产品机会验证

HarborMug return discovery decision

Recommendation: pivot the eligibility-preview concept and retest before committing delivery resources. Return uncertainty warrants further discovery. The current solution fails the supplied demand gate, even though comprehension and task speed meet their thresholds. Preserve merchant approval, make the request and review stages explicit, and investigate why people decline to submit.

All supplied interviews, journey observations, merchant data and experiment records are synthetic examples; this report evaluates those records and does not claim real customer testing or business improvement.

Outcome and decision boundary

Improve the return experience for US first-time mug shoppers and unused-return customers. Use one primary outcome: the share of return-policy visitors who need to contact support to understand eligibility. B1 records 12 contacts among 20 policy visitors (60%), but does not isolate eligibility-related contacts. Treat 60% as a broad support-contact baseline proxy, then classify contact reasons before claiming movement in the primary outcome.

Proposed target: reduce eligibility-clarification contacts by at least 20 percentage points from a verified baseline within four weeks of an approved limited pilot. This is a planning target, not a supplied or achieved result. Track request submission, approval misunderstanding, unresolved requests and total staff handling time as guardrails. Do not lower contacts by concealing support or obstructing returns. Routine merchant review after submission is separate from a clarification contact.

Evidence and problem validation

Source	What the supplied record shows	Interpretation and limit
I1	Checked policy twice; contacted Mia; 6 minutes	Eligibility uncertainty produces repeated checking and support work.
I2	Kept unused mug boxed while waiting; 8 minutes	Waiting prevents progress; keeping the item boxed is an observable workaround in the record.
I5	Short eligibility answer helps, but refund confirmation still needed; 5 minutes	Eligibility information alone may not satisfy the need for closure.
I3	Dispatch confused with delivery; emailed support; 4 minutes	Separate arrival-information opportunity.
I4	Added a second mug and still paid shipping; cart arithmetic; 3 minutes	Separate landed-cost opportunity.
B1	12/20 contacted support; 8/20 revisited	60% contact rate and 40% repeat-visit rate support journey friction; the record does not link individual sessions to interviews or classify contact reasons.

The supplied problem gate requires at least 3/5 independent reports of return uncertainty, at least two observed workarounds and B1 contact rate of at least 40%. I1, I2 and I5 provide 3/5 reports; their recorded workarounds are contacting support, waiting with the item boxed, and requesting approval. B1 is 60%, above 40%. The gate passes within the supplied records, treating the five interview entries as independent as the gate assumes. Recruitment independence and independent observation of each workaround are not documented; confirm them during the next discovery cycle.

Return uncertainty appears in three interviews, versus one each for arrival and shipping. The return interviews contain 19 reported minutes total (6 + 8 + 5); these are customer friction indicators, not measured staff handling time or expected savings. Frequency outside this small sample, financial severity and willingness to submit remain uncertain. The 2/5 submission result is the strongest supplied behavioral warning against assuming demand.

Job to be done: “When I have an unused mug I want to return, I want to know whether I can request a return and what happens next, so I can act without mistaking an eligibility answer for a refund promise.”

Opportunity solution tree

Root outcome: fewer eligibility-clarification contacts, subject to successful requests and safe merchant review. The adjacent branches preserve alternatives before convergence; they are lower priority because their evidence is thinner and their connection to the return outcome is indirect.

Opportunity: unmet need	Evidence	Candidate solutions	Experiments
O1. Know whether I can request a return and obtain a clear next step	I1, I2, I5; B1 corroborates general friction	S1: eligibility preview followed by request and pending-review receipt. S2: concise policy plus a direct human-review request without a separate preview.	E1: moderated task and teach-back for S1, then observe submission; apply the original solution gate. E2: fresh concierge cohort comparing reasons for submitting or declining S1/S2; use 3/5 submission as a directional gate per concept, not a statistical superiority claim.
O2. Know when my order will arrive	I3	Separate dispatch and delivery wording; explain the two stages in order help	E3: comprehension task asking shoppers to distinguish the two stages; proposed pass 4/5 correct. Defer delivery work until repeated pain is found in at least three independent relevant interviews.
O3. Know the total spend needed for free shipping	I4	Cart subtotal and amount remaining to the threshold	E4: cart arithmetic task using two mugs; proposed pass 4/5 identify $48 subtotal, $2 shortfall and $5 shipping. First confirm repeated need in relevant interviews.

O1 is the discovery priority. O2 and O3 remain documented alternatives, not approved build scope. Dispatch is within two business days and delivery is listed as 3–5 business days; the record does not specify the delivery interval's starting point, so clarify that before publishing a combined arrival promise. Two $24 mugs total $48, below the $50 free-shipping threshold; they do not qualify for free shipping.

Experiment observations versus thresholds

Gate	Required	Supplied observation	Decision
Approval comprehension	At least 4/5 distinguish eligibility from refund approval	4/5 (80%); one misunderstood	Pass exactly; retain an explicit approval boundary and investigate the error
Task speed	At least 4/5 finish within two minutes	4/5 (80%) under 120 seconds	Pass exactly
Concierge demand	At least 3/5 submit a request	2/5 (40%) versus 60% required	Fail by one submission, or 20 percentage points

All three solution requirements must pass; the combined gate fails. Task durations are 70, 95, 105, 115 and 180 seconds: mean 113 seconds, median 105 seconds, range 70–180 seconds. The aggregate record does not identify whether the slow participant also misunderstood approval or declined submission. Do not invent that connection. It also does not explain non-submission or demonstrate that submitted requests were eligible.

This supports a pivot, not a rollout and not abandoning the problem. The problem signal passes while the proposed behavior falls short. No retention, support reduction or financial lift has been established.

Prioritized assumptions and risks

Scores use the skill's mapper: risk × (1 − certainty). Inputs below are supplied judgments, not probabilities measured by the experiment. Keep them unchanged for traceability; T1 should inform the next scoring review.

Rank	Assumption / category	Risk	Certainty	Priority	Next validation and decision criterion
1	Users submit eligible requests after reading a preview — desirability	0.90	0.20	0.72	Revised concierge test: at least 3/5 submit without prompting. Independently review eligibility for each request; log all declines and reasons. T1 only proves 2 submissions, not eligibility.
2	Saved handling time justifies operations cost — viability	0.80	0.30	0.56	Time all handling and rework in baseline and concierge cases, calculate full cost; require positive conservative net benefit before delivery commitment.
3	Customers distinguish eligibility from refund approval — usability	0.90	0.40	0.54	Teach-back before submission: at least 4/5 distinguish the two without moderator help, alongside at least 4/5 under two minutes. Any design that grants or promises refunds from a preview is rejected.
4	Mia can route requests without automatic refunds — feasibility	0.70	0.60	0.28	Operations walkthrough: all scripted eligible, uncertain, outside-window, duplicate and routing-failure cases preserve pending review; zero automatic refunds, silent losses or duplicate requests.

The feasibility score is lower, but its control boundary must pass before any pilot. Mia owns routine review; Alex handles exceptions and routing failures. Neither owner has been contacted or assigned live work by this exercise.

Viability cannot yet pass. Use:

weekly net benefit = N × (baseline minutes − new minutes) ÷ 60 × loaded hourly cost − weekly tool cost − incremental error cost

Count review, escalation, duplicate handling and rework in the new minutes. Break-even minutes saved per request are 60 × (tool cost + incremental error cost) ÷ (N × loaded hourly cost), for positive N and hourly cost. Request volume, measured staff time, loaded cost and ongoing costs are missing. Do not substitute customer-reported interview minutes for staff savings. The merchant's weekly refunds/gross are 5% in both periods ($120/$2,400 and $144/$2,880); these figures do not establish return-request frequency, margin impact or savings from this concept.

Completed prototype draft

The complete screen and operations specification is in prototype.md. No customer-copy language is specified in the inputs, so this draft uses English. The proposed pivot moves from a short eligibility answer to an explicit request-and-review journey; the actual prior screen design is not supplied.

Screen 1 — Check a return

“Returns are available for unused items within 30 days. We can help you check whether you can request a return. HarborMug must review your request before any refund is approved.”

Fields: order reference; delivery date; item; “Is the item unused?” with Yes / No / Not sure. Button: Check return options. Provide Ask support throughout.

Screen 2 — Preview for the supplied example

“Your item appears to meet the return-request conditions. Your request is 14 days after delivery, within the 30-day window. You told us the item is unused. This is not refund approval. Send a request for HarborMug to review.”

Show editable answers and Send return request. Unopened packaging supports the example's screening result but must not replace confirmation that the item is unused. An uncertain or apparently ineligible result offers human review and does not promise acceptance.

Screen 3 — Request received

“Your return request has been sent for review. Your refund has not been approved. HarborMug will confirm the next steps after reviewing your request. Please wait for return instructions before sending the item.”

Show this receipt only after the request is durably recorded. Do not invent a response deadline, return address, label, postage policy, refund amount or payment action. If delivery fails: “We couldn't send your request. Please try again or ask support.”

Operations handoff: create one pending request for Mia, with order reference, item, delivery date, stated condition and submission status. Mia checks the order and policy and records the review outcome. Missing or disputed information, policy exceptions and routing failures go to Alex. No preview, submission or receipt triggers a refund. Send instructions only after the merchant has decided the next step.

Proposed ten-day discovery sprint

This is a future test plan; no new participants have been recruited and no operational changes have been made.

Days	Owner role	Work and output
1–2	Discovery lead + Mia	Verify problem-source independence and contact reasons; establish eligibility-contact baseline and staff-time measurement. Review interview behavior, not only preferences.
3–4	Designer + operations	Build the draft screens in prototype.md; rehearse eligibility, uncertainty, duplicate and routing-failure states with Mia/Alex roles. Lock task script and thresholds before testing.
5–7	Researcher + Mia role	Recruit five fresh relevant participants for E1 with a real return intent appropriate to the test. Observe unaided comprehension, duration and submission separately. Ask non-submitters what they chose to do and why. Run E2 in a separate fresh five-person cohort if feasible; otherwise defer the comparison and retain the same concept gates. Measure total handling time in a controlled concierge rehearsal.
8–9	Discovery lead + operations	Review evidence daily during testing; distinguish ineligible cases from demand failures. Reconcile missing, duplicate and failed requests; compute conservative cost scenarios. Do not change thresholds mid-test.
10	Operations decision owner	Review gates and costs; choose continue, pivot or stop and document the evidence.

For each task, record anonymous participant code, relevant return intent, scenario eligibility, prototype version, unprompted comprehension answer, duration, submission event, refusal reason, review disposition and handling minutes. Retain separate denominators for each cohort. Keep recruitment data outside the evidence table.

Continue to a small supervised pilot only if all original solution gates pass in the revised cohort, the routing rehearsal has no control failures, and measured costs support positive conservative benefit. This would authorize further learning, not establish a population effect.

Pivot again if the return problem remains supported but a solution gate fails and a specific observed barrier suggests a testable change. Test the direct human-review request if previewing adds uncertainty or friction.

Stop this solution if a second revised cohort still has fewer than 3/5 submissions without a credible fix, or measured economics remain nonpositive after a realistic lower-cost alternative. Reassess the opportunity if independent interviews and behavior no longer support the problem gate. These stop rules are proposed for the next cycle, not retroactive thresholds.

The gaps that change the decision are the reason for three non-submissions, unverified request eligibility, unclassified contact reasons, unverified interview independence and missing operational cost measurements. Until resolved, approve discovery work only.

Supporting deliverables

• prototype.md: complete customer screen copy, branches and operations acceptance checks.
• assumption-priorities.csv: skill-generated ranking and generic suggested test types; the tailored tests above govern this return workflow.
• calculations.json: rates, durations, ranking arithmetic, example calculations and cost model.
• evidence/brief.csv, evidence/evidence.json, evidence/merchant.json: unchanged supplied evidence for traceability.