A marketing report can look precise and still answer the wrong question. An ad platform shows strong ROAS, the CRM credits a different source, and total profit barely moves. The problem is often not one broken metric. It is the expectation that one measurement method should explain everything.
Google's 2026 measurement guidance explicitly combines three perspectives: attribution, incrementality experiments, and marketing mix modeling. They are not competing versions of truth. Each observes a different part of the system.
Good measurement does not search for one perfect number. It combines several imperfect signals and communicates the limits of confidence.
Attribution: who appeared on the path
Attribution assigns conversion credit across observable touchpoints. It is useful for daily work: comparing campaigns, investigating journeys, spotting a drop, or changing a bid, audience, or creative.
But attribution mainly asks “what appeared on the observed path before the conversion?” It does not always answer “would the conversion have happened without this advertising?” A person who already intended to buy may search for the brand and click an ad, giving the last interaction more credit than the demand it created.
Consent, devices, retention windows, and missing interactions also limit the data. Modeling can recover part of the missing picture, but an observational model is not the same thing as a controlled test.
Incrementality: what happened because of the intervention
An experiment compares a group exposed to marketing with a suitable control group. The difference estimates the additional effect—the outcome that probably would not have occurred without the campaign.
This is a strong way to test an expensive assumption:
- does a channel create new sales or capture existing demand;
- should the budget increase;
- does branded advertising add value beyond organic demand;
- does a new creative change business outcomes rather than click-through rate alone?
Experiments require discipline. Define the business metric, minimum detectable effect, treatment and control, duration, and decision rule before launch. Changing budgets and audiences halfway through weakens the conclusion.
Where individual randomisation is impractical, geographic testing may be useful. Meridian GeoX, for example, supports holdback, go-dark, and heavy-up designs across regions.
MMM: how the complete mix works over time
Marketing Mix Modeling estimates relationships between aggregated marketing activity, external factors, and business outcomes over time. A model can account for multiple channels, seasonality, pricing, promotions, distribution, and other drivers of sales.
MMM is particularly useful for budget planning and channels that are difficult to observe in an individual journey: outdoor media, television, events, offline sales, or a long B2B cycle. It works with aggregated data rather than following each person.
A model, however, cannot create truth from weak inputs. It needs sufficient history, variation in spending, reliable business outcomes, and explicit assumptions. If budgets never changed, it becomes difficult to separate their effect from everything else.
How the three methods work together
Think of three management layers.
Daily: attribution
Use it for campaign diagnostics, traffic quality, conversion paths, and operational optimisation. Do not describe every platform-attributed conversion as an incremental sale.
Periodically: experiments
Test the most consequential assumptions. An experiment can estimate the causal effect of a channel or budget change and provide a stronger reference point.
Strategically: MMM
Assess budget allocation, longer-term effects, and the interaction of channels. Strong experiment results can calibrate the model and reduce uncertainty.
When the methods disagree, do not choose the prettiest result. Check differences in the time period, outcome definition, channel coverage, and assumptions.
Where to begin without a large analytics team
You do not need to start with an advanced model. Build the measurement foundation first.
1. Define the business outcome
Decide what counts: paid revenue, contribution margin, a qualified opportunity, repeat purchase, or retention. A raw lead may be too early to guide investment.
2. Connect media, CRM, and finance
Agree on channel names, lifecycle stages, currencies, time zones, and refund rules. Validate the meaning of the data, not only its presence.
3. Separate observed and modeled results
A report should make clear which conversions were observed directly, which were modeled, and which conclusions depend on assumptions.
4. Plan one causal test
Choose a decision where being wrong is expensive: raising spend, keeping a brand campaign, or launching a new channel. Write the hypothesis before launch and do not move the success criterion after seeing the result.
5. Create a measurement calendar
Operational reporting may be weekly, an experiment follows its pre-defined window, and a budget model is refreshed less often. Different cadences protect the team from overreacting to random weekly movement.
Five questions for any impressive dashboard
- What business decision will this number change?
- Is this an observed association or evidence of causality?
- Which parts of the journey are invisible?
- What is the comparison or counterfactual?
- What outcome would make us reject the hypothesis?
If these questions have no answer, decimal precision only creates an illusion of control.
The main point
Attribution helps manage what is visible today. Experiments show what changed because of marketing. MMM helps allocate resources across the complete system. Mature measurement combines all three with CRM and financial data—and is honest about where it has evidence and where it has only a working hypothesis.