What attribution can and cannot tell you

Attribution assigns credit for an outcome to interactions that preceded it. It helps describe observed journeys, but it does not provide a complete record of every influence or prove that a campaign caused an additional sale.

Agree what is being measured

Define the meaningful action: a purchase, accepted enquiry, qualified opportunity or another event. Make sure the implementation distinguishes success from a button click or an attempted submission.

Connect the action to reliable business records where possible. A duplicate event, test order or spam enquiry can distort the report before any attribution model is applied.

Separate observations from assigned credit

Observed information may include a landing page, campaign parameter, referrer or an interaction recorded by an advertising platform. Consent choices, device changes, blocked storage and system boundaries can limit those observations.

A model applies rules or statistical methods to the information available. First-touch and last-touch are useful concepts, but not every platform offers every model. Google's attribution overview describes the models currently available in its Analytics attribution reports.

Keep the original observations alongside any inferred channel label where useful. A missing campaign parameter is not proof that a customer arrived directly, and a stored first touch is only the first touch within the coverage that system retained.

Understand why reports disagree

Platforms may use different attribution windows, eligible interactions, identity methods, time zones and modelling. Several systems can credit the same purchase without their totals being suitable for addition.

Use those reports for the decisions they can support. A platform can help optimise delivery within its own measurement view. Order or CRM records can establish realised business outcomes. Neither view is automatically a complete explanation of the other.

Keep acquisition context useful and proportionate

For an enquiry, relevant context may include the campaign, landing page, referring site and the page or CTA that opened the form. Capture it only when available and permitted, with clear handling for storage denial and missing data.

Keep customer answers and personal information out of general analytics events. An optional attribution problem should not prevent a valid enquiry from reaching the business. Our first-party data guide explains the ownership and purpose questions behind collection.

Use attribution alongside other evidence

Review channel results against qualification, sales progression, margin and customer retention where those measures exist. Avoid drawing conclusions from tiny groups or comparing outcomes before the relevant sales cycle has elapsed.

When the decision concerns incremental impact, consider an appropriately designed experiment or another suitable causal method. Attribution redistributes credit; that is a different question from estimating what would have happened without the activity.

Make the limitations understandable

Document the event definition, reporting period, model, window and important gaps. Change these deliberately and explain when a trend may reflect implementation changes rather than customer behaviour.

A useful measurement system supports better decisions without claiming certainty it cannot provide. The aim is to connect observed acquisition with real outcomes and keep the uncertainty visible enough to act responsibly.