How we measure retention
Most retention reporting measures the channel, not the customer. We measure the customer: what a cohort does over time, and what changed after we changed something.
The primary frame: cohorts, not campaigns
A cohort is every customer who made a first purchase in the same period. We track what that group does at 30, 60, 90, 180 and 365 days. Campaign-level revenue is an input, never the headline.
The metrics we hold ourselves to
| Metric | Definition we use | Why it is here |
|---|---|---|
| Repeat purchase rate | Share of a cohort with two or more orders by day N | The cleanest signal that retention work is landing |
| Cohort revenue curve | Cumulative revenue per customer by day N | Shows compounding, not one-off spikes |
| Time to second order | Median days between first and second order | Moves earlier than LTV does |
| Contribution per lifecycle moment | Revenue attributable to a specific moment, measured against a holdout where one is possible | Keeps the loop honest about which moment did the work |
| Program economics | Cost of rewards and margin impact against incremental behaviour | Loyalty has to pay for itself |
Attribution: what we will and will not say
- Where a holdout is possible, we run one and report against it.
- Where a holdout is not possible, we say so and report the change as directional.
- We do not report platform-attributed revenue as incremental revenue.
- We do not report a lift we cannot reproduce in the client's own data.
Reporting cadence
Weekly: experiment read-outs. Monthly: cohort trajectory and business review. Quarterly: a meta-review of which categories of change actually moved the curve.
Benchmarks
We do not publish retention benchmarks we have not collected under a disclosed method.
SOURCE REQUIRED · Any benchmark figures on this page require a documented dataset and methodology before publication. See the benchmarks methodology page.
Request a measurement review
We build and run lifecycle, loyalty and customer-data systems inside your existing stack. Tool-agnostic. Outcome-owned.
Request a call