A client showed us a dashboard last quarter with 11,400 enrolled members, a 38% redemption rate, and £214,000 in tagged member spend. The owner thought the programme was working. It wasn't. Once we stripped out the customers who would have come back anyway, the genuinely incremental revenue was closer to £41,000. Same programme, same numbers, completely different story.
Incremental loyalty revenue is the revenue that exists because of the programme, not the revenue that happens to flow through it. To measure it honestly you need a credible baseline (what these customers would have spent anyway), a way to correct for self-selection bias (your best customers join loyalty schemes regardless), and a monthly cadence so the number stays accountable rather than becoming a once-a-year board slide.
- Member revenue is not incremental revenue. Your top 20% of customers join programmes anyway.
- Enrolled members, redemption rate and total member spend are the three most misleading metrics in loyalty.
- A credible baseline does not need a data science team. It needs a matched non-member cohort and one quarter of pre-launch data.
- Self-selection bias inflates almost every member-vs-non-member comparison you will see in a vendor case study.
- Measurement is an ongoing programme function, not a calculation. Someone has to own the monthly review.
Three vanity metrics we refuse to report on their own#
These metrics aren't useless. They're useful as diagnostics. They become dangerous the moment they get presented as proof the programme is working.
| Metric | Why it looks like proof | Why it isn't |
|---|---|---|
| Enrolled members | Big number, grows every month, easy to put on a slide | Enrolment is a marketing cost, not a return. Many enrolled members never transact a second time. |
| Redemption rate | Suggests members are engaged and rewards are attractive | A high redemption rate often means you are discounting customers who would have bought anyway. That is margin erosion, not lift. |
| Total member spend | Largest revenue figure available, feels like programme contribution | Includes the spend your best customers would have made without the programme. Without a baseline, this number tells you nothing about incrementality. |
The control group problem nobody wants to talk about#
Every loyalty guide tells you to compare members against non-members. The problem, as causal inference practitioners point out, is self-selection: customers who opt in are systematically different from customers who don't. They visit more, spend more, and would have done so without the card in their wallet. Comparing the two groups directly is like comparing gym members to non-members and concluding the gym caused their fitness.
What we do instead is build a matched cohort. We take members and pair each one with a non-member who had similar pre-enrolment behaviour: visit frequency, average spend, recency. Then we measure the difference in behaviour after enrolment. It is not a randomised trial, and we say so plainly. But it removes the most obvious source of inflation and gives an operator a defensible number.
"If your loyalty report cannot answer the question 'compared to what?', it is not a loyalty report. It is a member newsletter with revenue attached."
The numbers we actually report each month#
- Incremental repeat rate: the gap between repeat visit rate of matched members and matched non-members, tracked monthly. This is the metric we watch most closely because it is the cleanest signal of behavioural change.
- Incremental AOV: average transaction value of members on programme-influenced visits, minus the matched non-member average. Useful for spotting whether reward structures are encouraging larger baskets or just discounting existing ones.
- Incremental revenue per active member: total incremental contribution divided by active members in the period. This is the number that goes on the operator's P&L conversation, not the enrolment count.
We pair these with two diagnostic figures (redemption rate and enrolment growth) but never lead with them. For a fuller view of what a healthy loyalty ROI looks like in practice, the ratio we look for is incremental revenue covering programme cost by month four and running at 3 to 5x cost by month twelve.
What we do when the numbers look wrong#
Measurement is only useful if it changes decisions. Three patterns come up repeatedly:
| Pattern | Likely cause | What we change |
|---|---|---|
| Flat incremental repeat rate after month 3 | Reward threshold too far away, or reward not valuable enough relative to AOV | Shorten the stamp ladder or restructure points tiers; test reward perceived value against a small cohort |
| Incremental AOV negative | Programme is cannibalising full-price visits with discounts | Shift from percentage rewards to fixed-value or product rewards; review redemption rules |
| Incremental revenue suspiciously high in month 1 | Pre-launch announcement pulled forward demand, or matched cohort poorly constructed | Rebuild the cohort with a longer pre-period; treat month 1 as anomalous and re-baseline at month 3 |
This is the loop most operators don't have time to run. It is also the loop that turns a loyalty card from a cost line into a compounding revenue stream. It is the work behind how Carrott runs the programme on your behalf, and the reporting cadence we build into every programme is designed around it.
What 'good' looks like at 6, 12 and 24 months#
| Timeframe | Incremental repeat rate lift | Programme cost coverage |
|---|---|---|
| 6 months | +4 to +8 percentage points vs matched non-members | 1x to 2x programme cost |
| 12 months | +8 to +14 percentage points | 3x to 5x programme cost |
| 24 months | +10 to +18 percentage points, with stable churn | 5x to 8x programme cost, declining acquisition share |
These are ranges from programmes we run in UK hospitality. They are not universal benchmarks, and any vendor quoting precise figures without context is selling, not measuring. Voyado's ROI guide and Brierley's framework both make the same point: the numbers only mean something inside your category and your customer base.
Can we measure incremental revenue without a control group?
Not credibly. You can use pre/post comparisons on the same customers, but those are heavily contaminated by seasonality and marketing activity. A matched non-member cohort is the minimum bar for a defensible number.
How much pre-launch data do we need?
One quarter at minimum, two is better. You need enough history to build matched cohorts and to distinguish programme effects from normal seasonal variation.
Is redemption rate ever a useful metric?
Yes, as a diagnostic. A very low redemption rate suggests rewards aren't attractive or aren't visible. A very high rate suggests you're discounting customers who would have bought anyway. Neither tells you anything about incremental revenue on its own.
How often should we review these numbers?
Monthly for operational metrics (repeat rate, AOV, active members), quarterly for the full incremental revenue calculation against the matched cohort. Annual-only reporting is too slow to catch a drifting programme.
What if our loyalty software doesn't export the data we need?
Most don't, cleanly. You either need a platform with a usable API for member-level transaction data, or you accept that your reporting will be partial. This is one of the reasons we built measurement into the managed service rather than expecting operators to wire it up themselves.