Measure CRM campaign ROI from incremental contribution after costs, not all revenue recorded after a message. Randomly assign eligible players to treatment and control, compare their average contribution over the same window, then scale the difference to the treatment group. Deduct campaign costs once.

This guide gives an operator-ready calculation, a constructed worked example, and checks for uneven group sizes, outliers, overlapping campaigns, and theoretical margin. For the wider reporting framework, use the CRM KPI definitions and formulas.

How do you calculate CRM campaign ROI without double-counting costs?

Choose and name your financial basis before calculating a return. GGR, NGR, and profit are different measures. A label such as "campaign revenue" is too ambiguous for a budget decision.

For this guide, contribution before campaign costs means revenue after agreed variable operating costs, but before the promotional and execution costs being evaluated. Finance should define the deductions, including how bonus-funded play is treated.

Incremental contribution before campaign costs = (mean contribution per assigned treatment player − mean contribution per assigned control player) × treatment player count.

Incremental profit = incremental contribution before campaign costs − incremental campaign cost.

Campaign ROI (%) = incremental profit / incremental campaign cost × 100.

Include promotional cost, channel delivery, and any incremental creative, support, or execution cost in that cost line. Where control receives a different intervention, use treatment costs above the control cost baseline scaled to the treatment group; costs common to both groups cancel. The promotional amount should be the finance-recognised economic cost, not automatically the face value issued.

If your outcome metric is already net of those same costs, compare that net outcome directly and do not subtract them again. Show the cost denominator separately. If campaign cost is zero, ROI is undefined; report incremental profit instead.

Separate two decisions:

  • Repeat this campaign? Compare its incremental contribution with the additional costs of running it again.
  • Fund the CRM function? Reconcile incremental contribution with platform, staffing, agency, and other programme costs over the reporting period. A campaign can cover its marginal costs without covering the whole function.

What is incremental revenue, and how is it different from attribution?

Attribution assigns credit to a touchpoint: for example, a deposit made after an email click. Incrementality asks what changed because the campaign ran. A player who would have deposited without the email can generate attributed revenue without generating incremental revenue.

A concurrent randomised holdout provides a baseline for that counterfactual. In completely randomized designs, the levels of the primary factor are randomly assigned to the experimental units (NIST/SEMATECH e-Handbook).

Set the comparison before sending

  1. Freeze eligibility. Define the player population, market, product, lifecycle stage, and applicable promotional exclusions.
  2. Randomise eligible players. Record assignment before exposure; consider stratifying by prior value or product to improve balance.
  3. Define the contrast. Does control receive business-as-usual messaging, or no discretionary promotion? Both can answer useful but different questions. Keep necessary service and compliance communications running.
  4. Fix the outcome and window. Use the same settlement rules and observation period for both groups. State how late-settled bets, reversals, and bonus liabilities are handled.
  5. Analyse everyone assigned. Include zero-activity players and failed deliveries in their original groups. Comparing only openers, redeemers, or depositors selects on behaviour affected by the campaign.

Set the holdout size from the variability of your chosen outcome, the smallest effect worth detecting, and the available audience. A fixed percentage or a universal minimum player count cannot establish reliability.

How are units assigned in a completely randomized experimental design?

In the US NIST/SEMATECH e-Handbook of Statistical Methods, checked 30 October 2026, the levels of the primary factor in completely randomized designs are randomly assigned to the experimental units. In a CRM test, the units are players.

How do you compare unequal groups and interpret the result?

Constructed example, not an EngageHut client result: 1,000 eligible players are randomly assigned to a seven-day experiment. There are 900 in treatment and 100 in control. These counts illustrate the arithmetic; they do not establish an adequate sample size.

All amounts below use the same contribution before campaign costs definition. The campaign's promotional and execution costs are excluded from those amounts and deducted separately.

CalculationTreatmentControl
Assigned players900100
Total contribution before campaign costs€45,000€3,500
Mean per assigned player€50€35

The estimated lift is €50 − €35 = €15 per assigned player. Applied to the 900 treatment players:

  • Estimated incremental contribution before campaign costs: €15 × 900 = €13,500.
  • Incremental campaign cost: €5,000, including promotional and execution costs on the stated basis.
  • Estimated incremental profit: €13,500 − €5,000 = €8,500.
  • Estimated ROI: €8,500 / €5,000 × 100 = 170%.

This is €1.70 of estimated incremental profit per €1 of campaign cost, or €2.70 of contribution before campaign cost. It is a point estimate, not a proven result without an uncertainty assessment.

Using all €45,000 as the return would produce 800% on the same cost base. That credits the campaign with the baseline contribution as well as the estimated lift. Subtracting the two groups' raw totals would also be wrong because their sizes differ.

Delivered impact versus a rollout forecast

Multiplying the same €15 difference by all 1,000 eligible players gives €15,000 of projected contribution if everyone were treated. The additional €1,500 was not delivered to the holdout. Label this a rollout forecast and forecast the additional cost too; do not insert it into realised campaign ROI.

Can you add campaigns together to report CRM contribution?

Only when the measurement design supports it. The same player can receive several campaigns, each affecting outcomes attributed to another. Adding isolated campaign lifts can double-count that value.

A persistent programme holdout can estimate the combined effect of the defined CRM programme against its stated baseline. Report its scope, exclusions, uncertainty, and costs. Do not estimate unmeasured campaigns by dividing measured lift by the percentage of campaigns tested: campaign counts do not tell you whether their audiences, costs, or effects are comparable.

The CRM features and integration checklist helps specify the assignment, exclusion, and data export capabilities this reporting needs.

How should you handle outliers and theoretical margin?

A large win or loss can dominate a gaming campaign's financial result. It may be a genuine outcome, not a data error. Removing it after seeing which group benefits changes the question and can manufacture a favourable answer.

Keep the raw result and predefine sensitivity checks

Start with reconciled raw contribution, player counts, group balance, and an uncertainty interval appropriate to the experiment. Investigate duplicate events, incorrect currency conversion, unsettled bets, and missing costs before interpreting extremes.

Use additional views to understand fragility:

  • Mean and distribution: show how concentrated contribution is. A median is a useful diagnostic but does not replace the mean for estimating total financial impact.
  • Predefined capped analysis: if winsorisation is appropriate, set common rules before reading treatment results, apply them symmetrically, and show the uncapped result too. The capped estimate answers a different question.
  • Player concentration check: disclose when a small number of players drive the estimate. A result that changes sign when one player is inspected needs uncertainty analysis, not selective deletion.
  • Repeat or extend appropriately: use a planned duration and stopping rule. Repeatedly checking until a positive result appears is not a valid decision rule.

A wide interval can leave a campaign inconclusive even when the point estimate looks profitable. Report that plainly. Small VIP audiences may require a longer planned test or a different design; a bigger percentage holdout alone does not guarantee enough information.

Add theoretical contribution as a separate view

For a casino game with a known theoretical edge:

Theoretical gross revenue = sum of each game's eligible turnover × that game's theoretical edge.

Use the actual game configuration and mix, with consistent treatment of cash and bonus play. A single generic margin can hide a shift into different games.

The UK Gambling Commission's RTP guidance distinguishes actual from theoretical RTP and explains why volatility and volume affect their difference. Theoretical revenue is an expectation; it is not cash profit or a guarantee for the next campaign.

Apply the same per-assigned-player treatment-minus-control calculation to theoretical revenue, then deduct the agreed variable costs and incremental campaign costs once. Constructed illustration: €100,000 of estimated incremental turnover at a specified 4% theoretical edge gives €4,000 of theoretical gross revenue, before those deductions.

For sportsbook, do not substitute quoted odds overround for realised profit. Settlement, event outcomes, bet mix, and the operator's expected-margin model need separate treatment. Ask trading and finance to approve the model.

How does the UK Gambling Commission say actual RTP is assessed?

The UK Gambling Commission's RTP guidance, checked 30 October 2026, says actual RTP is measured against the game's calculated figures, and the volatility of the game must also be considered.

Does the GB RTP tolerance narrow as play volume rises?

Yes. The UK Gambling Commission's RTP guidance, checked 30 October 2026, says "as the volume of play increases the tolerance will decrease".

What should you measure alongside campaign ROI?

A useful decision needs more than a single return percentage:

MeasureDecision it supports
Incremental profit and uncertaintyWhether the campaign adds value on the defined cost basis
Cost and budget at riskWhether the operation can afford the exposure
Retention after the offerWhether activity persists beyond the initial response
Cannibalisation and timingWhether the campaign moves existing activity across products or pulls it forward
Eligibility and suppression failuresWhether the campaign can be operated acceptably
Complaints, opt-outs, and support burdenWhether commercial results conceal a poor player experience

Use the same follow-up horizon when comparing campaigns. Higher short-term ROI does not automatically outweigh larger long-term profit, and more activity is not by itself a reason to increase promotional pressure.

Break down results using predefined player segments. Treat unexpected subgroup winners as hypotheses for another test, especially when many groups were inspected.

What should a campaign measurement brief contain?

Before launch, record:

  1. The decision: repeat, stop, change, or scale which intervention?
  2. Eligibility, exclusions, assignment unit, and treatment/control counts.
  3. Treatment and control experience, including other campaigns they can receive.
  4. Primary outcome, financial deductions, observation window, and settlement cutoff.
  5. Cost ledger and whether each cost is already inside the outcome metric.
  6. Sample-size rationale, planned stopping rule, and uncertainty method.
  7. Operational and player-protection guardrails.
  8. A named owner and the action to take if the result is positive, negative, or inconclusive.

After the test, add delivery failures, reconciliation gaps, raw and sensitivity results, and the final decision. This creates a repeatable record rather than a new calculation for every report.

If assignments and financial outcomes cannot be joined reliably, start with a player data and reporting assessment. If the calculation is sound but campaign decisions remain unclear, a CRM Growth Audit can examine measurement alongside targeting and lifecycle design.

Key numbers

Definitions quoted from primary sources, each checked on 30 October 2026.

FactSourceChecked
US (NIST/SEMATECH): "For completely randomized designs, the levels of the primary factor are randomly assigned to the experimental units."NIST/SEMATECH e-Handbook: completely randomized designs30 October 2026
GB (UKGC): actual RTP is measured against the game's calculated figures, and the volatility of the game must also be considered.UK Gambling Commission: how to calculate RTP30 October 2026
GB (UKGC): "as the volume of play increases the tolerance will decrease".UK Gambling Commission: how to calculate RTP30 October 2026

See the CRM Growth Audit

Frequently asked questions about CRM campaign ROI

What is a good CRM campaign ROI?

There is no universal threshold. Compare incremental profit, uncertainty, cost basis, payback horizon, and the opportunity cost of using that audience and budget elsewhere. A bonus-only return and a fully costed programme return are not comparable.

How large should a control group be?

Size it for the outcome's historical variability and the smallest commercially useful lift. Revenue and contribution can be much noisier than response rates. Do not copy the illustrative 90/10 split above as a sample-size recommendation.

Can you measure incrementality without a randomised holdout?

A well-designed quasi-experiment may estimate causal impact when randomisation is unavailable, but it requires defensible assumptions and specialist analysis. Before/after revenue or click attribution alone cannot separate a campaign from seasonality, acquisition changes, or other interventions. Label observational reporting accordingly.

Is incremental return on spend the same as ROI?

No. Incremental contribution divided by cost is a return multiple before subtracting that cost. ROI divides incremental profit after the relevant costs by the cost base. Name the numerator and denominator so the label can be checked.

Should actual or theoretical revenue decide the result?

Keep actual financial outcomes as the accounting record. Use theoretical contribution as a clearly labelled diagnostic for game mix and volatility, with its assumptions visible. Neither view removes the need for a valid experiment or player-protection guardrails.

How to cite this page: EngageHut, "How to measure CRM campaign ROI: the complete guide to incremental revenue, control groups, and outlier management", engagehut.com, https://engagehut.com/insights/crm-campaign-roi-measurement-guide/, updated 30 October 2026.