Activating Segmentation: From Insight to Action

Mike Moussallem, President of Explorer Research, joined Stuart Russell, Chief Strategy Officer, and Graham Burton, Chief Technology Officer at Plinc, for a conversation on why well-built segmentations so often stall before they reach the customer, and what it actually takes to get them working commercially.

That stall tends to look the same across organizations. The research is strong. The segments are well defined. The business is excited by the output. And then, six months later, the marketing a customer receives looks almost identical to what it was before the project started. None of this comes down to a lack of insight or a poorly built segmentation. The gap shows up in the handoff: the team that understands the customer often isn’t the same team deciding what actually gets sent to them.

Watch the discussion in full below. Or if you’d rather read than watch, we’ve summarized the key themes underneath.

Watch the Webinar

Two starting points: attitudinal research and first-party data

Segmentation tends to begin from one of two places, and each has a different natural strength. Attitudinal segmentation, the kind Explorer builds, focuses on needs, motivations and psychographic drivers. It does include behavioral elements, but they’re based on what people say they do rather than what they’ve actually done. This type of research works well when it’s feeding into new products, packaging decisions or media planning. However, what it doesn’t give you is a clear way to integrate into the wider business systems and workflows that shape day-to-day customer communication.

First-party data-led segmentation usually starts with transactional data: who’s spending the most, who’s spending less. From there it extends into behavior, which channels a customer engages with, what they browse but don’t buy, and into what happens after purchase, including returns and reviews. The real potential for greater impact comes when the segmentation moves beyond describing what a customer has done and starts predicting what they’re likely to do next. That’s the point where it can be used to change behavior, not just report on it.

Put the two together and the connection runs all the way through to the customer. As Mike put it:

“We’ve got all of the rich data, and then we’ve really got the why behind it with the attitudinal segmentation piece. If we can take those elements from both, now we’ve got a meaningful connection all the way through to that end customer, so we’re communicating with the right offer at the right time, with the right message. And that’s a cycle too. It can feed on itself and continue to refine and iterate over time.”

Where segmentation stalls

The pattern most organizations recognize is a segmentation that does what it was built to do, then sits on a shelf until it’s refreshed a few years later. That’s rarely down to the quality of the work. It’s a business silo problem: CRM and loyalty teams often aren’t in the room when a segmentation is being designed, so the fields they’d need for activation are never captured in the first place. Or they are involved, but the governance to carry a segment through to deployment simply isn’t there.

This was echoed by Stuart, drawing on a client-side anecdote of his own: a data science team had built a genuinely accurate lapse prediction model, and handed it over to the customer marketing team with a list of names.

“These customers are going to lapse, and… I need to know why, and what do I do about it? I don’t know what message to send. I don’t know the motivation behind the lapsing behaviors, so you really feel quite hamstrung. You’ve got the insight, you can set up the trigger and automation, but what do you actually do with it?”

What good looks like: scaled, real-time and never quite finished

Graham set out what the technical side of “good” actually involves. It starts with placing every customer and prospect into a segment, using both transactional and behavioral data alongside machine learning to predict where the people who never answered a survey most likely belong, while keeping a record of how each customer’s segment has moved over time so performance can be measured properly.

From there, it’s about accessibility. That means segment showing up in weekly trading reviews, not just in which categories are up or down but which segments are driving that movement. It means the segment reaching the email platform, the website for content personalization, the loyalty engine for tailored offers, and in some cases, the point of sale itself, so in-store teams can treat customers differently based on the segment they fall into.

Real-time matters more than it used to. The window to influence a customer’s behavior is shrinking, particularly online, where someone can land on a site, compare alternatives and be gone before a next-day email ever reaches them. And segmentation itself needs to keep moving: the segment a customer sits in today may not be the segment they’re in six months from now, whether because marketing has worked, preferences have shifted, or the market itself has changed with the season. Left static, a segmentation’s value degrades over time.

Getting from research to a base-wide segmentation

Two practical approaches came up for mapping a segmentation across a full customer base. The first, retrofitting, starts with an existing survey-led segmentation and a typing tool. A group of customers whose transactional and behavioral data is already known are surveyed and typed into segments, and machine learning models then predict segment membership, or predict likely survey responses, for everyone else, with a confidence score attached so a business can decide how aggressively to activate against each customer. A global fashion retailer used exactly this approach to build what Stuart described as one of the most sophisticated CRM emails he’s seen, personalizing not just product recommendations but tone and copy based on personas that had never previously been mapped to individual customers.

The second approach works in the other direction: using first-party data to shape the research before it’s even run, seeding an initial view of the segmentation from what customers are already buying and browsing, then designing sharper, more specific survey questions around it. One client using this approach also ran the same questions past a wider consumer panel and found a substantial segment they weren’t reaching at all. The research explained why: a gap in their product offering. That finding led to a strategic acquisition to close it, a use case that was never the original intent of the segmentation, but proved one of the most valuable outcomes of the work.

Turning segments into action

Knowing who to talk to and why is only half the job. Segments can inform tone of voice and content, supported by tools like a category affinity model to help decide what to talk to a customer about and a product recommendation engine to get specific. But content alone doesn’t change behavior for every customer. Changing behavior usually requires an incentive, and that’s where predicted value comes in: rather than offering every customer the same discount, future value modeling identifies where there’s genuine headroom to grow, so spend goes toward the customers most likely to respond rather than those who were going to buy anyway.

None of this works without measurement. Approach, target and control testing, with results available quickly enough to stop what isn’t working, is what proves the investment paid off and what feeds learning back into the next round of activation.

Handling more than one segmentation

A question from the audience raised a familiar scenario: a retail segmentation and an online segmentation that don’t quite map onto each other, with one typically winning out because that’s where the transaction happens. Graham’s answer was to treat the combination of the two as the true view of the customer, joining the data at a customer level through techniques like cookie matching, card matching or loyalty data, rather than treating it as a problem to solve by picking one segmentation over the other.

Closing thoughts

Mike closed with a short list worth holding onto: define a clear activation use case up front, even if it isn’t the primary reason for the segmentation. Get the right stakeholders involved from the beginning, ideally with a hybrid view where behavioral data and attitudinal insight inform each other. Build the first-party data foundation and use machine learning to scale it across the base. And hold the whole process to a governance standard that carries through to measurement.

“If anyone has a segmentation that’s sitting on a shelf somewhere, I’m confident there are other ways to bring it to life and make it work for your organization in a meaningful way.”

Thank you to everyone who joined us live. For those who couldn’t make it, we hope this recap gives you a feel for the ground we covered.

If this is a gap you’re navigating in your own organization, closing it doesn’t have to mean a major program or a long roadmap. You can start by testing the single highest-value move, proving it commercially, and deciding what to scale from there. That’s the thinking behind Growth Labs, a fixed-scope pilot from Plinc that turns one segmentation-led hypothesis into tested commercial evidence, using the data you already have.

Take a look at our Segmentation Activation Growth Lab one page summary to see how a pilot works in practice, or visit the Growth Labs page for the wider range of use cases it can help with.