groups, or they let the algorithm adjust in the middle of the test, and then the results are meaningless. Luna: So what's a realistic timeline for seeing results from a personalization initiative? Lucas: If you're starting with rule-based triggers, you can see a lift within a month or two. If you're training a machine learning model, you need at least a few months of clean data, and even then, the lift might be modest. The real gains come from continuous testing and refinement. Luna: And on that note, I'm curious about the data side. You need a lot of it, right? What's the minimum viable data for personalization? Lucas: Honestly, you can get started with just three things: email engagement, purchase history, and maybe a little browsing activity. If you have those, you can build a decent rule-based system. For machine learning, you'd want more — time on site, click paths, maybe customer service interactions. Luna: But you don't need to track every micro-movement. That can get creepy fast. Lucas: Right. The creepy factor is real. If a subscriber feels watched, they're not going to trust you. So you have to be transparent about what you collect and why, and give them an easy way to opt out. Luna: And that's a good point. Do you have an example of a brand that does personalization well without being creepy? Lucas: I think Spotify's Discover Weekly is a classic. It's clear that it uses your listening history, but it doesn't say 'We noticed you listened to this song 15 times.' It just presents new music in a nice way. The intent is to help, not to show off its predictive power. Luna: That's a great model. So let's talk about how a typical email marketing team can actually implement this. What's the first step? Lucas: Start with one behavioral trigger. Pick a specific action, like abandoning a cart or browsing a category without buying. Then craft a personalized follow-up email that references that action. Test it against a generic version. Once you see that lift, add another trigger. Luna: And that's where the 'start simple' advice comes in. You don't need a data science team to do that. Lucas: Exactly. And honestly, most email platforms have built-in tools for this. You just need to configure the triggers and the templates. Luna: So what about the future? Are we heading towards hyper-personalized emails that write themselves? Lucas: A little, but I think the bigger shift is towards using AI to generate content variants. Imagine sending a product description that dynamically rewrites based on what you know about the subscriber. That's already starting to happen. Luna: But then it's even more important to keep that human touch. Because algorithms can write, but they don't always know when to be funny or sensitive. Lucas: Right. And that's the real skill — knowing when to let the algorithm take the wheel and when to step in. Luna: Before we go further, I've got to say something. You know, we keep this show ad-free and it's just the two of us researching and recording every week. A few listeners chip in a few bucks a month on buy me a coffee dot com slash fexingo, and honestly, that's what keeps the lights on. It's a small group, but it makes a huge difference. Lucas: Yeah, and we're really grateful for that. If you're listening and it's something you'd consider, the link is in the show notes. No pressure, just know that it helps us keep doing this. Luna: Now, back to personalization. So we've covered the basics. But there's one thing that trips up a lot of people: how to measure the actual impact of a personalized email versus a generic one. Lucas: Right. You have to isolate the variable. Send the personalized email to one segment and a generic email to a similar segment at the same time, then compare metrics like revenue per email. But also look at long-term engagement, not just the immediate click. Luna: So it's not just about conversion. It's about whether that subscriber is more likely to open your next email. Lucas: Exactly. And that's where the real value of personalization shows up. It builds trust over time. A subscriber who feels understood is a subscriber who stays. Luna: And that's the kind of thing that can be hard to measure in a single campaign. Lucas: Absolutely. So if you're thinking about jumping into personalization, my advice is to start small, start with what you have, and keep testing. You don't need a massive data warehouse or a team of data scientists. You need a clear goal and a willingness to learn from the data. Luna: Well, and a healthy respect for the subscriber. Because at the end of the day, it's still a human on the other side of the screen. Lucas: That's the whole game. So next time you're tempted to over-engineer your personalization, remember: a simple, well-tested trigger probably beats a complex algorithm that nobody understands. Luna: And on that note, I think we've given everyone some good food for thought. Thanks for joining us.