Costs and pricing
Retail media return on investment, with attribution labelled as allocation
Retail media return on investment for England separates recorded sales, attribution, causal contribution and projections without inventing costs or results.
What to take away
- A defensible return uses a declared net-return ratio(G minus C) divided by C, where G is incremental contribution before campaign costs and C is incremental campaign cost for the same scope and period.
- Attribution is an allocation rule. Changing the lookback period or the eligible touchpoints changes the attributed figure without changing shopper behaviour.
- A causal estimate needs a counterfactual and a design that addresses rival explanations, following the Magenta Book analytical methods annex.
- This article gives no ROI figure and no England benchmark. Every input stays marked UNKNOWN until a named owner supplies evidence.
- Projected returns belong in a separate labelled account, away from recorded sales and causal results.
Set the perspective and the cost boundary
Decide whose return you are calculating. An advertiser's campaign return and a retailer's advertising revenue are different benefits and cannot be pooled.
Fix the activity, the period and the cost lines before any arithmetic. Carry that boundary through every later table so the comparison does not drift.
Cost should be defined in one line: media spend, platform fees, agency fees, creative production, measurement tooling and internal staff time. Say which items sit inside C and which stay outside.
Seasonal shifts recorded in the ONS retail industry statistics mean a comparison period can move a result without any campaign effect.
The retail media budget template shows how to carry unknown cost lines instead of filling them with averages.
Keep recorded sales in their own account
Record what the transaction system reports: the date range, cancellations, returns and tax treatment. Keep the source file and its reconciliation status.
A recorded purchase is an observation. It does not say whether an advert caused it.
Treat delivery events the same way. Ad requests, served impressions, viewable impressions and clicks measure different things. None turns into a sale because it sits beside a revenue column.
Recorded outcomes versus delivery events
Recorded outcomes
- What it is
- Transaction system report
- Date range
- Required
- Cancellations and returns
- Must be stated
- Tax treatment
- Must be stated
- Source file
- Preserve and reconcile
- Causal claim
- Observation only
Delivery events
- What it is
- Ad requests, impressions, clicks
- Date range
- Required
- Cancellations and returns
- Not applicable
- Tax treatment
- Not applicable
- Source file
- Preserve and reconcile
- Causal claim
- No sale inferred
Label attribution as allocation
Specify the attribution rule, the lookback period, the eligible touchpoints and the sales definition. Keep attributed sales in a separately labelled row.
Retailer media owners such as Tesco Media and Insight Platform, Nectar360 and Boots Media Group report attributed sales inside their own dashboards. Each dashboard applies its owner's allocation rule.
Marketplaces and off-site networks, including Amazon Ads and CitrusAd, apply their own rules too. Two dashboards can show different attributed totals for one order.
A ratio of attributed sales to advertising spend keeps its own label. It should not be renamed profit ROI.
If you need to compare rules, click-based or view-based attribution for retail media sets them side by side without choosing a winner.
Label attribution as allocation
- Specify the attribution rule
- State the lookback period
- List eligible touchpoints
- Define the sales definition
- Keep attributed sales separately labelled
- Do not relabel as profit ROI
Build a causal contribution estimate
The Magenta Book analytical methods annex covers experimental and quasi-experimental evaluation designs. It is UK government methodological guidance, not a legal duty for private campaigns.
Pick a design that answers your question: a holdout test, a matched comparison or an interrupted series. Record contamination risks and uncertainty alongside the estimate.
Where a test uses shopper-level data, you need a lawful basis under UK GDPR. The ICO guide to the UK GDPR sets out the conditions for that processing.
Example: net-return ratio with placeholder inputs
The table below is arithmetic only. No campaign produced these numbers, and the values are placeholders.
Net-return ratio with placeholder inputs
| Input | Placeholder value | Status |
|---|---|---|
| Incremental contribution before campaign cost (G) | £24,000 | UNKNOWN, pending finance sign-off |
| Incremental campaign cost (C) | £10,000 | Partly known: £7,000 media booked |
| Net-return ratio, (G minus C) divided by C | 1.40 | Illustrative only |
Work through it in five steps. Each step needs a defined metric, and retail media metrics from ad request to incremental sale supplies the definitions before any targets are set.
Net-return ratio with placeholder inputs
- Agree the scope and period, then fix the C boundary with finance.
- Reconcile recorded sales and delivery events back to source files.
- Label every attribution output as allocated sales.
- Obtain a causal estimate, or record the incremental result as UNKNOWN.
- Compute (G minus C) divided by C, or leave the ratio UNKNOWN.
If C is not positive, or either value is unsupported, the ratio stays UNKNOWN. G must not already deduct C.
Keep projections apart from evaluation
A forward scenario may use documented assumptions. Label its output projected, and record the source and owner of each assumed input.
Do not move a projected figure into the recorded column, and do not call it a causal estimate from a completed study.
Publish the evidence limits, the unanswered questions and the reviewer's name next to the decision. This is not financial or tax advice.
Common questions
What is the difference between attribution and a causal estimate?
Attribution allocates sales using a rule, a lookback period and a sales definition, so the result moves when the rule moves. A causal estimate needs a counterfactual and a design that addresses other explanations.
How is the net-return ratio calculated?
It is (G minus C) divided by C. G is incremental contribution before campaign costs; C is incremental campaign cost for the same scope and period. G must not already deduct C.
Can a projected return be shown as recorded performance?
No. A scenario may use documented assumptions, but its output is labelled projected. Keep the source and owner of every assumed input.
