Measurement
Ecommerce platforms attribution methods: model limits
Ecommerce platforms attribution methods: common credit rules, coverage limits and a merchant-ledger baseline for comparing channel credit.
Ecommerce platforms attribution methods should be compared on the same accepted-order population, period, timezone, currency, interaction evidence, exclusions and identity coverage. No attribution rule identifies a universal winner. Compare methods on that basis.
This desk comparison was researched on 6 September 2026 for one England-based seller of its own physical goods to adults. No campaign or platform account was examined.
What to take away
- Compare methods on one fixed accepted-order population before interpreting differences.
- Supplier models answer bounded channel-credit questions only when labelled with model and coverage.
- The merchant ledger reports observed orders, captures, settlements, refunds and net contribution without channel credit.
- Causal incrementality needs a pre-specified counterfactual design, not monitoring of a changed outcome.
- Match the method to the question it can answer.
Common comparison case
The decision is whether channel credit is informative enough to guide a controlled marketing test. The unit is one eligible accepted order, with attributed net contribution shown only after accountant-approved revenue and cost mapping. Retain unattributed orders and exclude only documented synthetic tests, duplicates and invalid events. Permission-denied or blocked tracking remains missing coverage.
Attribution rules compared
First click
- Credit rule
- first recorded interaction
- Useful question
- which entry gets credit
- Main limitation
- earlier contact missing
Last click
- Credit rule
- final recorded interaction
- Useful question
- which interaction closed path
- Main limitation
- earlier contribution disappears
Last non-direct
- Credit rule
- final non-direct interaction
- Useful question
- which external channel closes
- Main limitation
- direct and identity gaps
Linear
- Credit rule
- equal across recorded clicks
- Useful question
- how credit changes when shared
- Main limitation
- equal shares are a rule
Shopify's current marketing reports provide the first-party model definitions below. They are supplier rules, not England law or causal evidence.
Credit rule
- First click
- all credit to the first recorded interaction
- Last click
- all credit to the final recorded interaction
- Last non-direct click
- final recorded non-direct interaction gets credit
- Linear
- equal credit across recorded clicks
- Any click
- full credit to every clicked channel
Useful question
- First click
- which observed entry interaction receives credit
- Last click
- which observed interaction closed the recorded path
- Last non-direct click
- which observed external channel receives closing credit
- Linear
- how credit changes when all visible touches share it
- Any click
- whether one channel appeared in attributed paths
Main limitation
- First click
- earlier or unobserved contact is missing
- Last click
- earlier contribution disappears
- Last non-direct click
- direct activity and identity gaps alter allocation
- Linear
- equal shares are a rule, not measured contribution
- Any click
- totals can exceed observed order count
Illustrative example: five £100 orders with these touch paths. O1: PS > EM > D. O2: EM > D. O3: PS > D. O4: SO > PS > EM. O5: D > PS.
| Model | Paid search | Direct | Social | Total credit | |
|---|---|---|---|---|---|
| First click | £200 | £100 | £100 | £100 | £500 |
| Last click | £100 | £100 | £300 | £0 | £500 |
| Last non-direct | £200 | £300 | £0 | £0 | £500 |
| Linear | £166.66 | £116.66 | £183.33 | £33.33 | £500 |
| Any click | £400 | £300 | £400 | £100 | £1,200 |
Any-click credit sums to £1,200 because every clicked channel receives full credit.
A 30-day click window is a typical default because it covers many short purchase cycles and limits credit to interactions close to the order. Reproduce all models over that window before interpreting differences. Changing the window, order event or channel mapping makes the comparison invalid.
Add a merchant-ledger baseline
Report observed accepted orders, captures, settlements, refunds and net contribution without channel credit. This baseline answers what the merchant recorded, not why it happened. Shopify's sales-report definitions state that its sales reports do not track payments moving between customer and merchant, so its attributed sales need a separate reconciliation bridge.
Coverage reconciliation steps
- Report observed accepted orders, captures, settlements, refunds, net contribution
- Reconcile eligible ledger orders to each attribution source
- Show matched, unmatched, duplicated and late cases
- Break out permission-denied traffic only if permitted
- Otherwise keep permission-denied within unknown coverage
Google's global GA4 purchase-event guide shows that analytics values depend on a configured purchase event and can take time to appear in reports. It does not verify the merchant's implementation or match a platform order automatically.
Add a coverage table before comparing credit. Reconcile eligible ledger orders to orders visible in each attribution source, showing matched, unmatched, duplicated and late cases.
For an England-based seller, the Information Commissioner's Office guidance on consent and coverage governs how permission-denied traffic is classified.
Break out permission-denied traffic only when the privacy design and data permit that classification. Otherwise keep it within unknown coverage, because a model comparison on different visible populations answers a data-capture question before it answers a channel question.
Reserve causal language for an impact design
To test whether a campaign caused incremental contribution, pre-specify intervention, eligible population, assignment or comparison, outcome and analysis period. Add exclusions and the stopping rule.
HM Treasury's QPIE guidance explains that monitoring a changed outcome does not establish responsibility; a credible counterfactual is central to causal attribution. Its scope is government evaluation; use it as methodological discipline, not an ecommerce standard.
Decision
Use supplier attribution for bounded channel-credit questions, always labelled with model and coverage. Use the merchant ledger for observed commercial events. Use an appropriately reviewed counterfactual design for causal incrementality. Withhold a recommendation when identity coverage, permission, reconciliation or comparison quality is insufficient. No model should be selected because it gives the largest number.
Before you act
- Fix a 30-day click window, order event and channel mapping.
- Reconcile ledger orders to each attribution source.
- Show matched, unmatched, duplicated and late cases.
- Label every attribution figure with model and coverage.
- Keep permission-denied traffic inside unknown coverage.
- Pre-specify the intervention and counterfactual before causal claims.
Common questions
Why can any-click totals exceed the observed order count?
Any click credits every clicked channel, so one order can credit several channels and totals can exceed the order count.
What does the merchant-ledger baseline actually answer?
It reports observed commercial events, not channel credit or causes.
What makes a comparison unsuitable for a recommendation?
Insufficient identity coverage, permission, reconciliation or comparison quality.
