How to write a problem statement grounded in evidence

Dylan de Heer
To write a problem statement, identify who is affected, describe the difficulty in a specific context, explain the consequence your evidence supports, and make clear what you still need to learn. Keep the statement short enough to guide a decision about the work, with supporting evidence available separately.
For product and service work, this usually means describing a gap in the current experience before deciding how to change it. “We need a dashboard” names a proposed solution. It does not yet explain the difficulty that makes a dashboard worth considering.
This guide covers workplace product and service discovery. Academic research proposals may require a different structure, literature review or institution-specific format.
What a problem statement should do
A useful statement gives a team a shared question to investigate. It identifies the people and situation in scope and makes the reason for attention understandable.
NN/g’s discovery guidance treats a problem statement as a way to frame the problem, communicate scope and guide exploration. The University of Leeds’ practical guide also emphasises narrowing a description and challenging vague or unsupported wording.
The statement does not have to contain every interview finding, a finished causal explanation and a delivery plan. Those are different parts of the work. Its immediate job is to make the current difficulty clear enough that people can decide what to investigate next.
A good reader should be able to answer:
Who is experiencing the difficulty?
In which situation does it occur?
What is happening, compared with what those people need to do?
What consequence is supported by the evidence?
What important question remains unresolved?
Use that as a practical review, rather than treating a particular sentence formula as mandatory.
1. Start with observations, not a proposed feature
Gather the material that prompted the work: user interviews, support cases, project conversations, service records or observations of the task. Choose a boundary so you know which people, period and situation the statement describes.
Write down small, concrete observations first. “The support team keeps asking for clarification before starting work” is more useful than “our process is broken”, provided you can show which cases or conversations support it.
Keep three kinds of material separate while you work:
Observed or reported difficulty: what happened, or what someone described happening.
Interpretation: what you think might explain the difficulty.
Proposed solution: a possible change someone would like to make.
All three can be worth keeping. The distinction prevents a strongly expressed idea from becoming an unexamined fact.
For example, a stakeholder saying “we need a single intake form” establishes that stakeholder’s proposal. It does not by itself establish that missing forms are the cause of delays or that one new form is the best response.
Expected result: an evidence set with the difficulty, possible explanations and ideas clearly distinguished.

2. Name the people and the moment that matter
Replace “users” or “the business” with the group and activity you actually mean. A problem may affect several groups, but their needs can differ.
“Customers cannot get help” is too broad to tell a team where to start. “Customers submitting a first support request cannot tell which information the receiving team needs” is more specific and can guide useful questions.
Identify the point in the task where the difficulty occurs. Is it before submission, during a handoff, when someone reviews a result, or when they return to earlier work?
Do not expand the claim beyond the material reviewed. Evidence from a small set of onboarding conversations can identify a worthwhile problem to investigate; it cannot establish that every customer has the same experience.
If two groups experience different difficulties, decide whether one statement can accurately explain the relationship or whether you have two distinct problems. Avoid hiding unrelated work inside a long paragraph.
Expected result: a clear audience and context, with a visible boundary around the claim.
3. Describe the gap in plain, observable terms
State what the people are trying to do and what gets in the way. Prefer language a reader can investigate over labels that sound precise without explaining anything.
Consider these changes:
“The handoff is inefficient” becomes “The receiving team has to ask who owns the request before it can assign the work.”
“People lack visibility” becomes “The requester cannot see whether their request has been accepted or needs more information.”
“The experience is confusing” becomes “New starters use different sources for the same approval step and receive conflicting instructions.”
These are illustrative rewrites, not measured findings. A real statement should use the behaviour your own evidence supports.
Be careful with comparisons such as “too slow”, “too costly” and “low quality”. Explain the relevant expectation or consequence where you know it. If there is no agreed target yet, record that as an open question instead of inserting an arbitrary threshold.
Expected result: a description of the current difficulty that someone could check against the source material.
4. Explain the consequence without inventing impact
Why does the problem matter to the affected people or the organisation? It may create extra clarification, uncertainty, repeated work, an avoidable wait or a task people cannot complete.
Use numbers when you have a reliable basis for them. Check the population, period and definition before using a percentage or average. A participant’s estimate of time spent is a report, which may be useful, but it should not be presented as a measured organisation-wide cost.
When the impact is not quantified, say what is known and what needs measuring. For example:
The reviewed cases required additional clarification before assignment. We have not yet established how often this occurs across all incoming requests or how much waiting time it adds.
That is a usable evidence boundary. It helps the team decide what to inspect next without pretending the research is complete.
Avoid asserting the root cause merely because the consequence is visible. A missing field in a request, unclear responsibility and different expectations about readiness may require different responses.
Expected result: a reason to investigate, with reported, observed and measured evidence distinguished.
5. Write a concise draft and keep the evidence beside it
Once the components are clear, write a short paragraph in natural language. One practical structure is:
For this group, in this situation, this difficulty prevents or complicates this task. The evidence reviewed shows this consequence. We still need to understand this important uncertainty.
You do not need to use those exact words. Include only the elements that help the intended reader understand the work.
Keep the statement concise, then put the supporting notes, cases and sources below it or in a linked evidence record. A short statement with a clear trail is more useful than a confident claim that no one can retrace.

If the evidence comes from recorded conversations
In Weeve, review relevant saved recordings within a deliberate scope before asking for a draft. Select the relevant material and use a request such as:
Review these selected customer and project conversations. Identify concrete difficulties people described, the task and context for each, and the source passages. Separate observations from possible causes and proposed solutions. Show disagreements and important unknowns. Do not invent frequency, cost or completion claims.
This is a proposed research prompt, not a tested output. Check the returned passages and the material actually reviewed. A quotation can support what one person said; the broader statement needs an appropriate account of the whole evidence set.
Local chat runs locally. Optional external chat providers may receive relevant text, so choose the processing mode appropriate to the material. Do not call a generated draft approved research or a validated causal analysis.
For the broader process of moving from sessions to findings, use the research-synthesis guide.
Expected result: a short statement with traceable support and explicit unknowns.
Worked example: turn a requested form into an investigable problem
The following is fictional. It demonstrates the reasoning and does not describe a Weeve customer or a measured business outcome.
A team begins with this request:
We need a new form so service requests stop getting delayed.
There are two claims inside that sentence: a proposed form and a claimed relationship to delay. Neither should become the starting assumption for the investigation.
Imagine that a review of a small set of request handoffs shows the receiving team asking for clarification about the intended outcome and the person who can answer questions. Some requests use a form; others arrive through a conversation. The review does not establish the total delay across the service.
A more useful draft would be:
When new service requests reach the delivery team, the intended outcome and a contact for clarification are not consistently clear in the reviewed handoffs. The team has to seek that information before deciding how to assign the work. We need to establish how often this occurs and which part of the handoff leaves the information unclear.
This draft leaves several responses available: improve the existing form, change how requests are discussed, make responsibilities clearer or add a readiness check. The statement does not choose between them prematurely.
Keep a separate note of the original form proposal. It may become a good option once the team understands the difficulty. Separating it from the statement preserves the idea without treating it as the finding.
6. Review the statement with people who know the work
Ask an affected person, someone receiving the work and the person responsible for the investigation to read the draft. They may not agree, and that disagreement can be useful evidence.
Use concrete questions:
Does this describe the situation you recognise?
Which part is supported, and which part is an assumption?
Is an important affected group missing?
Does the wording imply a cause that has not been established?
Could more than one sensible solution address this difficulty?
What information would make us revise the statement?
If people disagree about the problem, preserve the differing accounts and investigate the reason. Rewording a sentence until nobody objects is not the same as resolving the underlying difference.
Expected result: a reviewed statement with named open questions, rather than artificial agreement.
Common problems to correct before the project begins
A solution disguised as a problem. Remove the requested tool from the core claim and ask what people cannot do satisfactorily today. Keep the tool as a proposal to assess later.
An unsupported root cause. Replace the causal assertion with the observed difficulty and record the explanation as a hypothesis.
A list of unrelated complaints. Choose one bounded difficulty or separate the work into distinct investigations.
False precision. Remove invented percentages, costs and time savings. State the known consequence and the measurement needed next.
A statement that never changes. Review it as new evidence appears. Keep a short record of material revisions so people can understand why the scope changed.
Once the problem is clear enough, choose the next learning step or decision. If the work moves into requirements, the requirements-elicitation guide explains ways to investigate needs and resolve differences. The problem statement remains the reference point for why that work matters.



