Most loyalty programme ROI calculations are measuring the wrong thing. The standard formula, member revenue minus non-member revenue, tells you that your members spend more than non-members. It does not tell you whether your programme caused that difference. Members self-select into loyalty programmes precisely because they already buy more frequently. If you are not controlling for that pre-enrolment behaviour, you are measuring correlation and calling it causation, and every number on your dashboard is inflated as a result.
- Key takeaway: The standard loyalty ROI formula has a selection bias problem that most operators never account for.
- Key takeaway: Incrementality requires a control group, not just a member-vs-non-member comparison.
- Key takeaway: Short-cycle and long-cycle incrementality are different questions and need different measurement approaches.
- Key takeaway: A number sitting in a dashboard changes nothing. Someone has to act on it on a regular cadence.
- Key takeaway: Cohort LTV trajectory over 12 to 24 months is a more honest loyalty metric than month-one revenue lift.
Why Most Loyalty ROI Calculations Are Lying to You#
The formula most guides publish is: Incremental Revenue = Member Revenue minus Baseline (Non-Member) Revenue. Voyado and LoyaltyLion both explain it clearly. The problem is not the arithmetic. The problem is what you are comparing. A customer who signs up for your loyalty card on their third visit in a fortnight is not a representative baseline. They were already a high-frequency buyer before the card existed. Comparing their post-enrolment spend to a non-member who visited once six months ago and never came back produces a flattering number that has almost nothing to do with your programme.
This is the selection bias problem. It is well-documented in causal inference literature, and practitioners on Reddit's CausalInference community flag it regularly when operators ask how to measure loyalty incrementality. The fix is not a better formula. It is a better control group.
The Three Methods That Actually Isolate Incremental Revenue#
There are three approaches that produce numbers worth trusting. Each has a different data requirement and a different level of operational lift.
| Method | What it does | Data needed | Practical difficulty |
|---|---|---|---|
| Pre/post enrolment comparison | Compares each member's spend in the 90 days before joining versus 90 days after | Transaction history with timestamps and member join dates | Medium: requires clean historical data per customer |
| Matched control group | Pairs each member with a non-member of similar pre-enrolment visit frequency and spend level | Enough non-member transaction data to find statistical matches | High without a data team; the gold standard when done correctly |
| Holdout group test | Withholds loyalty enrolment from a randomly selected group and compares outcomes over time | Ability to randomise enrolment, which most operators cannot do retrospectively | Low ongoing, but requires planning before launch |
The pre/post enrolment comparison is the most accessible for mid-market operators. If your platform records when each customer joined and you have transaction data going back at least 90 days, you can run this without a data science team. Talon.one's restaurant-specific guidance mentions matched comparables as the gold standard, but for most independent hospitality groups, the pre/post method with a 90-day window on each side is a credible and achievable starting point.
Two Different Questions: Campaign Lift vs. Cohort LTV#
Incrementality is not one question. It is two, and conflating them is how operators end up with misleading reports. Short-cycle incrementality asks: did this specific campaign drive a visit that would not have happened otherwise? Long-cycle incrementality asks: is this cohort of members compounding lifetime value over 12 to 24 months in a way that non-members are not? The first is a campaign metric. The second is a business metric. Most loyalty reporting only attempts the first, which is why loyalty programmes often look good in monthly reviews and still get cut when the board asks for proof of value.
What Good Looks Like: Repeat Rate, AOV, and LTV#
Before you can prove incrementality, you need to know which metrics to track. Epsilon's guidance on loyalty measurement and Propello's list of essential metrics both converge on the same core set: repeat purchase rate, average order value among members versus non-members, and customer lifetime value by cohort. Of these, repeat rate is a more honest loyalty metric than points redeemed because it measures behaviour, not programme mechanics. A customer redeeming points is using the system. A customer returning without a redemption trigger is demonstrating genuine preference.
Rivo's analysis of loyalty programme statistics puts the repeat purchase rate for loyalty members significantly above non-members across ecommerce categories, but the gap narrows considerably when you control for pre-enrolment behaviour. That narrowing is the honest number. It is smaller than the headline figure, and it is the one worth reporting to a finance director.
From Vanity Metric to Board Metric: The Cadence Problem#
Here is the part no measurement guide addresses: a number sitting in a dashboard changes nothing. EY's framework for demonstrating loyalty ROI is thorough on methodology and silent on execution cadence. The operators who get value from loyalty measurement are the ones running it as an ongoing discipline, not a quarterly exercise. That means a weekly operational check on active member count and recent redemption rate, a monthly business review comparing cohort repeat rates, and a quarterly strategic review of LTV trajectory by enrolment period. Without that cadence, the measurement is an audit, not a management tool.
This is the specific pain point that comes up repeatedly among hospitality ops directors: the programme exists, the data exists, but no one has the time or mandate to run the analysis on a regular basis. The loyalty software was bought. The measurement framework was never operationalised. The result is a dashboard that gets opened before board meetings and ignored the rest of the time. If keeping members active enough to generate measurable lift is already a stretch, running rigorous incrementality analysis on top of it is unrealistic for a small marketing team.
The Incrementality Test: A Practical Checklist for Operators#
- Do you have transaction-level data with timestamps and member join dates going back at least 90 days? If not, start collecting it now. You cannot run a pre/post comparison without it.
- Are you comparing members to non-members, or members to their own pre-enrolment baseline? The second comparison controls for selection bias. The first does not.
- Have you segmented your member base by enrolment cohort? Members who joined 18 months ago should look different from members who joined last month. If they do not, the programme is not compounding.
- Is repeat rate tracked separately from redemption rate? A member who visits without redeeming is more valuable as a signal than one who only visits to use a reward.
- Is someone reviewing these numbers on a fixed cadence, with a mandate to act on what they find? If the answer is no, the measurement exercise is producing information, not outcomes.
- Can you articulate, in one sentence, what would have to be true for you to conclude the programme is not working? If you cannot define failure, you cannot define success either.
The Question Every Measurement Guide Ignores: Who Acts on It?#
The measurement guides, including the detailed ones from Fundle and EY, all stop at the number. They do not address what happens next. In practice, the incrementality analysis is only useful if someone with operational authority sees it, interprets it correctly, and has the time and tools to respond. For a founder running a three-site hospitality group while managing suppliers, staff rotas and a refurbishment, that is a significant ask. For an ops director whose marketing team is two people, it is equally unrealistic. The measurement problem and the operational problem are the same problem.
That is the case for treating loyalty as a managed outcome rather than a software subscription. How Carrott runs loyalty programmes for operators is built around this gap: the analysis, the cadence, and the operational response are part of the service, not homework the operator has to do after purchasing a platform. The incrementality story is only worth telling if someone is there to tell it on a regular basis, and to act when the numbers move.
What is the difference between loyalty revenue and incremental loyalty revenue?
Loyalty revenue is the total spend attributed to programme members. Incremental loyalty revenue is only the portion of that spend that would not have occurred without the programme. The gap between the two is usually significant, because members tend to be higher-frequency buyers before they ever join. Measuring incrementality requires controlling for pre-enrolment behaviour, not just comparing members to non-members.
How do I build a control group for loyalty measurement without a data science team?
The most accessible approach for mid-market operators is a pre/post enrolment comparison: compare each member's spend and visit frequency in the 90 days before joining against the 90 days after. This is not as statistically rigorous as a matched control group, but it controls for the main selection bias problem and can be run in a spreadsheet if you have clean transaction data with timestamps and join dates.
Is repeat rate a better metric than redemption rate for proving loyalty ROI?
For incrementality purposes, yes. Redemption rate tells you that members are using the programme mechanics. Repeat rate tells you that members are returning, which is the behaviour the programme is supposed to drive. A member who visits without redeeming a reward is a stronger signal of genuine loyalty than one who only returns when a reward is available.
How long does it take to see meaningful incrementality data from a loyalty programme?
For short-cycle campaign incrementality, four to eight weeks of post-campaign data is usually sufficient. For long-cycle cohort LTV incrementality, the signal becomes meaningful at around six months and is most reliable at 12 to 24 months. Programmes that are only evaluated on month-one metrics will almost always look better than they actually are.
My loyalty programme has been running for a year but I have never measured incrementality. Where do I start?
Start with the data you have. Pull transaction records for all members and identify their join dates. Calculate average monthly spend and visit frequency for the 90 days before and after enrolment for each member. Segment by enrolment cohort. If spend and frequency increased post-enrolment and that increase has held or grown over time, you have a credible incrementality case. If the increase faded after the first redemption, the programme is driving trial, not retention.