Project: review Meridian pickup survey findings

Last updated: 7 Oct 202618 min read

project

Review a fictional Meridian Parcel pickup-window survey before its product memo is released. The target is recent pickup users, but the invitation frame contains only 240 unique contacts who enabled email notices. Build a survey contract, pilot the instrument, reconcile the response ledger, and draft a claim with its exact scope. The model may assist with drafts and checks; the form engine enforces branches and deterministic code calculates counts.

Inspect wording, branches, and pilot results

Replace the leading draft 'How frustrating is our confusing pickup window?' with a neutral clarity item about the most recent pickup. Keep ability to plan as a separate construct. Q6 routes customers without a recent pickup away from Q7; cannot-recall answers do not enter a planning follow-up. Five pilot participants review Q7-v1, and two interpret 'pickup window' as locker collection rather than the pre-confirmation time range. The team clarifies Q7-v2, repairs the branch, retests it, and freezes the instrument before the 240 invitations are sent.

Reconcile the evidence

Meridian receives 84 submitted forms, including seven partial and 77 complete. Twelve complete respondents correctly skip Q7 because they had no recent pickup. The other 65 are eligible complete respondents, and 39 select a narrower window: 39/65 is 60%. The simple submission fraction is 84/240, or 35%; it does not prove absence of nonresponse bias. One open answer contains a depot worker's name, which is removed before coding. A preference for flexible windows stays as a contrary case. The memo cannot say that 60% of customers want a narrow window.

Frame: 240 unique email-notice contacts
Pilot: 5 participants; 2 misread Q7-v1; Q7-v2 retested
Submitted: 84; partial: 7; complete: 77
Q7 branch-ineligible: 12; eligible complete: 65
Narrower choice: 39 / 65 = 60% of eligible complete respondents
Simple submission fraction: 84 / 240 = 35%

Performance and review cost

Deduplicate N invitations in O(N) expected time, validate a V-question and E-branch form in O(V+E), and reconcile R responses in O(R). The release packet should hold the invitation frame, instrument version, pilot corrections, branch dispositions, exact Q7 wording, counts, private-text review, and the approved claim. A missing denominator or unreviewed version change blocks release even if the generated summary reads well.

Common Mistakes

Do not treat email-notice contacts as all pickup users.
Do not field leading wording without respondent review.
Do not count branch skips as missing answers.
Do not publish an identifying free-text quote or an unsupported population claim.

Related lessons

prompt engineering

survey research