Marketing attribution strategy: matching claims to evidence

Frame a paid-search allocation decision

A marketing attribution strategy can frame a paid-search allocation decision, but this article does not supply campaign data or choose an allocation. It shows what a buyer should define before deciding whether evidence supports a change.

No campaign result is assumed. The focus is what evidence could justify a change, who may approve it, and which unresolved limits mean an allocation should stay as it is.

The buyer frames the question as one bounded allocation choice, not a search for a universally best channel. Before opening platform reports, the buyer records what may change, the outcome that matters, the authorised approver and the point at which the decision will be reviewed again. A blank planning template can structure that record.

The record fixes seven items before any tool discussion:

The buyer also maps context, inputs, activities, outputs, outcomes, assumptions and risks. The Government Analysis Function's Theory of Change toolkit was created for public interventions rather than commercial attribution. A private buyer may adapt its sequence, while making clear that this does not imply government endorsement.

The 2027 label sets a review horizon, not a forecast of adoption or performance. Platform, privacy and advertising sources can change quickly, so the buyer plans to check each source again on publication day. The record also sets a date to reconsider the allocation and names an authorised decision owner.

Checks: define the outcome and eligible population

The buyer checks whether the outcome and eligible population can be identified consistently. Unless both are fixed and tied to a reproducible source, reports may describe different events or groups, making any apparent agreement unsuitable evidence for the allocation decision.

The outcome record names the event, unit, source system, query or report version, deduplication rule, relevant currency, inclusion and exclusion rules, latency and data owner. These details let a later reviewer reproduce what was counted. The measurement jobs stay separate because allocated credit does not by itself establish a causal effect.

The buyer distinguishes order value before cancellations from settled contribution after refunds and direct costs. Those figures come from different definitions and cannot be exchanged without changing the question. The buyer also makes clear whether the outcome is an accepted order, a page view or another event, rather than allowing a platform's configured event to decide silently.

For England, the buyer chooses one population rule that suits the decision. It could use an organisation's registered office, a trading site, a customer address or a campaign targeting setting. Unknown and excluded cases stay visible. UK evidence may inform an England operating decision, but its source geography must not be relabelled as England.

The Government Analysis Function's guidance on communicating quality and uncertainty asks official-statistics producers to describe bias, uncertainty and distortive effects in source data. It is not a marketing standard. The buyer applies the useful discipline of placing material weaknesses beside a result, rather than leaving readers to discover them in a methods appendix.

Checks: compare evidence with the claim it can support

The buyer checks what each source can establish and keeps unlike outputs separate. Platform credit, reconciliation, modelling and tests answer different questions; none should be treated as interchangeable simply because each produces a report or estimate.

The buyer compares order identifiers with campaign records to find agreement, differences and unmatched cases. This is a reconciliation check, not proof that a channel caused an order. The record also guards against reconciliation pitfalls, including combining platform totals that use incompatible scopes.

Google's current Analytics attribution overview describes attribution as assigning credit to touchpoints and lists models available in that product. It is first-party product documentation that helps explain Google Analytics reporting, not independent proof of business impact. The buyer keeps the platform scope and configuration with each result.

Google's Ads attribution-report documentation explains that reporting can differ by event time, network coverage, source and lookback window. Credits exported from separate platforms must not be added as if they share an eligible population and denominator. Each result remains a separate observation, recorded with its configuration and extraction date.

Marketing mix modelling uses aggregate time-series data and a specified model. Google's Meridian documentation is a current first-party record for one open-source implementation, but it cannot validate the buyer's variables, priors, geographic units or conclusions. A statistics reviewer examines model choice, confounders, collinearity, missingness, uncertainty and sensitivity before results affect spending.

For a causal question, the buyer plans the counterfactual before an intervention. HM Treasury's Quality in Policy Impact Evaluation guidance distinguishes monitoring an outcome from establishing whether an intervention caused it. The current Magenta Book covers evaluation scoping, design, conduct, use and dissemination for government.

A commercial buyer may apply the emphasis on early design and proportionate methods while keeping their public-sector scope clear. Movement from a before period to an after period alone does not isolate marketing effect. A controlled test needs a stated comparison and must not be generalised beyond the population, period or channels tested.

Checks: clear permission and independent review gates

The buyer checks whether the measurement design changes what information is collected from people or devices. Privacy review comes before collection decisions because the technology, purpose, exception or consent route, and later processing all matter to the assessment.

The ICO's final storage and access technologies guidance was updated on 29 April 2026 after changes made by the Data (Use and Access) Act 2025. It covers cookies, tracking pixels and device fingerprinting under PECR and, where relevant, the UK GDPR. A qualified reviewer assesses the actual setup; a vendor setting is not a legal conclusion. The data protection review records its findings.

Research, service communication and promotion are not interchangeable. The ICO's electronic-mail marketing guidance was updated on 28 April 2026 and explains PECR concepts for email, text and similar stored messages. Any UK GDPR lawful-basis decision remains separate from the rules that apply to a PECR channel.

Privacy and PECR approval do not replace security, accessibility, advertising or commercial checks. The NCSC's SaaS security guidance says the customer retains responsibility for configuring and using a cloud service. W3C's WCAG 2.2 Recommendation provides a technical accessibility reference; specialist testing still needs to cover the buyer's real reporting and approval tasks.

Finding: no single report establishes causal return

This article provides no campaign finding that justifies changing a paid-search allocation. The evidence sources described here do not establish causal return by themselves. An allocation should remain pending until the outcome, evidence design and independent review gates are sufficiently clear.

Any public performance claim receives its own review. ASA and CAP's substantiation guidance, updated 12 March 2026, says marketers should hold evidence for objective claims capable of substantiation. A modelled or platform-attributed result must not be presented as an observed causal fact. The buyer checks advertising claims separately from the allocation analysis.

Financial use also needs distinct definitions. Gross value, refunds and direct costs remain separate; contribution, attribution assumptions and full economic cost are not merged. HM Treasury's Green Book 2026 is mandatory in its stated public-sector settings, not for ordinary private marketing budgets. Its treatment of appraisal, evaluation and uncertainty can still inform the buyer's review.

The accountant or finance lead approves financial definitions and treatment. Thresholds remain blank until the outcome definition, evidence design, acceptable uncertainty and consequences of a wrong choice have approval. The buyer neither borrows a threshold from another decision nor invents one to make a result appear conclusive.

Rules carried forward to the next allocation

For the next allocation decision, the buyer connects the decision to its evidence through a short chain: decision, outcome, population, options, assumptions, method, uncertainty and action gate. Each link must be explicit enough for a later reviewer to see what was measured and why that evidence was considered relevant.

Name an accountable decision owner, measurement lead, source-system owner, privacy reviewer, security reviewer, accessibility reviewer, advertising or claims reviewer, finance reviewer and commercial owner. A small organisation may assign several roles to one person, but the distinct decisions remain necessary. An evidence-level review helps align the claim with the evidence available.

Pause if source coverage changes materially, identifiers cannot be reconciled, a platform model or window changes, consent or lawful processing remains unresolved, or the counterfactual is contaminated. Pause as well if uncertainty exceeds the buyer's approved limit or an independent gate fails. Record these stop rules before the result is known.

Version the question, dataset, transformation code, platform configuration, model, assumptions and decision. A rerun should show which component changed; an earlier result should not be quietly replaced. Use a UK compliance map alongside the record of the outcome, population, period, channels, sources, counterfactual, uncertainty, owner and stop rule.

Ask qualified reviewers to mark unresolved privacy, PECR, security, accessibility, claims, finance and contract issues. Choose an attribution method or commission implementation only after those gaps are visible. This sequence keeps the buyer's decision in view and limits conclusions to what each evidence source can actually support.