Interview instruments — Part 3

Six remote sessions (30–40 min) reconstructing one real save→decision episode from the last 30 days, then a think-aloud usability pass on this prototype. Oversampling the two highest-lift engine segments: new/low-trust (23.7% of corpus, 2.81× trust lift) and fit-uncertain (24.5%, 3.99× fit lift).

Recruitment groups

G1Purchased from wishlist ≤30d1 participant

G2Revisited but didn't buy2 participants

G3Saved, never revisited2 participants

G4Bagged but didn't purchase1 participant

Episode reconstruction probes

What were you doing right before saving? What did you expect to happen next?
If asked at that moment “what are you saving this for?” — your honest chip?
What changed by your next visit — and could the app tell you?
What information did you look for outside the app before deciding?
What exactly stopped the purchase that day — and what would have made buying feel safe?
Workarounds: screenshots to friends, notes app, offline checks, waiting.

Usability tasks on the prototype

#	Task	Watch for
T1	Save a kurti for an occasion; continue browsing normally	Prompt must never block; skip works; answer ≤5s or skipped
T2	Two days later: find the item you saved	Time-to-find; do change badges register?
T3	Your size sold out — get to a state where you'd still buy	Which recovery path: swap / alert / similar
T4	Decide it's the one — bag it and start checkout	Sheet usage before committing; hesitation points

Key cross-check

Does the declared save-reason (chip chosen in T1) match the stated blocker at the decision point? This validates or falsifies the routing premise of the MVP. Mismatches are reported honestly and feed the risk register (“self-report misprediction”), not hidden.

Full moderator scripts incl. screener questions and synthesis template live in the project repo (research/interview-instruments.md).