Which question is the report meant to answer?
Three questions that sound alike are in fact different: which channel first brought a user to the site, which channel brought a particular session, and which channel should receive credit for a key event such as an enquiry. Separating these three questions is the first step in marketing analytics. In GA4 these questions are answered by reports with different scopes, so the same channel label in two reports does not mean the numbers can be compared or added together. Before comparing channels, make sure each enquiry is counted once: see GA4 lead tracking.
Whenever you take figures from a report, record the dimension, metric, date range and settings that were used. This seems tedious until someone asks why last month’s figures differ from those in the presentation and nobody remembers which report they came from.
Decide which view fits the business question before exporting figures into a shared dashboard. If the question is how new customers find the company, a first-user view makes sense; if it is which campaigns bring visits right now, a session view does. Mixing the two on one screen without labels guarantees confusion.
Are your events and campaign tags reliable?
Decide together with the sales or product team what counts as a meaningful conversion: a submitted form, a qualified enquiry, a paid order. Then test how that event is actually collected. Does it fire once, does it fire on the right page, and is a repeated submission counted twice? Write the definition down and share it with everyone who reads the report, so that “conversion” means the same thing in every meeting.
Use consistent campaign names and check where links really lead after any redirects; a redirect that drops campaign parameters will quietly move traffic into the wrong channel. Document the known limitations of tracking where they matter: the effect of consent choices, the gaps when a person switches devices and the traffic your analytics cannot see.
Keep what customers tell you themselves (“a colleague recommended you”, “I saw your video”) separate from instrumented events. These are different kinds of evidence, and the report should show them side by side. Merging them into one total produces a figure that looks precise but cannot be explained.
How do you handle discrepancies and make a decision?
The advertising platform, the analytics tool and the CRM will almost always show different numbers for the same campaign. Compare them using consistent definitions and explain why they differ before deciding that one of them is wrong. If the CRM and analytics are not connected at all, start with system integration.
Check the usual causes first: time zones, attribution windows, refunds and cancellations, and the moment an event is recorded, whether at the click, the visit or the payment. Often all three sources are correct, each by its own rules.
Use the reconciled data together with lead quality and the cost of serving customers from each channel; a channel that brings many cheap enquiries that never become orders may be worth less than it seems. And keep the limits of any attribution model in mind. A model shares out credit according to its own assumptions. It describes how credit is assigned, but on its own it cannot show how many sales a channel added that would not have happened without it. Write down which limitations applied when the decision was made, so it can be revisited when better data becomes available.
Checklist: attribution reports
- Record the scope of each report and the attribution settings used.
- Check that the intended event is collected correctly and protected against duplicates.
- Reconcile the definitions used by the advertising platform, analytics and the CRM.
- Review budget decisions together with lead quality and the known limitations of the data.
Example: search first, video later
A person first finds a service through search and a week later returns to the site from a link in a video. A first-user report credits search, while a session report shows the return visit under video. Both reports are correct; they simply answer different questions.
The team labels both reports clearly and checks in the CRM whether the enquiry was qualified before discussing how much each channel contributed.
