The last click is a receipt, not the whole sales story

Attribution can help operate campaigns while missing the work that created demand; decide what each report can establish before using it to move budget.

The content director brings a customer interview to the budget meeting. The buyer remembers a research report, a conference conversation and a colleague’s recommendation. The performance dashboard gives the sale to a branded search advert clicked on the day of purchase.

Neither record must be fraudulent for the accounts to disagree. They answer different questions. One describes remembered influences; the other assigns credit according to an observable path and a reporting rule.

The trouble begins when the CMO treats either account as a complete explanation of what caused the purchase.

Establish what the number means

Google’s attribution documentation describes models that allocate credit to recorded interactions. A last-click model concentrates credit near the end of that path. Other models distribute it differently.

That allocation is useful for comparing campaigns within consistent reporting conditions. It is not the same question as whether the customer would have purchased without the marketing activity.

Put the definition under the chart: event counted, eligible touchpoints, observation window and attribution model. A report without these qualifications invites a stronger interpretation than its data can support.

Then identify missing observations. A procurement conversation, forwarded document or offline recommendation may never enter the tracked journey. Missing does not mean influential, but it does mean the measured journey is incomplete.

Give each source a bounded job

Use platform and web reports to investigate delivery, response and operational changes. They can reveal that an advert stopped producing enquiries, that a landing page failed, or that a campaign attracted a different mix of visitors.

Use customer interviews to discover possible influences and buying situations. Ask buyers to reconstruct events before displaying your channel list. Otherwise the interview can become a recognition exercise in which the customer politely endorses every activity mentioned.

Use sales records to inspect opportunity quality, progression and commercial value. A form submission that becomes a poor-fit enquiry should not carry the same planning weight as a credible buying process.

These sources should challenge one another. Agreement strengthens a working explanation; disagreement identifies an uncertainty to investigate. Combining them does not automatically produce a precise causal estimate.

An illustrative budget decision

Imagine a specialist training company spending €12,000 on a research guide and €8,000 on search advertising. Its dashboard assigns 30 sales to search and two to the guide. The finance director proposes cancelling the next guide.

Those invented figures justify investigation, not immediate exoneration of content. The guide may create demand, it may serve buyers already convinced, or it may be largely unused. The dashboard alone cannot distinguish these possibilities.

The CMO first checks whether guide readers can be observed across the relevant buying period. Next, sales reviews whether the guide addresses an actual objection and whether prospects use it. The team also examines search terms to separate people seeking this company from people comparing unfamiliar providers.

Suppose the guide mainly repeats material buyers already know. Protecting its budget merely because attribution is imperfect would be as weak as cancelling it on last-click evidence alone. Measurement uncertainty does not make every activity valuable.

Ask a counterfactual question

For an expensive disputed activity, write the alternative explicitly: what would happen if we reduced it, delayed it, or delivered it to a different eligible group?

Controlled tests can help, but they are not effortless truth machines. In their 2015 advertising research, Randall Lewis and Justin Rao found that noisy sales outcomes made precise return estimates difficult even in large experiments.

For a small company, that means choosing a test whose possible answers would affect a real decision. A short, underpowered test cannot settle a question about a slow buying cycle simply because the spreadsheet has a significance column.

If a credible experiment is unavailable, state the weaker evidence you will use. A matched-period comparison, buyer interviews and sales-stage analysis may support a cautious judgement. Label the judgement rather than presenting it as measured incremental revenue.

Avoid double counting the same success

Different channels can claim the same order under their own rules. Adding all their attributed revenue together can therefore exceed actual company revenue.

Reconcile campaign reporting to the commercial ledger before discussing total return. Keep the ledger as the account of sales, and channel attribution as a set of explanatory views. Differences should be visible rather than silently forced to match.

Also distinguish attribution changes from business changes. A new model can move credit between channels without creating a single additional customer. Annotate such changes so the board does not celebrate an accounting rearrangement as improved demand.

Make the budget judgement explicit

For each major activity, record its intended role, the evidence that it is performing that role, the largest unresolved uncertainty and the next spending decision. This can fit on one page without pretending that every role has the same measurement standard.

Protect an activity when its strategic role and evidence justify doing so, with a review date and a bounded commitment. Reduce it when the evidence is weak relative to the alternatives, while acknowledging what the organisation may stop learning or creating.

The CMO’s responsibility is to make a defensible allocation under uncertainty. A clean attribution chart helps operate the plan. It cannot relieve leadership of deciding which explanation is credible and what evidence would change its mind.