The customers who were quietly leaving

A sudden jump in churn late in the year had no obvious cause. The team could see the rate move, but could not say which customers were most at risk, why they were leaving, or whether the spike would persist. Churn was a single blended number that nobody could act on.

The client is a high-volume on-demand delivery platform with a large, active customer base. Most orders were still paid in cash on delivery, but a growing share of customers were moving to the platform's own digital wallet.

Churn was managed as one undifferentiated rate, with no view of the kinds of customer sitting behind it. So when churn spiked late in the year, there was no way to explain it — or to know whether it was a one-off or the start of a trend.

Using a rolling three-month window, every customer was placed into one of four behavioural segments by the share of their orders paid through the wallet — from wallet-first, through mixed, to entirely cash.

We then compared churn, order frequency and order value across those segments, tracked how the segment mix shifted month by month, and isolated the late-year churn spike against an external disruption to cash availability — the test for whether payment behaviour really explained who was leaving.

Payment behaviour was one of the clearest churn predictors the business had. Cash-only customers also ordered less often and spent slightly less per order. And the risk was concentrated: most of the base was still cash-only, while the stickiest group — wallet-first customers — more than doubled across the year.

When an external shock to cash availability hit late in the year, it was the cash-only segment that churned hardest — jumping several points in a single month while wallet customers held steady. The pattern held across the platform's largest markets and among its most frequent users, which meant it was structural, not a quirk of one place or one period.

The business could finally treat churn as something to act on. Moving customers from cash to wallet wasn't a payments preference — it was a retention lever, and the analysis sized how much retention was on the table.

It gave the team a defensible basis to prioritise wallet adoption among at-risk cash-only customers, and an early-warning view of which segments would be most exposed the next time cash availability tightened. The work didn't reduce churn on its own; it identified where the reduction opportunity was, and how large it could be.

Customer Segmentation is one of five Hira Deep-Dive Sprints — a focused, four-week engagement that finds the behavioural lines that actually predict value and churn, and turns them into something the business can act on. It's the right fit when you know a single average is hiding very different customers.