Retail media measurement should answer three different questions: what the media delivered, what commerce outcome received credit, and what changed because the media ran. It should then connect the incremental outcome to the cost and contribution of the products sold.
The challenge is that retail media networks use different data, identities, formats, sales scopes, attribution windows, new-to-brand definitions, and reporting delays. A brand can collect many precise-looking numbers that are not comparable. Strong measurement begins with transparent definitions and a decision, not a dashboard export.
Quick answer: Measure retail media by documenting the audience, exposure, product scope, sales channel, attribution window, returns treatment, and cost; separating attributed from incremental outcomes; selecting a credible counterfactual; and calculating contribution after media, retailer, fulfillment, discount, and return costs.
Table of Contents
- Industry principles for retail media measurement
- The retail media measurement ladder
- 1. Write the measurement question first
- 2. Document the sales scope
- 3. Treat new-to-brand as a defined metric
- 4. Build a credible counterfactual
- 5. Control bias and contamination
- 6. Match the method to the decision
- 7. Connect media outcomes to contribution
- 8. Reconcile online and in-store outcomes
- 9. Build cross-retailer comparability carefully
- 10. Create a measurement governance record
- How Eva manages retail media measurement
- Retail media measurement FAQ
Industry principles for retail media measurement
The IAB/MRC Retail Media Measurement Guidelines emphasize transparency and consistency. The IAB and IAB Europe Guidelines for Incremental Measurement in Commerce Media organize methods around credible counterfactuals, control of bias, and separation of signal from noise.
Those principles matter because a sale after an ad exposure is not automatically caused by the ad. The customer may already prefer the brand, search for the product, or buy on a regular cycle. Measurement quality depends on estimating what would have happened without the investment and stating the uncertainty honestly.
The retail media measurement ladder
| Level | Question | Example evidence |
|---|---|---|
| Delivery | Did the media reach the intended environment? | Impressions, viewability, completion, frequency, and spend |
| Engagement | Did the shopper interact? | Click, product view, search, detail-page visit, and add to cart |
| Attributed commerce | Which sales received model credit? | Attributed orders, units, revenue, ROAS, and new-to-brand |
| Incrementality | What would not have happened without media? | Incremental sales, buyers, penetration, or visits versus counterfactual |
| Economics | Was the added outcome valuable? | Contribution, payback, marginal return, and customer value |
1. Write the measurement question first
A launch campaign may ask whether media created trial among category buyers who had not purchased the product. A search campaign may ask whether a higher placement added sales rather than shifting organic orders. An off-site video campaign may ask whether exposed regions produced more new buyers and branded search.
The question determines the outcome, comparison, window, and method. Do not use one last-click ROAS report to answer awareness, acquisition, incrementality, and profit at the same time. Assign a primary question and a small number of supporting diagnostics. Extra metrics are useful only when they explain the result or protect a guardrail.
2. Document the sales scope
Define whether reported sales include the advertised SKU, the brand’s products, related products, online sales, physical-store sales, pickup, delivery, marketplace transactions, or a combination. Record the order status, currency, tax, shipping, discount, cancellation, and return treatment. A larger sales scope usually produces a larger attributed result, so it must be visible.
Also document the attribution window and event that starts it. Click and view windows should be labeled separately. For in-store media, state how exposure is estimated or observed. For cross-device and off-site programs, record the identity and match methodology in the limits allowed by the provider. Never compare two networks without reconciling these definitions.
3. Treat new-to-brand as a defined metric
New-to-brand can mean no recorded purchase from the brand within a particular retailer lookback window. The exact lookback, channel, identity, and product scope matter. It does not necessarily mean the customer has never purchased the brand anywhere or that the media caused the first purchase.
Use the metric to understand the credited customer mix, then combine it with incrementality and customer value. Track the product purchased, repeat behavior where available, contribution, and whether the customer was already buying the category. A campaign that acquires new brand customers at a sustainable cost can deserve a different budget than one that mostly reaches loyal buyers.
4. Build a credible counterfactual
The counterfactual estimates what would have happened without the media. Randomized experiments can provide strong causal evidence when assignment, sample, compliance, and contamination are managed well. Geo tests, audience holdouts, product tests, time-based tests, and store tests may be practical depending on the network and objective.
When randomization is not available, model-based comparisons, matched markets, econometric methods, and hybrid proxies can be useful, but they rely on assumptions. Record how groups were selected, what variables were controlled, what other marketing changed, and how uncertainty was estimated. A method should be as rigorous as the budget decision requires.
5. Control bias and contamination
Selection bias occurs when exposed customers differ from the comparison group before media. A network may reach heavier category buyers, more active shoppers, or markets with higher distribution. Seasonality, price, promotion, inventory, competitor activity, and other campaigns can also affect the result.
Contamination occurs when the control group receives the campaign through another format, device, retailer, market, or shared household. Monitor overlap and record concurrent investment. Use pre-period balance checks and guardrail outcomes. If the test cannot isolate the intended difference, report the limitation rather than forcing a precise lift number.
6. Match the method to the decision
| Method family | Useful for | Main caution |
|---|---|---|
| Randomized experiment | Causal test with controlled exposure | Scale, compliance, contamination, and implementation |
| Matched comparison | Geo, store, product, or audience comparison | Unobserved differences and match quality |
| Econometric model | Portfolio and historical contribution | Data volume, assumptions, and correlated media |
| Platform model | Fast optimization inside a network | Method transparency and cross-network comparability |
| Hybrid proxy | Directional decision where perfect design is unavailable | Lower causal confidence and risk of overstatement |
Use stronger methods for decisions with more money, risk, or strategic importance. A small creative test may rely on platform conversion evidence. A major annual retailer allocation may justify an incrementality experiment and broader model. Measurement cost should be proportionate, but lack of perfect data should not excuse unclear definitions.
7. Connect media outcomes to contribution
Incremental revenue is not incremental profit. Subtract product cost, retailer or marketplace terms, fulfillment, discounts, returns, media, creative, data, measurement, and variable operations as appropriate. If the campaign changes product mix, use the actual mix rather than one average margin.
Calculate marginal contribution for the added outcome. A campaign can have a lower attributed ROAS and still be more valuable if it creates incremental new customers with strong retention. Another can show high ROAS while mostly capturing existing demand. The economic model should preserve both short-term contribution and strategic customer value.
8. Reconcile online and in-store outcomes
Retail media can influence store, pickup, delivery, and marketplace sales. Document which channels are observed and which are missing. Distribution changes, planogram changes, store availability, and local promotion can create apparent media effects. Use store-level or regional operational data when the campaign reaches physical retail.
Do not force online and in-store metrics into one number without showing the source. Report the channel mix and confidence. For in-store formats, verify opportunity to see, store compliance, screen or placement uptime, and product availability. Media cannot create a sale when the product is not on the shelf.
9. Build cross-retailer comparability carefully
Create a common layer for spend, impressions, clicks, attributed sales, new-to-brand, incremental outcome, and contribution, but retain the network-specific definition beside it. Normalize currency and time zone. Map products and categories consistently. Keep click and view attribution separate and label online-only versus omnichannel sales.
Use ranges or confidence levels when definitions cannot be reconciled. The goal is a better allocation decision, not artificial precision. A network with narrower sales scope can look weaker than one that credits the full brand across a longer window. Transparency lets the team interpret the difference instead of rewarding the most generous methodology.
10. Create a measurement governance record
- Business question and decision owner
- Campaign, retailer, audience, product, format, and dates
- Exposure, sales, attribution, and new-to-brand definitions
- Counterfactual and method selection
- Pre-period balance and concurrent activity
- Returns, cancellations, inventory, price, and promotion treatment
- Media and full variable cost
- Result, uncertainty, limitation, and next action
Preserve the record with the result. This makes learning reusable and prevents a future team from comparing tests that used different rules. Review definitions whenever a network changes its interface, methodology, identity, or product. Measurement governance is an operating responsibility, not a one-time analytics project.
How Eva manages retail media measurement
Eva connects retailer media reports with marketplace and store sales, product economics, inventory, pricing, content, promotions, returns, and broader channel activity. Senior operators define the decision and measurement approach before investment, then reconcile the reported result with contribution and operational reality.
This reduces the risk of optimizing the network with the most generous attribution. Eva helps brands compare opportunities across Amazon, Walmart, Target, Instacart, Shopify, Google, Meta, and other environments while preserving the limits of each dataset. The result is a portfolio decision the business can defend.
Retail media measurement FAQ
What should retail media measurement include?
Include delivery, engagement, attributed commerce, customer evidence, incrementality, and economics. Document the sales scope, window, identity, product set, returns, cost, and method.
Is retail media ROAS enough?
No. ROAS can be useful for platform optimization, but it does not show incrementality, full variable cost, contribution, or customer value. Use it inside a broader decision model.
How is retail media incrementality measured?
Methods include randomized experiments, matched comparisons, geo or store tests, model-based counterfactuals, econometric analysis, and hybrid proxies. The appropriate choice depends on the decision, data, scale, and risk.
Can retail media networks be compared directly?
Only after reconciling sales scope, attribution, new-to-brand, identity, time zone, currency, product mapping, returns, and cost. Keep network-specific definitions visible when they cannot be normalized.
What is a credible counterfactual?
It is a defensible estimate of what would have happened without the media. Credibility depends on comparable groups, control of bias, limited contamination, appropriate timing, and transparent assumptions.
Related Eva resources: Marketplace Expansion, Full-Funnel Commerce Playbook, Retail Media Strategy, Amazon Marketing Cloud Measurement, and Instacart Advertising Strategy.
