Thinklytics

Specialty Apparel Retailer · Retail & E-Commerce · Nashville, TN · 12 weeks

Churn prediction model identified $3

A specialty apparel retailer with 1.8 million loyalty members couldn’t identify which customers were likely to churn until after they stopped buying. We developed a churn prediction model that flagged at-risk members 90 days in advance. This allowed the retailer to run targeted win-back campaigns that recovered $3.1 million in annual loyalty revenue.

Challenge

The retailer considered a member churned only after 12 months of inactivity. This delay meant they missed any chance to act early. There was no system to identify at-risk members, no way to intervene, and no clear tracking of the loyalty program’s actual ROI beyond just counting members.

Approach

We developed a churn model using purchase recency, frequency, monetary value, and engagement data from email, app, and in-store activity. Each week, we scored all 1.8 million members and identified those likely to churn within 90 days. We integrated these scores into the marketing platform to trigger targeted retention campaigns automatically.

Outcome

The model flagged 94,000 high-risk members in its first run. We launched targeted win-back campaigns and won back 68 of every 100 members who received outreach, securing $3.1M in retained annual loyalty revenue. This project also set up the first reliable way to measure loyalty program ROI.

Spotting Issues Early So We Don’t Get Hit With Big Delays

We scrapped the old 12-month inactivity check and jumped onto a 90-day churn prediction model instead. Every week, we dug into customer behaviors that shouted “at risk” and scored members accordingly. This gave the retailer fast heads-ups so they could jump in and stop churn before it even started.

How We Got Automation Up and Actually Saw What Made a Difference

Here’s what we did: we fed the churn scores directly into the marketing system. So the moment someone looked like they were about to bail, we’d fire off a win-back campaign immediately. We also kept a control group for each campaign. That way, we could track exactly how many folks we brought back and establish the real ROI of the loyalty program.

Results

  • $3.1M Annual loyalty revenue recovered
  • 68 of 100 High-risk member recovery rate
  • 94,000 At-risk members identified in first scoring run
  • 90 days Early warning window before predicted churn

We used to wait a full year of no activity before counting someone as churned, which meant we were always behind. Thinklytics built a model that flags customers likely to leave 90 days early. That helped us save $3.1 million in the first year.

Director of Loyalty and CRM, Specialty Apparel Retailer

In the client's terms

The question

Is the data in a state that will survive what we want to build on it?

The situation

A member counted as churned after 12 months of inactivity, by which point there was nothing to act on. There was no at-risk list, no intervention, and no measure of loyalty programme return beyond membership count.

What we found

The definition of churn was the problem. Twelve months of silence is not an early warning, it is a post mortem, and the programme had been measuring its own size rather than its effect.

What it produced

The first run flagged 94,000 high-risk members. Win-back outreach recovered 68 of every 100 contacted, holding $3.1M in annual loyalty revenue. The programme also acquired its first reliable measure of return.

The roadmap

A churn model on purchase recency, frequency and value plus engagement across email, app and store, scoring all 1.8 million members weekly and identifying those likely to lapse within 90 days. Scores were written into the marketing platform so retention campaigns trigger automatically rather than waiting on a monthly review.

The work behind this result

Related case studies

Frequently asked questions

When should a loyalty member count as churning?

Early enough to act. This retailer used 12 months of inactivity, which is a post mortem rather than a warning. Weekly scoring against behaviour that precedes lapse gives a 90-day window instead.

Does win-back outreach actually recover members?

At this specialty apparel retailer, 68 of every 100 high-risk members contacted were recovered, holding $3.1M in annual loyalty revenue. The first scoring run flagged 94,000 members across a base of 1.8 million.

How do you measure loyalty programme return?

With a control group on every campaign. That is what gave this retailer its first defensible measure of programme return, having previously measured only membership count.