Product Skills: Product Routing and Discovery Tools

HarborMug next discovery cycle

Decision: pivot the discovery focus; continue research and defer a returns build. Investigate whether care and return-policy uncertainty affects purchase decisions before committing to a returns solution. The supplied evidence supports care, arrival and total-cost questions, but contains no return-journey interviews or evidence of return friction. A returns redesign would therefore outrun the evidence.

All supplied business data are synthetic examples. No new customer research or operational changes have been performed.

Outcome and decision boundary

Retain the locked outcome: raise US session-to-order conversion from 2.5% to 3.0% by December 15, 2026, with refunds / gross revenue ≤5%. That is a 0.5 percentage-point, or 20% relative, improvement. Both supplied weeks have 2.5% conversion; the higher order count reflects higher traffic, not better conversion. At 4,800 sessions, 3.0% would mean 144 orders, 24 above the supplied current week, assuming the same traffic denominator. These are targets, not forecasts.

Refunds / gross is 5% in both weeks, already at the ceiling. The data do not identify geography, so confirm the session and order figures represent the US population before treating them as the goal baseline. The weekly refund ratio is an operating guardrail, not a cohort return rate; refund timing can lag sales.

The supplied return example is unopened and requested 14 calendar days after delivery, within the stated 30-day unused-return window. This supports eligibility review only: merchant approval is still required. It does not establish customer confusion, refund approval, processing time or a system defect.

Evidence and contrary explanations

Source	What it supports	Contrary evidence or limit	Consequence
P1–P4, four US mug shoppers	All looked for dishwasher compatibility, compared total price and struggled to estimate arrival; calendar checking is an observed workaround	Summaries repeat identical wording; severity, purchase consequences and verbatim accounts are absent	Treat as directional repeated signals; obtain distinct behavioral examples
T2, moderated comprehension, n=4	Two initially assumed dishwasher safety	Does not establish whether correction changes product fit, abandonment or returns	Test desirability separately from comprehension
T1, concept comparison, n=4	Three accepted $29 delivered when shown early	One did not; no randomized control or real purchase outcome	Weakens the claim that free shipping is necessary; does not prove price disclosure increases conversion
Discovery log	Four weekly, outcome-linked interviews; two assumptions marked invalidated	A recorded label is not stronger than its underlying evidence	“Care is understood” is contradicted; “care clarity matters to choice” remains unresolved
Return example and policy	A concrete unused-return scenario exists	No return-experience research, ticket volume, effort or abandonment evidence	Investigate the opportunity; do not assert it is validated

A plausible contrary hypothesis is that shoppers understand care once told but reject hand-wash-only mugs because the product does not fit their routine. Clearer copy could then reduce purchases while preventing unsuitable orders. Arrival or delivered price might instead dominate choice. Return reassurance might also create an incorrect expectation of automatic refunds. Capture these outcomes explicitly rather than counting favorable reactions alone.

Riskiest assumption

Use risk × (1 − certainty) to prioritize learning, following the discovery assumption-mapping framework. This ranks uncertainty about a consequential belief; it is not an estimate of revenue impact.

Rank	Supplied assumption	Risk	Certainty	Priority
1	Care clarity matters	0.90	0.20	0.72
2	Arrival card is understood	0.80	0.40	0.48
3	Margin supports disclosure	0.70	0.70	0.21
4	Shipping logic is feasible	0.50	0.80	0.10

Operational hypothesis: uncertainty about care prevents target shoppers from judging product fit, and accurate disclosure lets them make an informed choice without assuming a refund is guaranteed. T2 identifies misunderstanding, not demand for a solution. The proposed returns hypothesis—“uncertainty about unused-return eligibility or approval obstructs purchase or return tasks”—is unscored because the supplied evidence does not justify a numerical certainty estimate.

The top ranking survives an illustrative adverse change of 0.1 in care certainty (score 0.63) combined with a 0.1 decrease in arrival certainty (score 0.56). Larger revisions could change the ranking; reassess after the cycle.

Tree and cadence review

The supplied tree passes the structural gate: one measurable outcome, three needs, two solutions per targeted opportunity, tests on every solution, and no orphan solutions. It compares arrival disclosure with a destination panel, a care card with an image-side comparison, and early total with a calculator. Structural soundness does not validate any branch. Its generic comprehension criteria also do not test whether care clarity matters to decisions.

Keep the three existing branches and focus this cycle on care. Keep returns as a research question outside the validated tree until evidence supports a customer need. Do not label a proposed returns branch as observed fact. The preserved ost.json has not been expanded with invented research.

Cadence scores 100/100, HEALTHY: four distinct participants across four observed weeks, a four-week streak, full outcome linkage and two resolved tests. The tracker’s next action is to book next week’s touchpoint. Healthy cadence permits continued learning; it is not solution validation. discovery_log.json preserves completed activity only; planned sessions must not be recorded as completed interviews.

Next-cycle brief: ten working days

Accountability: Mia coordinates support evidence and research scheduling; Alex owns the decision and any policy approval. Assign a product/research facilitator and prototype builder on day 1. These are proposed responsibilities, not completed assignments.

• Days 1–2: confirm US metric denominators, approve the scoring rubric and draft copy, review available return-support evidence without treating absent records as zero incidents. Book at least one touchpoint in each cycle week. Freeze the hypotheses and thresholds below before exposure.
• Days 3–4: prepare both existing care concepts and the separate return task. Recruit five new US mug shoppers who have recently considered an insulated mug, including at least two who normally rely on dishwashers. Recruit three separate recent unused-return requesters for return-journey interviews. Do not substitute prior P1–P4 participants for new evidence.
• Days 5–7: run individual moderated sessions. Start with past behavior and an unaided product assessment, then show both care concepts in alternating order. Do not mention the hypothesized pain before the unaided task. Keep price and arrival information identical between concepts. Run return interviews separately to avoid contaminating care findings.
• Days 8–9: review evidence daily, code failures and negative cases, compare concepts on first exposure, and separate prompted answers from unaided success. Recheck policy accuracy and the refund guardrail definition.
• Day 10: Alex reviews the decision table, counterevidence and missing evidence; record continue, pivot or stop, then update the log and tree with actual findings. Book the following weekly touchpoint.

This cycle uses small qualitative tests to reject weak concepts and improve the problem framing. It is not powered to estimate a conversion effect. Its behavioral tasks are simulated purchase choices, not actual orders.

Experiment A: does care clarity resolve a consequential uncertainty?

Unit and sample: one new eligible shopper; five completed sessions. Record both concepts per participant but analyze each concept separately. Alternate order 3/2 and report that imbalance. Within-person comparison is directional and affected by learning; first-exposure results must be shown separately.

Procedure: ask the shopper to assess whether the mug fits their actual routine and make a simulated buy/decline decision with a reason. Ask for a recent comparable purchase example. Show a care concept and ask them to explain how they would clean and use the mug, then repeat their decision and reason. Finally ask what they expect if they request a return. Do not reward purchase intent or explain the correct answers until scoring is complete.

Predeclared pass conditions, all required:

1. At least 3/5 independently identify care uncertainty before disclosure as a material obstacle to deciding fit, supported by a specific behavior or recent example; moderator-prompted agreement does not count.
2. For a concept to pass comprehension, at least 4/5 correctly state both “hand-wash only” and “not microwave safe” without help after viewing it.
3. At least 3/5 show a specific, accurately reasoned resolution of the initial uncertainty in their simulated choice after disclosure. An informed decline counts as resolution, but must be reported separately from a purchase choice.
4. 0/5 infer that opening/using the mug guarantees a refund, and the policy review finds no inaccurate promise.

Record denominators, first-exposure counts, corrections, informed declines and contrary examples. Do not add participants after seeing unfavorable results to reach a pass. Missing or ineligible sessions are replaced before analysis; fewer than five eligible completions means inconclusive. These are explicit qualitative gates, not statistical confidence or evidence of conversion lift.

Decision: if all conditions pass, continue with the qualifying concept to further validation; select by first-exposure errors and misconception severity rather than a one-person preference margin. If care matters but comprehension or safety fails, pivot the presentation and retest with new participants. If fewer than 3/5 show a consequential care obstacle, stop prioritizing care as the main conversion lever and investigate arrival/total-cost alternatives. Any false refund promise blocks advancement until corrected.

Experiment B: is there a returns opportunity worth adding?

Interview three separate recent unused-return requesters. Reconstruct the last journey: trigger, eligibility check, request, approval, next step and effort. Ask what actually happened and what they expected, then test this scenario: an unopened mug, delivered 14 days ago, with a return requested today. Ask whether it can be reviewed under the policy and whether a refund is guaranteed.

Problem gate: at least 2/3 independently recount the same eligibility/approval uncertainty plus a concrete consequence, such as repeat contact or delayed action. If met, propose a need-based branch (“Customers cannot determine unused-return eligibility and the approval step”), linked to the new evidence, with two candidates: an eligibility summary and a guided request checklist. Each must carry its own test before the tree is linted again.

Copy safety gate: 3/3 correctly identify the 30-day unused window and merchant approval, with 0/3 believing the draft guarantees a refund. A comprehension failure triggers copy revision even if the problem gate passes. If the problem gate is missed, stop the returns-solution commitment for now; the small sample does not establish that no broader returns problem exists. If recruitment cannot yield three eligible interviews, mark the returns decision inconclusive rather than replacing it with shopper opinions.

Interview prompts and scoring record

Use these prompts verbatim where practical: “Tell me about the last insulated mug you considered. What did you check before deciding?” “What would make this mug fit or not fit your routine?” “Show me where you would look for that information.” “How would you clean and use it?” “What would you expect after requesting a return?” For returners: “Walk me through your last unused-return request. Where did you hesitate, and what did you do next?”

Record participant code, segment, concept order, unaided obstacle, concrete example, comprehension before/after, simulated choice and reason, return expectation, assistance, contrary evidence and pass/fail against each criterion. Use the supplied blank worksheet; do not populate it with fictional new results.

Draft customer-facing prototype copy

No separate target language was supplied; these drafts use English. They are research materials, not published copy.

Care card

Will it fit your routine? This 500 ml stainless steel insulated mug is hand-wash only. It is not microwave safe. If you need a dishwasher-safe mug, consider whether hand-washing works for you before ordering.

Image-side care comparison

Cleaning: hand-wash only. Microwave use: not safe. Check these care requirements before you choose.

Unused-return summary for the separate return task

Need to return an unused mug? The unused-return window is 30 days. Refunds require merchant approval. Request a review of your return eligibility; a request does not guarantee a refund.

Mia and Alex must settle the policy’s window start point and request instructions before publishing a return flow. The supplied policy does not specify return shipping costs, refund timing or the handling of used goods; the drafts make no promises on those points.

Decision after this brief

Pivot now, with bounded continuation of discovery. Preserve the conversion outcome, test the top-ranked care assumption, and establish whether a returns opportunity exists. Stop treating free shipping as a prerequisite and stop short of approving a returns build. Neither the four-person evidence nor the cadence score justifies shipping or claiming sales impact.

A passing qualitative cycle permits further validation, not rollout. A later causal test must define the baseline population, minimum detectable effect, allocation, power, sample size, runtime and refund-maturation window before launch. No causal experiment is authorized or sized here. The bundled statistical sample-size calculator is unavailable, so no powered-test claim is made.

The method follows Torres, Continuous Discovery Habits (weekly outcome-linked touchpoints and competing solutions), and Bland & Osterwalder, Testing Business Ideas (risk and evidence-based test sequencing). The challenge for the decision review is: what observed behavior shows that care or returns uncertainty changes the decision, rather than merely attracting attention?

Supporting deliverables

evidence/ preserves all six supplied materials. calculations.json contains arithmetic and assumption scores; assumption_ranking.txt is the mapper output. cadence_report.json and ost_lint.json contain the machine checks. ost.json and discovery_log.json preserve the current tree and completed discovery record. research_worksheet.csv is the blank next-cycle recording template. No new research results have been added.