Product & Project Delivery
Write a Product Problem Statement Before Your Team Commits to a Solution
Rewrite a proposed feature as a bounded user problem using a complete fictional case, deduplicated evidence, negative cases and a decision record that preserves what remains unknown.
Write a product problem statement around the person, situation, obstacle and consequence you can support with evidence. Keep the requested feature in a separate proposal field. Then state what you still do not know and which finding would make you rewrite the statement. A clear statement should help your team decide what to investigate before it commits engineering time.
A request such as “add an overdue badge and automatic reminders” already contains a solution. It does not establish who needs help, whether an item is actually overdue, or why the existing process fails. This guide from LandOffer.ai uses a completed fictional case to show how to make those distinctions. All people, records, observations and decisions below are authored teaching material; no customer research or product test was performed.

Start with the user's unfinished task
Ask what the user is trying to accomplish at the moment the difficulty appears. “Use a better dashboard” describes a proposed interface. “Tell the next shift which returned kits still need counting” describes a task that could be supported in several ways.
UCSF's Product Toolkit presents a problem statement as a concise description of a current issue that directs investigation. Its guidance asks teams to examine who, what, where, when and why, and to consider user needs before solutions. That provides a useful starting discipline without requiring your company to adopt UCSF's workshop format.
Be specific about the context. A shift volunteer reconciling a multi-piece kit after an unattended return may face a different problem from a borrower checking an account at home. Putting both into “users cannot manage returns” hides the distinction. Keep the two users separate until evidence shows they share the same obstacle. A reviewer should be able to identify whose experience the statement describes.
Write the immediate consequence at the same level of confidence. If staff ask a colleague to recount a kit, record that additional check. Do not turn it into lost revenue, customer churn or hours saved unless your evidence actually establishes those effects.
Rewrite the feature request using the same evidence packet
CedarKit is an invented service for lending reusable activity kits through two community desks. Product lead Ada receives this request from operations lead Joel: “Add a red overdue badge and automatic borrower reminders so staff stop chasing missing kits.” The product team has not accepted that request for implementation.
The authored packet contains six handover observations involving four volunteers, selected specifically because a bundle return was awaiting reconciliation. Four handovers included a question about whether the pieces had been counted; two included a physical inspection. These are overlapping descriptions of six episodes, not eight independent incidents. The two inspections occurred within the four questioning episodes.
A separate fictional register contains 24 return visits from October 19–23, 2026: nine await counting and 15 have been reconciled. Five of the nine pending visits display “Returned”; four display “Count pending.” The original request treats a pending count as a missing kit, but the records do not establish that pieces are missing or that the borrower still has them.
Ada first writes: “Borrowers and desk staff cannot see accurate return status, causing overdue kits and repeated support contacts.” She rejects it. Borrowers have not been studied, “accurate” has no agreed definition, and overdue status depends on rules the packet does not supply. The statement also attributes support contacts to a cause without checking their circumstances.
Her completed revision reads:
Problem statement, version 2: Rotating CedarKit desk volunteers taking over multi-piece returns after unattended drop-off cannot reliably distinguish “received but not counted” from “count completed” in the current handover information. In four of six selected pending-return handovers, volunteers asked whether a count had happened; two of those four also inspected the physical kit. This uncertainty creates additional checking before they can give the next shift a dependable account of the return. The pattern is documented only in this authored sample; its frequency across all returns and its cause remain unestablished.
The scope is deliberately narrower than the feature request. A kit might have been received without having been counted. A count might be complete but communicated poorly. A reminder aimed at the borrower could address neither condition. The revision preserves that uncertainty. Joel's suggestion remains useful evidence of an operational concern, while its suitability as a response still needs investigation.
Ada also records the associated user need: “When I take over an unattended bundle return, I need to know whether its contents have been checked and what remains unresolved, so I can give the next shift a dependable handover.” This is a need to understand a state, not a requirement for a particular badge, notification channel or database field.
Keep observations, interpretations and unknowns traceable
The GOV.UK Service Manual's research-analysis guidance separates what people saw or heard from the findings they derive and the actions they propose. Apply that separation to your statement. A summary should retain a route back to the underlying record rather than turning several interpretations into one apparent fact.
The following completed evidence register belongs to the same fictional CedarKit packet. Its purpose is to show which claims survive inspection, including evidence that limits the proposed problem.
| Record | Authored evidence | Supported reading and boundary |
|---|---|---|
| E01: handovers | Six selected episodes, four volunteers; R01, R03, R06 and R08 include a count question; R03 and R06 also include physical inspection | Confusion exists in these episodes. Selection focused on pending bundles; this is not an estimate for all users or returns |
| E02: return register | R01–R09 await counting; R10–R24 reconciled. R01, R03, R06, R07 and R08 carry “Returned” | Nine pending records and five potentially ambiguous labels. Neither count verifies missing pieces; label ambiguity is an interpretation, not an established cause |
| E03: support links | Seven tickets concern five visits: two each for R03 and R06; one each for R08, R14 and R21 | Tickets are not seven unique affected returns. Three linked visits are pending; two are reconciled and need separate explanation |
| E04: contrasting episodes | R04 and R09 display “Count pending”; observed volunteers do not ask the count question | A useful contrast, not proof that changing the wording would remove uncertainty elsewhere |
| E05: reminder request | Joel proposes borrower messages; no borrower interviews or reminder evaluation in the packet | A stakeholder solution hypothesis. Borrower need, channel suitability and consequences remain unknown |
| E06: operating rule | Volunteers refer to an end-of-shift count; no approved count deadline or overdue definition is supplied | Investigate the current rule before describing pending counts as overdue or specifying automatic messages |
Link tickets to return-visit identifiers so a reviewer can distinguish repeated contacts from separate affected visits. Here, seven support tickets collapse to five return visits, and only three of those visits are pending counts. The ticket data may justify examining the handover problem, but it cannot support “seven missing kits.” It also does not establish how many borrowers were affected.
The register's negative evidence changes the question. Two pending-return handovers with clearer wording did not generate a count question, while two support-linked visits were already reconciled. Ada preserves both facts. Removing them would make the label explanation look stronger and obscure a potentially different support problem.

Make uncertainty specific enough to change the statement
The GOV.UK user-needs guidance recommends grounding needs in user evidence and framing them around a problem rather than a preferred solution. It also distinguishes broader needs from the more constrained stories used to organize delivery. CedarKit's statement therefore stops before behavioral requirements, implementation choices or acceptance tests for a new feature.
Ada turns three assumptions into questions. First, does the handover information lack a distinct count state, or do volunteers struggle to find an existing one? Second, is the state maintained consistently across both desks? Third, do borrowers need a different explanation of receipt versus reconciliation? Those questions could lead to different problem definitions. They should not all be compressed into “the UI is confusing.”
She records what would weaken version 2: evidence that volunteers can consistently find the relevant state but cannot trust a count performed earlier. That would shift the investigation from discovering status toward confidence in the counting process. Evidence that the difficulty occurs only at one desk would narrow the context further. Evidence of a physical storage mix-up could reveal a problem outside the proposed interface altogether.
Do not label all unknowns as research blockers. A decorative color preference can wait. Whether the borrower still possesses the kit cannot wait if the team is considering a message that implies non-return. Give the uncertainty priority according to the decision it could change, rather than the ease of answering it.
Close the discussion with a bounded problem decision
In the fictional review on October 27, Ada, Joel and research partner Priya agree that version 2 is suitable for a focused investigation. This is an authored agreement, not a real meeting or a commitment to build. Joel keeps his badge and reminder idea in a proposal note, linked to E05; it is not relabeled as an established need.
Their completed decision record states:
- Accepted problem scope: handover understanding for rotating volunteers dealing with pending multi-piece returns at the two desks. Single-item returns, borrower account views and general inventory accuracy are outside the current statement; none is covered by the six selected pending-bundle observations.
- Evidence accepted: E01's bounded observations and E02's record states support investigating uncertainty. E03 is deduplicated by visit and retains the two reconciled exceptions. No prevalence, monetary loss or causal claim is accepted.
- Next question and ownership: Priya will prepare a proposed observation approach comparing how volunteers locate and interpret receipt and count information. Joel will identify the owner of the current counting rule. Neither recruitment nor staff time has been arranged.
- Decision boundary: Ada will revisit the statement after those questions are answered. No reminder implementation, status redesign, engineering estimate, launch date or reduced-contact target is approved by this record.
This is a completed problem-definition decision even though research remains undone. The team has chosen a bounded question, retained the proposed solution without endorsing it, and identified which missing evidence could change direction. It has not solved the user problem merely by agreeing on the wording.
GOV.UK's research-planning guidance explicitly advises turning unfounded assumptions into research questions and selecting activities that answer the important questions. Use that principle to connect a statement to the next learning decision. The suitable activity and participants depend on your actual context; CedarKit's numbers are illustrative, not a prescribed sample size.
Review the statement before it becomes a build instruction
Read the statement aloud with someone who did not write it. Can they identify the user, the triggering situation and the obstacle? Can they locate the supporting records? Could a different explanation fit those records? If they can only describe the feature you expect to ship, separate the solution from the problem again.
Ask one further question: what would the user be able to do differently if this problem were addressed? For CedarKit, the answer is a dependable handover of receipt and counting status. It is not automatically fewer tickets, a red badge or a reminder campaign. Those might later become hypotheses with their own evaluation, or they might be discarded.
Before your next solution discussion, bring one revised statement, its evidence register and a short record of what remains unapproved. Keep the rejected wording so reviewers can see why the scope changed. For more workplace guides, visit LandOffer.ai. Before the team estimates a solution, trace each factual clause in your statement to a record. Rewrite unsupported claims as questions, and identify which question must be answered before a build decision.
Sources and Further Reading
- UCSF Product Toolkit: Problem Statement, checked October 27, 2026; no page revision date stated.
- GOV.UK: Learning about users and their needs, checked October 27, 2026; last updated March 23, 2017.
- GOV.UK: Analyse a research session, checked October 27, 2026; last updated May 24, 2016.
- GOV.UK: Plan user research for your service, checked October 27, 2026; page reports last updated October 15, 2026.


