On this page
An attribution report can assign revenue to a rebate campaign without showing that the campaign created that revenue. It describes credit under a chosen model. Incrementality asks a different question: how much would change if the offer were absent?
Both views can be useful, but they support different claims. A merchant needs a claim-evidence matrix so that a dashboard label, internal memo, or published case study does not quietly turn journey credit into causal proof.
Distinguish journey credit from causal effect
Imagine a fictional customer discovers a product through a creator, returns through search, and later clicks a rebate email before purchasing. Depending on the attribution model, credit may be assigned differently across those observed contacts. The purchase itself does not change when the model changes.
Shopify’s marketing reports describe attribution models for understanding channel contributions across customer journeys. That is distinct from a randomized comparison designed to estimate what would have happened without the offer.
The distinction matters when a customer was already planning to buy. A rebate email can receive credit for the final observed contact even if the customer would have purchased at the same time without it. Conversely, an offer may influence a purchase while receiving little credit under a model that emphasizes a different touchpoint.
Do not dismiss attribution because it is not causal proof. It can help trace observed journeys, inspect campaign tagging, and understand where reported credit appears. Its value improves when the claim matches the measurement rather than exceeding it.
Reconcile differing attribution views
When two attribution reports disagree, compare their definitions before assuming one is wrong. Check the attribution model, observation window, customer-identification coverage, included sales channels, date convention, and revenue treatment. A report grouped by purchase date can differ from one grouped by campaign interaction date without either being a duplicate of the other.
Use a reconciliation note for the fictional campaign:
| View | Fictional reported amount | What the number describes | What it does not establish |
|---|---|---|---|
| Last-touch campaign attribution | $30,000 | Revenue credited under that model and scope | That $30,000 would disappear without the reward |
| Another documented attribution view | $22,000 | Credit under a different assignment rule | That the campaign lost $8,000 in real sales |
| Before-versus-after store revenue | +$40,000 | Observed movement across the selected periods | That the rebate caused the whole change |
| Valid experimental estimate | +$6 per assigned customer | Estimated causal difference for the tested offer and population | The effect for every future audience or reward design |
These values are hypothetical and intentionally describe different quantities. They should not be added together. Attribution views allocate or associate the same commercial activity in different ways; an experimental estimate concerns a counterfactual difference.
A difficult case is an order appearing under several platforms’ self-reported campaign totals. Adding those totals can double-count the same revenue. Reconcile the underlying order scope where permitted and appropriate, and label unresolved coverage. Do not treat a sum of platform claims as a larger store sales total.
Write revenue claims that match the evidence
Write the claim first in plain language, then ask what evidence supports it. “The campaign was associated with $30,000 of attributed revenue under last-touch reporting” is narrower than “the campaign generated $30,000 in extra revenue.” The second requires evidence about the counterfactual, not just a different verb in the report.
| Intended statement | Evidence needed | Suitable wording when that evidence is absent |
|---|---|---|
| Campaign received journey credit | Defined attribution report with known scope | State the model and the attributed amount |
| Sales increased after launch | Comparable descriptive series | State the observed change and concurrent events |
| Offer created additional revenue | Credible causal design with uncertainty | Do not claim incrementality from attribution alone |
| Offer will improve future sales | Evidence supporting generalization to the planned setting | Present a testable expectation, not a guaranteed result |
Include uncertainty and limitations near causal estimates. If the experiment covered returning customers for a short period, say so. A sitewide annual revenue claim cannot be obtained merely by multiplying a narrow pilot estimate without additional assumptions about traffic, seasonality, capacity, and persistence.
Keep the claim consistent across internal and external materials. A careful analytics report can become misleading when a sales deck removes the words “attributed,” “estimated,” or “in this test.” The evidence owner should review the final sentence that will be used, not only the underlying spreadsheet.
If the team needs an incremental answer, design the measurement before the next offer is exposed. That may mean a customer holdout or another justified causal approach. More elaborate attribution models can improve journey description without resolving the absence of a counterfactual.
The completed matrix should make it easy to choose an honest statement. Attributed revenue, observed growth, and estimated incremental revenue can sit in the same readout as long as their meanings remain distinct. A merchant then gains a fuller picture of the campaign without asking one measurement method to prove a question it was not designed to answer.
A useful final review is to remove the metric label and ask someone to restate the claim aloud. If “attributed sales” becomes “sales we created” in ordinary discussion, the report needs a clearer explanation near the number. Put the counterfactual question beside the attribution result: “This report does not measure how many of these purchases would have occurred without the offer.” The sentence is specific to the measurement gap and avoids implying that the recorded sales are unreal. They are real observed sales; the unresolved question is the campaign’s additional contribution to them.
Review the reporting capabilities you would use alongside a rebate campaign measurement plan. Explore capabilities.
Source references
This guide is general information, not financial, legal, tax or regulatory advice. Eligibility, card availability, permitted use and responsibilities depend on the applicable offer and card terms.