AI Customer Segmentation: How to Build Segments Around What Customers Will Do Next

AI customer segmentation uses machine learning to predict what each customer is likely to do next – churn, buy again, spend more – and groups customers by that predicted outcome. Every customer in the segment carries a score for how likely the outcome is.

If you run a CRM or lifecycle program, your segments probably already update themselves every day: a VIP list, a lapsed list, a first-time-buyer list. None of them was built to spot the VIP who’s about to leave, and that’s the gap AI segmentation fills.

You don’t need to throw out those rules to use AI. AI segmentation adds a second kind of segment alongside them, built around what you want to happen next.

By the end, you’ll know which AI segment I’d advise you to build first, and how to tell whether it’s working.

AI customer segmentation is a move-on from rule-based segmentation, which selects the customers who’ve taken a set action, such as spending $500 in the last year or it’s 90+ days since the last order. It then treats everyone inside that segment in the same way. With AI customer segmentat, you select customers on a predicted outcome instead, such as “likely to churn,” and each of them comes with a score, so you can see who’s very likely to leave and who’s only borderline.

To make those predictions, AI segmentation works from the signals you’re probably already collecting:

• Purchase history, frequency and recency
• Browsing behavior and engagement
• Product affinity
• Customer lifetime value
• Channel preferences
• Store purchases, loyalty status and service history, where you’ve connected them

If you’d like the groundwork first, our glossary covers customer segmentation and, more specifically, predictive segmentation.

Take a specialty coffee retailer whose CRM team runs a VIP segment for anyone who’s spent $500 or more in the last 12 months. The segment updates every day, and it does exactly what the team built it to do.

One VIP spent $620 last year across ten orders. Her recent orders have been getting smaller, though, and she hasn’t bought anything in seven weeks, so the VIP email lands in her inbox the same week she starts buying beans from another roaster.

Another customer bought one bag in April and a grinder in May, then came back for a second bag the week after. She’s still under the line, so she hears nothing from the VIP program at all.

The problem is, the rule is doing what it was built to do: it answers who has spent, but it can’t pick out who’s about to leave or who’s about to spend more.

Relevance is already a gap for customers. In SAP’s Global Engagement Index 2026, 58% of consumers say they “think that most marketing emails they receive aren’t relevant.” A spend rule keeps her in the VIP segment, so she keeps getting VIP emails while she tries a new roaster. What she needs is a win-back message before she settles in, but a spend threshold puts her in the same segment as the VIP who’s still buying every month.

City Beach, an Australian youth fashion retailer, made the same move. Its marketing team had been segmenting customers by lifecycle stage, lifetime value and RFM (recency, frequency and monetary value), and as the business grew, the team moved to a predictive marketing approach.

To find the same gap in your own program, look for segments where a spend or recency line stands in for a question about what the customer will do next. Those are the segments worth predicting.

Build predictions on the whole customer record

A prediction is only as good as the customer record behind it. If your record holds opens and clicks but not what customers bought in your stores, how they use your loyalty program or what they’ve told your service team, you’re building predictions on half the story.

SAP Engagement Cloud connects customer data across CX and operational systems, including offline sales, store purchases, loyalty status and service data, alongside engagement data. It also gives your team AI predictions of what each customer is likely to do next, such as buying or churning.

Our guide to AI marketing use cases goes further into why the data behind AI matters.

City Beach worked in that order. The team set out to merge each customer’s online and offline shopping behavior and purchase history into one profile, then brought in their loyalty program, points of sale and customer service touchpoints, and only then turned to AI to predict churn.

Mike Cheng, City Beach’s Head of Digital, describes how they approached it:

If your customer data sits in five systems that each hold a different version of the same customer, I’d fix that before building any AI segment on it.

Predict customer behavior

A predictive model gives each customer a score for an outcome, such as how likely each one is to buy or churn, what they’ll be worth over time or which channel they’ll respond to. SAP Engagement Cloud’s predictive real-time segmentation builds segments from those predictions.

You’ll also come across clustering, where AI finds groups of similar customers in your data. That can be interesting, but then your team still has to work out what to do with each one. There’s more on why that matters when you pick your first segment, further down.

Create dynamic customer segments

Both kinds of segment update automatically as customer data changes, which is what dynamic segmentation means. A rule-based segment moves a customer when they cross your threshold, whereas an AI segment moves them when their score changes, so someone can join “likely to churn” before crossing any line you’d set.

At the coffee retailer, the April customer would join “likely high spenders” the week she came back for her second bag, while she was still well under the VIP line.

Activate segments across customer journeys

A segment starts paying off once your team puts it to work in a campaign or journey. In SAP Engagement Cloud, pre-built tactics use AI segments to define audiences and drive channel selection based on each customer’s likelihood to engage in each channel.

Traditional segmentation stays useful, but AI adds a different way to select customers: on what they’re likely to do next, with a score for each one.

Some segments should stay rule-based:

• consent and suppression lists
• regions and languages, and
• the loyalty tiers your customers can see

The above need to be exact and easy to explain. Use AI segments where the question is what a customer will do next, such as who’s likely to churn, who’s ready to spend more, and who’s likely to buy again.

So where should you start? Pick a call your team already makes on a regular schedule, like who gets this month’s retention offer, and build your first AI segment to answer it. 

A cluster labeled by its averages – mid-spend, weekend browsers, slightly lapsed – describes a group of customers, but your team still has to decide which campaign it gets, which offer, and whether to act at all. 

When you start from the decision instead, every customer in the segment comes with a reason to act and a score for how likely the outcome is. Each of these five segments is built around one prediction:

To see how the first one plays out, go back to the coffee retailer. Its CRM manager is planning next month’s retention offer, and the standing rule sends it to everyone who hasn’t bought in 45 days, usually around 6,000 customers.

That list probably includes a thousand or so regulars who reorder every six weeks, on the same list as the customers who are drifting away. So she starts from the decision: who should get the retention offer? The outcome that answers it: which customers are likely to stop buying in the next 30 days.

While she sets the segment up, her team connects the data that prediction needs, from order history and returns to service contacts. By send day, the segment is built and reviewed, and the offer goes to the customers most likely to lapse.

The lapsed-but-interested segment is a win-back segment you build from recent browsing and purchases. For tier upgrades, SAP Engagement Cloud’s customer loyalty solution has pre-built tactics, and our guide to lifecycle marketing covers segments by relationship stage.

A score for every customer

Before AI segmentation, your team wrote the thresholds, tuned them every few months and made an educated guess about who was slipping. With AI segments, each customer carries a likelihood score, so you can set the cutoff for who’s in and vary the offer inside the segment: the customers most likely to churn get your strongest offer, and borderline ones get a reminder.

Instead of debating whether the lapsed line should sit at 60 days or 90, your team decides where the score cutoff goes and which offer each band of the segment gets.

In SAP Engagement Cloud, Customer Data connects store, loyalty and service data with engagement data, Personalization builds segments from AI predictions, and AI Marketing puts them into pre-built tactics.

It also takes some of the dread out of send day. You can see who’s in the segment, and how likely each customer is to act, before anything goes out, so you can check that your strongest offer is going to the customers most at risk of leaving before you spend it.

Spend where it matters

Before, a retention offer went to everyone over the line, including customers who were probably going to reorder anyway. Your strongest offer now goes to the customers most likely to leave. To see how many of them it keeps, run the holdout test in the build plan below.

You use the AI segment to decide who should hear from you, and personalization to decide what each of those customers receives. AI segments can shape:

• Product recommendations
• Content and offers
• Journey paths
• Timing and channels
• Loyalty experiences
• Re-engagement

Back at the coffee retailer, the April customer could get beans that suit her new grinder and a short brewing guide, while the VIP who’s drifting away gets a reason to come back in place of the standard VIP offer.

For the bigger, enterprise-wide picture, see our article on hyper-personalization, and for the basics, our glossary entry on personalized marketing.

Getting both decisions right matters to customers. In SAP’s Customer Loyalty Index 2026, 60% of consumers say they “are more loyal to brands that make them feel personally valued.” The regular who’s drifting to another roaster won’t get that from the standard VIP email, so she needs both halves: the segment that finds her, and the content that gives her a reason to come back.

I’d set up your first AI segment in six steps. Keep it small: one decision, one segment and one journey.

1. Choose the decision and the outcome to predict. Start from something your team will do differently, like a retention offer, then pick the outcome that should decide it.
2. Connect the data that prediction needs. Bring in the store, loyalty, service and commerce data that sits outside the system your segments run on.
3. Build and review the segment. Select only customers whose recorded marketing consent covers the channel you’ll use. Before the segment triggers an offer, a marketer checks its size, the score threshold and a sample of members, confirms the offer and its limits, and signs it off.
4. Put it to work in a journey. Connect it to the campaign it was built for.
5. Measure it two ways. Hold back a random share of the AI segment from the offer, which shows how much the offer itself changed their behavior. Then compare the AI segment’s results with your old rule’s over the same period, which shows whether the segment does better than the rule.
6. Refine. Feed what you learn back into the threshold and the offer.

I’d treat the review before sign-off as the step you never skip. A marketer scrolling a sample of twenty names might find a customer with an open complaint about a cracked carafe, who shouldn’t get a “we miss you” offer this week, so exclusions like open service cases, recent complaints and suppression lists go into the segment before it runs.

If you want a worked example to adapt, the Personalization Playbook includes a play for launching a new product, where AI builds the audience most likely to be interested in the new offering.

The coffee retailer still sends its VIP offer every month. With scored segments behind it, the customer drifting to another roaster hears from the team before she’s gone, and the April customer gets beans for her new grinder long before she’d have crossed the VIP line.

If you want to try this yourself, pick the one offer your team debates most each month, and make that your first AI segment.