Attribution and incrementality: two different marketing questions

A marketing review can contain several confident answers to the same apparent question. An advertising platform reports strong return on ad spend. Web analytics gives another channel credit for the conversion. Finance sees little change in total revenue. The disagreement does not necessarily mean that one system is broken. The systems may be answering different questions.

Attribution assigns credit to observed marketing interactions under a defined rule or model. Incrementality asks what changed because of the marketing activity compared with what would otherwise have happened. Attribution can help teams operate campaigns. It does not, by itself, establish the additional sales or profit caused by them.

For a business leader, the distinction matters when approving a budget. For an analyst, it determines the data, assumptions, and study design needed. The aim is not to dismiss platform reporting. It is to use each type of evidence for the decision it can reasonably support.

Why the distinction matters now

Marketers work across retail media, search, social, email, websites, and offline activity. Each environment can observe different interactions and apply different reporting rules. Meanwhile, measurement tooling continues to evolve. Google's open-source Meridian framework, for example, provides marketing mix modeling for estimating channel effects and evaluating budget scenarios under explicit assumptions.

These developments make measurement more accessible, but they do not remove the need for careful interpretation. A richer model is still dependent on the available data and the credibility of its assumptions. A platform-generated number and an independent business outcome should remain distinguishable in the review.

The practical opportunity is to build a measurement plan around business questions instead of collecting more disconnected performance dashboards.

Read attributed performance in its own terms

Before comparing two channel reports, confirm the conversion event, attribution window, time basis, and treatment of returns or cancellations. One report may assign a sale to the date of the ad interaction, while another uses the purchase date. Some reporting includes view-through credit; other reporting may focus on clicks.

The same sale can be credited by more than one platform. Adding platform-attributed revenue across channels can therefore exceed actual sales. Reconcile business revenue through a governed order or finance model, and keep each platform's credited revenue as a separately labeled measure.

Use attribution for questions such as which campaigns are reaching their stated audience, how attributed conversion rates are changing, and where operational investigation is warranted. Do not silently rename attributed return on ad spend as incremental return. That language change would imply evidence the underlying report does not provide.

Understand the counterfactual

Incrementality requires an estimate of what would have happened without the intervention. A well-designed experiment can compare eligible groups assigned to treatment and control. Google's Conversion Lift documentation describes this general treatment-and-control approach for supported studies. Eligibility, design, and available reporting vary, so a team should confirm the specific implementation before planning around it.

A valid comparison needs more than a before-and-after chart. Seasonality, promotions, availability, competitor activity, and price changes can affect sales at the same time as advertising. An increase following a campaign is consistent with an effect, but it is not sufficient proof of one.

Experiments also have limitations. Groups need enough relevant activity, the intervention must differ meaningfully, and spillover between groups can weaken the comparison. Define the outcome and evaluation plan before looking at the result. Otherwise, selecting the most favorable segment after the fact can produce a misleading conclusion.

An illustrative lift calculation

Suppose an illustrative experiment randomly assigns 10,000 eligible customers to treatment and 10,000 to control. Count at most one qualifying purchase per customer during the defined observation window. If 500 treatment customers and 400 control customers make that purchase, the observed rates are 5% and 4%: a one-percentage-point difference and a 25% relative increase.

Under the study's assumptions, the point estimate is 100 incremental purchases for the treatment group. That estimate is not the same as the total 500 purchases, and the example does not establish statistical certainty. Review the uncertainty interval, actual exposure, contamination, and study execution before acting on it. For unequal group sizes, compare normalized rates or the appropriate estimator, not raw counts.

If contribution before campaign cost were $20 per incremental purchase and incremental campaign cost were $1,500, the illustrative contribution after that cost would be $500. Change the margin, lift estimate, or cost boundary and the decision changes. Use finance-approved contribution definitions and avoid presenting a point estimate as guaranteed profit.

Know where marketing mix modeling fits

Marketing mix modeling, often shortened to MMM, examines aggregate outcomes alongside marketing inputs and other relevant factors over time or geography. Meridian supports this type of analysis, including response curves and budget scenarios. Its causal interpretation depends on assumptions about the data and relationships being modeled.

MMM can complement experiments when leadership needs a broader allocation view. It should not be treated as an automatic substitute for a test or a precise explanation of every individual purchase. Correlated channel spending, limited variation, missing controls, and changes in the business can make effects difficult to separate.

Use experiments, operational attribution, and modeling as complementary evidence. Investigate disagreement rather than averaging incompatible estimates into one reassuring number. A channel can look strong under attribution because it reaches existing high-intent buyers while showing a smaller incremental effect in a particular test.

Build a governed measurement dataset

The foundation should connect actual outcomes, media cost, campaign definitions, calendar, geography, product availability, pricing, and relevant events at a compatible grain. Keep source extracts and revisions traceable within the agreed retention policy. Separate observed values, modeled estimates, and assumptions.

Choose the outcome before choosing the model. Revenue may be appropriate for one decision; contribution, new-customer purchases, or qualified opportunities may matter more for another. Define new customer consistently across channels, and explain where identity coverage limits the measure. An unknown customer is not automatically a new customer.

For retail and commerce, inventory constraints are especially important. Advertising cannot produce a fulfilled sale for an unavailable product. Record stockouts and supply changes so the team does not confuse a supply limitation with weak marketing response. The data will not eliminate every confounder, but it makes the remaining uncertainty easier to discuss.

Show the evidence in the budget review

A useful executive grid keeps attributed performance, experimental or modeled incremental estimates, uncertainty, and operational constraints in separate columns. Include the measurement period, evidence age, and the decision the result can support. Avoid a single ranking that hides material differences in study quality.

Present the next action as a bounded decision: maintain the current allocation, run a focused test, investigate a discrepancy, or evaluate a modest budget change. Large reallocations deserve stronger evidence than small reversible experiments. State what observation would cause the team to reconsider.

Do not assume that a result from one season, audience, or spending level will hold everywhere. Response can change as reach expands or conditions shift. Keep the measurement plan connected to the current business rather than treating an old lift estimate as a permanent property of the channel.

Begin with the decision under debate

Choose one disputed budget question and document what the current reporting can and cannot answer. Reconcile the outcome, review attribution definitions, and decide whether the next useful step is an experiment, a model, or better source data. The best measurement program is not the one with the most methods. It is the one that gives leaders appropriate evidence for the next decision.

VeriVal helps teams connect marketing, commerce, and financial data into governed measurement and decision workflows. If campaign reports are clear but the business impact is not, start a conversation.

References

• Google Ads: Conversion Lift
• Google: an introduction to Meridian