Why customer segmentation is the foundation of every analytics model you'll ever build

Ian Liddicoat is CPO at Kleene.ai. He's spent years building analytics models for brands across retail, travel, and financial services. We asked him to explain where segmentation fits in the analytics stack, and why it matters more than most businesses realize.

Segmentation is applicable to everyone. It doesn't matter whether you're B2C or B2B, what sector you're in, or how sophisticated your analytics setup currently is. At its most basic, it's about grouping consumers who share similar demographics, lifestyles, behaviors, interests, and purchasing habits together, with a complete understanding of their value to the business. The goal is differentiated messaging: understanding where value is being created, and then using that understanding to move customers from one value-based segment to another over time.What you end up with is a segment that has a label, a profile, and a description built from the underlying data. For a B2C business, you'd typically have 20 to 30 of these, each with a name and a profile. And then that segment label appears in every other model the organization should build. That's the part people often miss. Segmentation isn't a standalone exercise. It's a component that shows up in your media mix model, your attribution model, your pricing model, because you want to understand the relative behavior of a given segment at any point.Value is an intrinsic part of the segments we build. A segment or a lifetime value model is a pretty foundational layer for analytics for any organization.

On the technical side, the actual statistical modeling hasn't changed that much. We use either K-means clustering, which is the more traditional technique, or a hierarchical clustering technique, both in Python. What's changed is the machine learning capability you put on top of those segments. It's a bit like AI more broadly: a lot of the equations have been around for a long time. What's changed is compute horsepower, the availability of data, and the ability to put large language models on top of analytics, which is also something we're doing.

If we build a price elasticity model, for example, different segments and different consumers within a segment respond to price-led messages differently. It's about optimizing the relationship between the two things because they don't operate independently. As the analytics structure becomes more complex, you want to start looking at the interrelationship between different models. That's where the big shift has happened over the last year or two: having the ability to apply machine learning to the whole topic.

A large language model like KAI will understand an individual model and its characteristics. So if it's a segmentation model, you don't need to individually query each segment and its profile. You can ask KAI to tell you about a given segment, what its characteristics are, what sort of messaging it responds to or doesn't respond to, and through which channels. And then as things become more complex, KAI starts looking at the interrelationship between different models simultaneously.

For a CEO, the question they're actually asking is: what is really driving my business? Is it price? Is it media? That's what the orchestration layer and KAI are there to answer, in plain language, without needing to reconcile six separate model outputs manually.

Our ETL history means we see a lot more data than most organizations for a given business. If we haven't done any modeling work for a client yet, we will already understand their data. If we're doing inventory management or demand forecasting and we already have access to their warehouse, we probably built their third-party connectors. We know a huge amount about their data, their business, and their customer journey. That puts us in a very strong position to build not just sophisticated models, but models that really drive sales and revenue.

That's what we're here for. The point of intersection between transactional data, sophisticated analytics, and natural language is pretty unique. It's designed to put businesses in a position to control and optimize the levers of profitability.