Sessions
25,000
TA-P2-001 synthetic
Project 2 · Funnel & experiment
Synthetic funnel + simulated A/B readout
Decision: whether a review-prominence PDP badge should ship under this simulated evidence. Stakeholder: Product Operations / experiment reviewer. Not Agoda production data.
Synthetic clickstream
Simulated experiment
Grain: session / event / BKK PDP
Sessions 25,000 · Events 77,096
Overview synthetic
Events
77,096
TA-P2-002 synthetic
Purchases / CVR
722 / 2.89%
TA-P2-003 synthetic
Denominator: 25,000 sessions
Evidence check
PASS
artifact reconciliation
Funnel diagnosis synthetic
| Step | Count |
|---|---|
| search | 25,000 |
| view_listing | 19,948 |
| select_dates | 2,782 |
| begin_checkout | 1,302 |
| purchase | 722 |
Largest drop typically view_listing → select_dates in this synthetic funnel. Project 1 association motivates the badge hypothesis only — not causal proof.
Simulated A/B decision and action simulated
Decision
DO_NOT_SHIP
primary readout
No significant uplift detected on primary metric.
Control vs treatment CVR
3.64% vs 3.66%
Arms: 7,151 / 7,266 BKK PDP sessions
p-value · 95% CI
0.93612
Absolute CI [-0.59%, +0.64%] · relative uplift +0.69%
Power vs design
14,417 / 339,614
underpowered vs design
α=0.05 · 80% power · 5% relative MDE · ~239 days design estimate — not a launch plan
Bounce proxy guardrail
pass
Delta 0.4173 pp vs +2.0 pp · bounce proxy, not product bounce
AOV guardrail
pass
Delta 6.0105% vs -5.0% · simulated THB, not realised revenue
guardrails_ok
True
Conjunction of both checks; unavailable never passes
Device segments (exploratory · multiple-testing risk)
Device splits are exploratory and carry multiple-testing risk; they are not separately powered and cannot overrule the primary decision.
| Device | Control | Treatment | Rel. uplift | p-value |
|---|---|---|---|---|
| app | 4.54% | 6.23% | +37.19% | 0.065933 |
| mobile | 3.00% | 2.50% | -16.65% | 0.17594 |
| desktop | 4.33% | 4.27% | -1.34% | 0.926471 |
Segments cannot overrule DO_NOT_SHIP.
Caveats
- All funnel counts are synthetic; the A/B readout is simulated.
- Not Agoda production, not a live marketplace result, not realised revenue, not causal proof from Project 1.
- Simulation (14,417 PDP sessions) is underpowered versus design (339,614).
- Bounce metric is a proxy; AOV uses simulated THB amounts.
- Offline artifact: no CDN and no remote network requests.