Required CSV Format (One row per user)
Upload your experiment CSV to immediately generate the leaderboard and Shapley split:
- test_id: Experiment identifier (multiple tests in one file are analyzed separately).
- user_id: Unique user identifier.
- variant: Name of the variant (e.g., control, variant_a, variant_b).
- is_success: 1 if converted/succeeded, 0 if not (or true/false).
- revenue: Total revenue generated by the user (0 if not converted).
Statistical Methodology
CR = Conversions / Users · Avg Rev = Revenue / Conversions · RPU = CR × Avg Rev.
Noise = √(p̂(1−p̂)(1/nA + 1/nB)), Z = (CR_B − CR_A) / noise, P(beats control) = Φ(Z).
Shapley Attribution: CR impact = ΔCR × mean(Avg Rev), AOV impact = ΔAvg Rev × mean(CR); they sum exactly to ΔRPU.
Frequently Asked Questions (FAQ)
What is the advantage of ranking by RPU instead of just Conversion Rate?
A variant can increase conversion rate while cannibalizing average order value (e.g. heavy discounting), resulting in net negative revenue. Ranking by RPU accounts for both volume and monetization simultaneously.
Are my experiment data files stored on your servers?
No. Trustometer processes all CSV calculations in-memory in real-time and discards them immediately. Zero data retention.