Customer Selection
A fashion retailer wanted to know who would be making purchases over the next two weeks. We compared five selection methods using exactly the same data. The difference between the roughest and the most precise approaches: a threefold difference in revenue per customer.
Objective
The retailer wanted to answer one specific question: Who will be making purchases over the next fourteen days? Not “Who were our biggest customers in the past?” but who is currently the most valuable for this promotion. The task for our data analyst: to demonstrate, using actual figures rather than assumptions, how much a more granular or predictive selection yields compared to a simple approach.
Challenge
Who your best customer is depends entirely on your goal. One retailer may want to engage as many customers as possible, while another may target the biggest spenders. “Best” is therefore not a characteristic of a customer, but rather a strategic choice.
In practice, however, selection is often still based on an arbitrary hard threshold—such as a revenue threshold—without taking into account the full customer profile or the actual objective. This means you’re leaving value on the table: you’re directing your campaign toward those who happen to fall above a certain threshold, rather than toward those who will yield the greatest return.
The Stratics Approach
Instead of setting a limit, we focused on one specific question: Who will make a purchase in the next fourteen days? We then compared five selection methods side by side using exactly the same data, from the most basic to the most sophisticated. Each step adds one layer to the customer profile. The most precise layer is a predictive model that estimates the likelihood of a purchase for each customer, powered by the 360° customer profile in MIP.
These five levels are not a one-size-fits-all solution. Our analyst carefully considered the specifics and selected them in consultation with the client: which thresholds, which variables, and which definition of a purchase. There are numerous other possible configurations to test. These were the most relevant for this client, and that only became clear after engaging in a real dialogue with the client.
Approach
Result
Revenue per customer based on a sample of 50,000 customers. The baseline is a random selection from among those who have made a purchase at least once (€5.31). By comparison, a purely random selection from the entire database yields €2.68; that buyer filter alone doubles that figure.
Simulation based on real data from a fashion retailer. Source: Stratics’ own analysis, validated by Yann Cadoret.
"Your best customer isn't separate from your goal. Ask the right question, and the difference is measurable: three times more revenue per customer."
The foundation and the model do not replace each other; they reinforce each other. A hard threshold captures the largest share of profit; RFM and ranking refine that; and a predictive model built on top of a clean 360° customer view captures what the rest miss. The profit is cumulative.
In the Spotlight
The more refined (smaller) the selection, the greater the difference. With a selection of 50,000 customers, the predictive model yields €16.16 per customer, compared to €5.31 if you were to choose customers at random from among your buyers.
Simulation based on real data from a fashion retailer. Source: Stratics’ own analysis, validated by Yann Cadoret.
We'll frame the question clearly and use your own data to show you the benefits of a more granular selection.
Schedule a call