Retention and Churn
Date: 2026-08-16
The same thing counted from opposite ends. Which one you report changes the conversation — retention frames a base to build on, churn frames a leak to plug — and for non-subscription retail neither is defined until you decide what “gone” means.
What it is
Retention is the proportion of customers still active in a period. Churn is the proportion who stopped.
Mathematically trivial, rhetorically not. “70% retention” and “30% churn” are the same fact and produce different meetings.
The definition problem in retail
A subscription has an unambiguous churn event — a cancellation. Retail doesn’t. A customer who hasn’t bought for four months might be gone, or might be mid-cycle.
So churn has to be defined, and the definition is a business decision:
customer last ordered 120 days ago
consumables, 45-day cycle → almost certainly churned
apparel, 6-month cycle → entirely normal
furniture, 4-year cycle → early days
Set the threshold from your own Time Between Orders distribution, not from a round number. A common rule: churned once they’ve passed roughly twice the median repurchase interval without ordering — long enough that most genuine repeaters would have returned.
Whatever you choose, write it into the metric definition, because a monthly retention figure with an undefined churn window means nothing — Metric Design.
Why it dominates the economics
On the running model — £22.50 CAC, £15 contribution per order:
orders per customer contribution LTV LTV:CAC
1.6 £24.00 1.07
1.9 £28.50 1.27
Moving from 1.6 to 1.9 orders per customer — roughly a ten-point improvement in repeat rate — lifts LTV:CAC by nearly 20%. That’s a larger effect than most acquisition optimisation achieves, on spend you’re already making.
That asymmetry is the argument for lifecycle work: acquisition improvements compete against an efficient market, retention improvements compete against nobody.
Retention is not one number
The most common analytical error is a single blended figure:
- It varies enormously by acquisition channel. Discount-led acquisition retains worse, so a shift in Channel Mix moves blended retention with no behavioural change
- It varies by cohort. Mixing cohorts of different ages means the aggregate moves with acquisition volume rather than behaviour — a big new-customer month lowers blended retention — Cohort Analysis
- It varies by first product. The category someone enters through predicts whether they return
Report it as a curve by cohort, not as a percentage.
Revenue retention differs from customer retention
Two different questions, and both matter:
customer retention are they still buying?
revenue retention are they spending as much?
Fewer customers spending more is a different business from more customers spending less, and a customer-count figure alone can’t tell them apart — Cohort Revenue.
Where it’s measured wrongly
- Only identified customers have retention. Anonymous traffic can’t be tracked, so retention describes your best customers — Anonymous and Identified Users
- Identity decay reads as churn. A returning customer with an expired identifier looks like a new one, which understates retention and overstates acquisition simultaneously — Browser Privacy Restrictions
- Survivorship Bias — measuring only customers still present inflates it
- Returns not deducted, so a customer who bought and sent it back counts as retained
Related: Repeat Purchase Rate for the single most diagnostic version of this, and Lifecycle Stages for acting on it.