Inspect paid-search mobile traffic on the short landing page
Its traffic mix grew while its conversion rate declined. It is the largest observed contributor, not a proven cause.
Evidence: 71.43% of segment-rate underperformanceParas-built system · worked example
Where are customers getting stuck?
Abandonment Intelligence · ranked investigation and action layer · course practice data
OUTCOME: REVIEW
Top three actions now
Audit the v1-to-v2 tagging transition before changing routing
The tag version changes between the two periods. Validate classification continuity with the metric owner.
Gate: review required before actionAdd this segment and tag transition to the regression set
Preserve the correction so the next run checks complete weeks, exclusions, benchmark tolerance, and context version.
Owner: analytics system + metric ownerCurrent conversion
6.9%
Eligible, complete week
Movement
−1.5pp
8.4% in prior week
Rate underperformance
140
Signups below prior segment rates
Benchmark delta
0.0pp
Independent tie-out passed
Where the movement concentrates
Eligible sessions10,000 current week
100%
Expected signupsat prior segment rates
830
Actual signupscurrent complete week
690
140-signup segment-rate gap · paid search contributes 100
Ranked investigation list
| Finding | Affected population | Evidence | Next owner/action | State |
|---|---|---|---|---|
| Paid-search short-page mix and rate moved together | Mobile paid-search sessions | Traffic share rose; rate fell to 4.8% | Growth + metric owner: inspect tag continuity and page routing | REVIEW |
| Tag version changed inside the comparison | Paid-search classification | v1 prior week; v2 current week | Analytics: reconcile mapping before a causal claim | REVIEW |
| Benchmark and calculation agree | Weekly aggregate | 0.0pp difference for both complete weeks | System: preserve as an automatic validation | PASSED |
How I designed it
A rate is not a finding.
The output had to move from a broad funnel observation to a specific, reviewable action. Each specialist investigated one breakpoint. The synthesis layer ranked only findings that could survive these six questions.
PopulationWho or what is affected?
Broken flowWhich rule, route, or step needs inspection?
DurabilityIs it persistent or a one-period fluctuation?
EvidenceWhat source and validation support it?
OwnershipWho investigates, approves, and closes it?
ConfidenceWhat can we say, and what remains unproven?