A funnel is needed to find the place of loss.
The overall “campaign conversion” metric is almost always too rough. It mixes several different processes: reaction to advertising, landing, registration, payment route, confirmation and further user behavior.
Therefore, it is useful to keep the chain of events separate.
Basic diagram
| Phase | Question |
|---|---|
| Impression → Click | Does the message attract the right audience? |
| Click → Registration | Landing confirms the expectation? |
| Registration → FTD | Does the user go through the payment process? |
| Raw FTD → Confirmed FTD | Does the action meet the offer criteria? |
| FTD → Retention/LTV | Does the audience retain value further? |
Each transition has its own reasons for the drop. There is no point in remaking creativity if the problem arises after registration.
Start with the earliest change.
If yesterday all the stages were stable, and today the FTD fell, find the first coefficient in the chain that has changed. It narrows the search.
For example, CTR and CR registration are the same, but registration → FTD has declined. Then the first thing to check is payment methods, GEO, technical errors and product changes, not the banner.
Don't mix the segments
The middle funnel may look normal when one source grows and the other falls. The minimum useful incision is the source × GEO × device. Deeper detail is needed only when the volume allows us to draw conclusions.
How to link costs to the funnel
For paid traffic, add to the stages cost: CPC, cost per registration, cost per raw FTD, cost per confirmed FTD. The last metric usually better reflects the real price of the result, if there is a noticeable difference between raw and confirmed.
With RevShare, an NGR or other available financial metric is added to this. With CPA – confirmed actions and rate. With Hybrid, both parts.
LTV is taking time.
You can't honestly compare the LTV of yesterday's cohort to the audience that came in a few months ago. Use the same observation window. If there is little data, it is better to limit yourself to earlier metrics and not to depict the accuracy of the long forecast.
Diagnostic order
- Find the first changed stage.
- Break it down by source/GEO/device.
- Check for changes in Creative and Landing.
- Check integration and attribution.
- Verify the terms of the offer.
- Only then start the change.
Conclusion
A strong funnel is not a single number, but a sequence of causes. Use the material for practice. Where more often is lost click → registration → FTD and cohort analysis.
