Branding agencies attribution methods: four ways

Branding agencies attribution methods answer different questions, so choose one from the decision rather than its apparent sophistication. The four common approaches are multi-touch attribution, marketing mix modelling, brand lift studies and geo holdout or incrementality tests. Attribution methods are not increasingly impressive versions of the same calculation, and none of these four proves causation on its own.

This guide compares the four attribution methods most often used on branding-agency engagements affecting organisations in England. The branding-agency market includes names such as Interbrand, Landor and Wolff Olins, and these methods are common across the sector. When you shortlist an agency, ask which of the four sits behind the results it presents, and whether the claim is attributed or causal. UK government evaluation guidance is the analytical reference here, not a requirement for private agencies.

What to take away

Common comparison basis

Each approach is assessed against the same fields: question answered, eligible population, outcome unit, period, comparison logic, main evidence, central limitation and permitted wording. The outcome definition and unit must remain identical if results are placed side by side.

Multi-touch attribution

Multi-touch attribution allocates recorded outcomes across paid, owned and earned touchpoints using first-contact, last-contact or weighted rules. It can reveal which touchpoints precede conversion, but it has no counterfactual and cannot distinguish branding work from price, distribution, product or wider changes.

Use it for reporting and commercial allocation. Say "attributed" rather than "caused" when the method only applies a credit rule.

Marketing mix modelling

Marketing mix modelling uses historical sales and marketing data to estimate the contribution of channels such as brand campaigns, price, distribution and promotions. State the model specification, data sources, time period and assumptions. Apply the same outcome unit and period to every model in the comparison.

This can support budget planning and channel comparison, but it does not prove that branding caused the estimated contribution. Model outputs are not causal certainty.

Brand lift studies

Brand lift studies measure changes in brand awareness, consideration or preference by comparing an exposed group with a control group. State the exposure definition, survey instrument, sample and control construction. Platforms such as Meta Brand Lift and Google Brand Lift offer standardised versions.

This method is useful when a campaign can be split into exposed and control groups, but it rarely isolates a single agency's work. A lift is not an effect size unless the design supports a credible counterfactual.

Geo holdout and incrementality tests

Geo holdout or incrementality tests compare treated and control regions. HM Treasury's Magenta Book, the UK government's evaluation guidance, sets out experimental and quasi-experimental families, their assumptions and limitations. Spillover, simultaneous changes, selection and small numbers can still limit causal claims, so qualified evaluators should decide whether a design is viable.

These tests can support causal language only when the comparison is credible. A causal estimate does not by itself explain delivery or establish value.

What to ask a branding agency

Selection rule

Use multi-touch attribution for reporting and allocation. Use marketing mix modelling for budget planning across channels. Choose brand lift studies for brand awareness and consideration effects. Reserve incrementality language for an assured geo holdout or comparable counterfactual design.

Before release, record the method, version, population, geography, period, unit, assumptions, uncertainty and stop condition. The AQuA Book calls for verification and validation across analysis. Stop causal wording if comparison assumptions fail, exposure cannot be reconstructed or the eligible population changes without a defensible bridge. Reopen the choice when the decision, outcome definition or available comparison evidence changes materially.

Before you act

Common questions

What is the main limitation of descriptive monitoring?

Descriptive monitoring has no counterfactual. It can reveal an operational movement, but it cannot distinguish branding work from price, distribution, product or wider changes. Use it for delivery management and for deciding where investigation is needed, and say observed rather than generated when the design supports description only.

When is contribution analysis appropriate?

Contribution analysis is useful when experimental control is limited, provided the claim remains bounded. It builds a theory of change, checks whether intended activity occurred, inspects expected intermediate evidence and tests competing explanations. Its result is not an effect size, and contrary evidence should be published with the judgement.

What should you record before releasing any attribution result?

Before release, record the method, version, population and geography. Also record the period, unit, assumptions, uncertainty and stop condition. The AQuA Book calls for verification and validation across analysis. Stop causal wording if comparison assumptions fail, exposure cannot be reconstructed or the eligible population changes without a defensible bridge.