01 — Set the table
We name the eligible population, exact entry, intended outcome and observation window. Product, analytics and engineering each describe the journey so differences surface early.
02 — Check the deck
Events are reconciled with the tracking plan and observed app behaviour. We look for duplicates, missing properties, late timestamps, identity breaks, retries and platform divergence.
03 — Read the hand
Only then do we calculate step conversion, elapsed time, exits and useful cohort cuts. Each material finding is tested against at least one plausible alternative explanation.
04 — Call the next move
The workshop sorts work into tracking repairs, product changes, research questions and experiment candidates. Owners and dependencies are recorded; weak evidence is labelled rather than polished away.
Access without unnecessary exposure
Read-only access is preferred. We use client-managed accounts, avoid exporting raw identifiers and remove working extracts at the end of the agreed retention period. If access cannot leave your environment, we adapt the process around queries your analyst executes.