Example: three drops inside one funnel
This is a hypothetical situation from a work note, not a real product or client. One week records 100,000 entry visits, 68,000 product-page views, 18,000 checkout starts, and 9,000 completed payments. Step conversion is 68%, 26.5%, and 50%.
The total conversion rate may point attention towards checkout. Yet entry to product page, product page to checkout, and checkout to payment respond to different conditions and need separate comparisons.
| Step | People | Step conversion |
|---|---|---|
| Entry visit | 100,000 | — |
| Product-page view | 68,000 | 68% |
| Checkout start | 18,000 | 26.5% |
| Completed payment | 9,000 | 50% |
Check 1: the traffic mix changed
Break entry traffic down by source, customer status, device, and region. A larger share from a low-intent source can lower the total rate while every source remains stable. That is a composition change, not direct evidence of a page problem.
Keep the size of each group beside its within-group rate. Rates without weights hide the effect of traffic composition on the total.
Check 2: the product path changed
Review releases, entry placement, routing rules, and error rates. People may bypass an old step or stop at a new prompt or login requirement. Put release time and funnel movement on the same timeline.
- Check whether the movement is confined to the new version.
- Look for traffic entering a new branch.
- Compare errors, timeouts, and empty states.
Check 3: the measurement changed
An unchanged event name does not guarantee an unchanged firing rule. A front-end rebuild, deduplication rule, cross-device sign-in, or default field can alter the count. Sample raw events and compare event time, identity, and page state with the tracking dictionary.
Check 4: behavior changed
After composition, path, and measurement issues are excluded, it becomes reasonable to examine price, content, demand, or competition. Search terms, product mix, and return intervals can guide the next enquiry or experiment.