Tracked affiliate revenue does not show how much new revenue the program caused. It shows the revenue credited to affiliates under your tracking rules. Incremental revenue is the difference between what customers spent with the program and what they would have spent without it. An order report observes the purchase; it cannot observe that same customer's alternative purchase history.
Keep both measures. Attributed revenue helps reconcile orders and commissions. Incrementality helps decide whether more affiliate spending will add enough sales to justify its cost.
What the tracking record establishes
Affiliate systems connect an order to a recorded signal, such as a creator's link or code. For example, Shopify Collabs documents creator-linked visits and orders. Its invite programs let merchants set link parameters, customer discounts, and commission percentages.
Those settings describe how the program operates. They do not show whether a customer already intended to buy.
For each report, write down the rules that produced its total:
- Which link or code qualifies an order for credit?
- How long after a qualifying interaction can a purchase receive credit?
- Which partner receives credit when signals conflict?
- Does reported revenue include discounts, tax, shipping, cancellations, or returns?
- Can the same order appear in several channel reports?
Use your provider's documented settings. Do not assume every affiliate system has the same window or gives codes priority over links. If link parameters and commission tracking are being mixed together, separate the jobs of affiliate links and UTMs before reconciling totals.
Modash's ecommerce affiliate guide discusses two useful warning signs: codes spreading to third-party sites and customers returning through another channel. These explain why recorded credit can differ from a customer's route to purchase. Neither observation supplies the missing comparison with no affiliate activity.
Four orders, two different questions
The following orders are hypothetical. All amounts are USD merchandise revenue after discounts and refunds, excluding tax and shipping. No further returns are assumed. The imagined tracking system credits A, B, and C once each; it does not credit D. The customer histories are illustrative, not verified observations.
| Order | Revenue | Recorded journey | Tracking status | Causal interpretation |
|---|---|---|---|---|
| A | $100 | A first-time buyer follows a creator's demonstration and uses its link. | Tracked | Potentially incremental; purchase without the creator remains unknown. |
| B | $80 | A repeat buyer has a cart ready, then finds the creator's code on a coupon site. | Tracked | Existing purchase intent makes substitution plausible; no causal verdict. |
| C | $120 | A first-time buyer clicks a paid search ad, then uses an affiliate link before checkout. | Tracked | Contribution is unknown; several contacts preceded the purchase. |
| D | $60 | A buyer reports seeing a creator's video, then purchases directly without a recorded link or code. | Untracked; affiliate connection reported | Potentially incremental, but the report cannot establish causality. |
The categories overlap. An order can be tracked and potentially incremental. An untracked order can also be potentially incremental. All four orders have an unknown causal effect at the individual level.
A first-time customer is evidence of customer status, not proof that the affiliate caused acquisition. A repeat customer's purchase could also be incremental if the promotion changed what, when, or how much they bought.
Here is what the order table supports:
| Hypothetical calculation | Formula | Result |
|---|---|---|
| Affiliate-attributed revenue | $100 + $80 + $120 | $300 |
| Revenue across all four orders | $100 + $80 + $120 + $60 | $360 |
| Share credited to affiliates | $300 / $360 × 100 | 83.3% |
| Incremental revenue | Revenue with program minus revenue without program | Unknown |
The share rounds to one decimal place. It measures credit within this tiny example, not the program's share of caused sales. Do not add A and D together and call the result incremental revenue. Their stories suggest questions to test; they do not supply answers.
What a comparison can estimate
Google's Conversion Lift documentation explains the causal distinction through treatment and control groups. It compares outcomes when ads are present with outcomes when they are withheld. That documentation concerns Google Ads. It does not mean an affiliate report has a built-in lift experiment.
For an affiliate test, define the change you can control. You might test an additional recruitment offer or a new promotion. A test of that addition estimates its effect relative to the existing program. It does not estimate the value of every affiliate relationship you already have.
Where feasible, randomly assign eligible units before the change starts. Use a comparable control condition and measure store revenue across channels in both groups. Counting only affiliate-coded purchases would miss sales that moved to direct checkout or another channel.
For equal-sized randomized groups observed over the same period, the illustrative calculation is:
| Hypothetical test | Value |
|---|---|
| Total net revenue in treatment group | $30,000 |
| Total net revenue in control group | $27,000 |
| Estimated incremental revenue for treatment group | $30,000 − $27,000 = $3,000 |
| Estimated lift relative to control | $3,000 / $27,000 × 100 = 11.1% |
The percentage rounds to one decimal place. These are invented totals showing arithmetic, not a study result. They assume equal group sizes, consistent revenue definitions, complete outcome measurement, and no exposure crossing between groups. Unequal groups require an appropriate adjustment. The totals alone cannot establish statistical uncertainty or justify rolling the estimate out to a larger audience.
Use the guide to measuring incremental impact when attribution is incomplete to choose a study design before launching a holdout.
When the test cannot settle the budget decision
Public creator posts make exposure difficult to contain. A shopper in a control region can still see a post or receive a shared code. If both groups encounter the promotion, the comparison no longer represents a clean with-versus-without test.
Low order volume creates another limit. Google's study setup guidance identifies duration, traffic split, conversion actions, historical data, and expected lift as factors in study power. Affiliate studies also need a feasibility assessment suited to their design. There is no universal minimum order count that makes every test decisive.
Treat the following as planning checks:
- Define the smallest revenue change that would alter your spending decision.
- Estimate whether the available sample can detect that change.
- Choose the observation period and return treatment before seeing results.
- Record overlapping promotions, stock shortages, and exposure across groups.
- Report uncertainty alongside the estimate. A wide interval that includes no effect leaves the decision unresolved.
A before-and-after increase during a seasonal sale is descriptive evidence. It cannot by itself separate affiliate activity from seasonality, discounting, or other campaigns. Keep the incrementality field marked unknown when the comparison cannot support a causal claim.
Keep commissions and disclosures on their own terms
Do not turn an unresolved incrementality estimate into an ad hoc reason to reject credited orders. Reconcile payouts against the agreed program terms. Use causal evidence to inform future budgets and program design.
For US-facing endorsements, FTC guidance calls for clear, conspicuous disclosure of affiliate relationships. It warns that an affiliate-link label alone may not communicate payment. A personalized discount code may also fail to explain the financial relationship. Measurement uncertainty does not remove that relationship. This is operational guidance, not legal advice.
At your next revenue review, add separate fields for attributed revenue, the attribution rules, and the incremental estimate with its uncertainty. If no credible comparison exists, enter "unknown" for incrementality and name the specific program change you want to test next.
