Marketing attribution: useful evidence without false certainty

Distinguish assigned credit from causation

A report may credit a search ad with a conversion because of the model’s rules and the observed path. That does not prove the person would not have bought without the ad. They may already know the brand, have received a recommendation or have spoken with a salesperson outside the tracked journey.

Google Analytics describes attribution models as methods for assigning credit to touchpoints. Keep that definition visible when interpreting a report. The model answers a specific measurement question. A decision about incremental business value is broader and may need experiments, customer context and an understanding of what is missing from the observed data.

Map the information you cannot see

List the parts of the buying process that sit outside your reporting. Offline conversations, shared devices, later visits and incomplete tracking can all complicate the picture. The exact gaps depend on implementation, consent choices and the customer journey. Avoid assuming that every unassigned conversion belongs to the most recent campaign.

Document what the report includes and how it handles unavailable information. Google Analytics also documents modeled key events, so reported values may involve estimation rather than only directly observed events. That does not make the report useless. It means the team should understand its construction before treating a small difference between channels as decisive.

A hypothetical B2B buying journey

Consider a manager who first hears about a consultancy at an industry event, reads an article later and eventually searches for the company name before submitting a form. A channel report may emphasise the final search visit. The salesperson’s notes may reveal the event and the article as important context.

Neither source is complete. The manager may not remember every influence, and the analytics system may not connect every visit. A sensible review combines those perspectives and asks whether similar patterns recur. It would be premature to stop event activity solely because the last recorded digital touchpoint looks more efficient.

Use several views for different questions

Channel reporting can help monitor changes in observed acquisition. CRM information can show qualification and eventual commercial progress. Customer interviews can reveal language and perceived influences. Controlled experiments can help test particular causal questions when the business has enough volume and a suitable design.

Each method has a cost. More detailed tracking can require integration work and ongoing governance; experiments can take time and may be unsuitable for very small samples. Choose evidence proportionate to the decision. A minor creative adjustment and a large budget reallocation should not demand the same level of confidence, but neither should be justified with a number whose meaning nobody understands.

Use this attribution review checklist

Before moving budget, write the decision you are considering and the evidence that would change your mind. This prevents the meeting from becoming a search for a chart that supports an existing preference.

• Which attribution model and reporting period are being used?
• Are the compared channels measured on a consistent basis?
• How much time normally passes between enquiry and purchase?
• Are you reviewing raw enquiries or qualified commercial outcomes?
• Which touchpoints or customers may be missing from the observed path?
• Are modeled values or recent incomplete data involved?
• What do CRM records and customer conversations add?
• Could a limited experiment reduce uncertainty before a larger change?

Make a decision that matches the confidence

When evidence is incomplete, use a reversible adjustment and define a review point. State the working hypothesis, the expected signal and the conditions for changing course. This is more useful than presenting a precise allocation as if the model had discovered a universal truth about customer behaviour.

Keep an attribution note alongside commercial reporting, including changes to tracking and model settings. Compare patterns over suitable periods rather than reacting to every fluctuation. Strong measurement supports judgement; it does not remove the need for it. The useful question is whether the combined evidence gives the team a better basis for action than it had before.