One tool rarely sees the entire chain.
Web analytics is good at showing behavior on your own site. The tracker links the source and click ID. Affiliate platform knows confirmed events and payments. The ad office is an expense. If you try to choose the “main truth” in one place, some of the context is lost.
What is needed is not maximum complexity, but a clear distribution of roles.
Minimum stack
| Tool. | The main task |
|---|---|
| Web analytics | login pages, site events, devices |
| Tracker/SubID | source, campaign, creative, click ID |
| Partner platform | registration, FTD, confirmed events, revenue |
| The advertising office | expense, impressions, clicks |
| Table/BI | Consolidate costs and confirmed results |
First, agree on the names.
If in one report the campaign is called KZ_01elsewhere kz-mainin the third 124567, the gluing turns into handmade. A good UTM/SubID scheme should be clear before launch.
Discrepancies are a normal part of the system
Different platforms can count clicks differently, use different timezone and attribution. It is important not to achieve absolute coincidence at any cost, but to understand the reason for the discrepancy and control its stability.
If yesterday the difference was 3%, and today 30%, this is a signal to check.
Do not collect data “just in case”
Each parameter must answer the question on which the decision is made. If the team never watches utm_termMaybe it doesn't need to be complicated. But. source, campaign, creative, GEO The device is often very useful.
Reduce the cost with a confirmed result
A Raw FTD may look good, but if some of the events are not confirmed, the actual cost is higher. Therefore, it is better to build a financial report on the event for which the calculation actually takes place.
Retention and cohort – a separate layer
Long metrics should not be confused with instant metrics. They need a date of engagement and the same observation window. Otherwise, older users will always have more accumulated results.
Conclusion
The analytics stack should help answer “which segment created the result and why,” rather than simply assembling the maximum of events. Related material: UTM and SubID, postback S2S and attribution.
