Research
Assumption mapping explained: choose what to test before you commit

Dylan de Heer
Assumption mapping is a way to compare the beliefs a plan depends on by how important they are and how much supporting evidence you have. Its purpose is to help you decide what to investigate next. A belief that is essential to success and weakly supported usually deserves more attention than a minor detail you already understand.
The map makes a discussion visible; it does not turn a belief into a fact. A confident team, a frequently repeated statement and a precise-looking score can all coexist with a serious evidence gap.
This explainer focuses on the importance-and-evidence approach for product and service teams. It shows what the dimensions mean, how to interpret the result and how to use a fictional example without mistaking prioritisation for validation.
What an assumption map is for
A product plan relies on conditions that may not yet be established. People must want the outcome, the team must be able to deliver it, and the proposed operating model must be workable. Assumption mapping brings those dependencies into the open so that learning effort follows the decisions at risk.
David J. Bland describes the method as making hypotheses explicit and comparing their importance with the available evidence. His Strategyzer guide distinguishes desirability, feasibility, viability and adaptability, then focuses investigation on consequential beliefs with little support. Strategyzer: assumptions mapping.
For an ordinary product discussion, the useful output is a small set of prioritised questions with their current basis and next action. A completed board is not, by itself, a research finding or permission to implement the idea.
The two dimensions: importance and evidence
Importance asks what depends on the belief
Ask what would change if the assumption were false. Would the proposed solution become impractical? Would the audience no longer need it? Would you change a small detail, or would you abandon a substantial commitment?
Keep this question tied to a particular decision. “Customers like convenience” is broad and difficult to place. “The intended users can complete the required step without an administrator present” identifies a condition that could change the operating model.
Importance is not the same as the seniority of the person raising the point, the excitement around an idea or the effort required to investigate it. A difficult question can be essential, and an easy question can be peripheral.
Evidence asks what supports this specific statement
Name the material you are relying on and what it actually establishes. An interview may show that a participant describes a problem. An observed task may show how that person handled one situation. A controlled technical test may show that a component worked under stated conditions.
Those sources answer different questions. A participant requesting a feature does not demonstrate that they will use it in routine work. A successful component test does not establish demand. Keep the source, its conditions and its limits alongside the assumption.
Avoid equating the quantity of material with its relevance. Several summaries of the same interview are still derived from one conversation. A large collection of general comments may tell you less about a narrow dependency than a directly relevant observation.
How to read the map
You can sketch two axes or use a table if a shared board is unnecessary. Make the direction of the evidence axis explicit: different versions place “more evidence” on different sides. Refer to the meaning of a position rather than telling people to work on a particular corner without checking the labels.
Importance to the decision | Current supporting evidence | Useful interpretation |
|---|---|---|
High | Limited or poorly matched | A candidate for early investigation, especially before a difficult-to-reverse commitment |
High | Relevant and reasonably strong for the question | Keep the basis visible and check whether it still applies when conditions change |
Lower | Limited | Record the uncertainty; investigate when it becomes consequential rather than because it is easy |
Lower | Relevant | Usually less urgent for this decision; avoid spending all the team's learning effort here |
These are relative judgements for a stated context. They are not measured probabilities, fixed thresholds or a universal order for every project. An urgent dependency, an existing commitment or an inexpensive decisive check can affect what the team does next.
If two people place the same assumption differently, ask which consequence or source they are considering. The disagreement may reveal two different assumptions hidden in one sentence.
Turn a broad belief into something you can investigate
Consider the statement “Customers will use the new portal”. It combines several uncertainties: whether the portal serves an important need, whether people can access it, whether they understand the task and whether the organisation can support it.
Separate the parts that could lead to different decisions. For example:
The intended user can obtain the information needed to submit a request.
The approver can act without relying on a separate informal message.
The operations team can handle requests that fall outside the ordinary path.
Each statement still needs refinement for a real study. Define the intended users, task and relevant conditions. Avoid combining a user's preference, a technical dependency and a financial forecast into one item merely because they concern the same feature.
An assumption is also different from an agreed requirement. The requirement describes an expected outcome or constraint; the assumption states something the current plan relies on without sufficient confirmation. See project assumptions vs risks for the wider planning distinctions.
A worked example: an onboarding portal
Imagine a fictional team considering a self-service onboarding portal. It is deciding whether to invest in a limited implementation. The examples below are invented planning material, not customer findings or measured product results.
Assumption | Why it matters | Current basis in the fictional example | Next question |
|---|---|---|---|
New users can supply the required account information themselves | The self-service workflow depends on that access | A manager says the information is “usually available”; no direct user task has been observed | Where do intended users obtain each required item, and which items need another person? |
Every exception can be resolved by the current support team | The proposed operating model assumes existing capacity and responsibilities | The team has reviewed ordinary cases but has not defined the exceptions | Which exception types occur, who handles them and what work is involved? |
Users prefer a blue confirmation button | It affects a reversible interface detail | A few internal colleagues express a preference | Is this detail consequential enough to investigate before the access and support questions? |
The first two assumptions could change the proposed workflow substantially. The button preference is easy to discuss, but that does not make it the most useful uncertainty to investigate now.
Suppose the team later observes a user who must ask a colleague for a required identifier. That observation challenges the unrestricted self-service assumption in that situation. It does not establish how often the problem occurs across the whole audience. The team might refine the intended scope, investigate other user groups or redesign the handover.
The useful change is in the next decision. The team has a more specific question and a visible reason to investigate it, rather than a board that simply looks more complete.
Confidence is not supporting evidence
Confidence describes a judgement. Evidence is the material offered in support of it. Keep them separate when someone says they are certain because “we have always known this” or because several colleagues agree.
Ask for the source behind the statement. Was the person present? Is the account about the same user group or an earlier version? Does it describe observed behaviour, a preference or an expectation? The answer may be useful even when it weakens the original claim.
Bland's public discussion of quantifying assumption maps cautions against decimal precision that creates an impression of certainty. Treat that as the method author's guidance, rather than evidence that every numerical approach is invalid. For a simple team discussion, an explanation of the source and its limits can be more useful than an unexplained confidence score. David Bland: assumptions mapping fundamentals.
If the team uses a rating scale, define its meaning and keep the underlying evidence accessible. A change from one number to another should not become the only record of what was learned.

From a position on the map to a learning question
Choose an investigation that could change the dependent decision. Start by stating what you need to observe, who or what could provide that evidence and what different results would mean.
A claim that an umbrella keeps someone dry is not answered by showing that it opens. In the same way, asking whether people like a proposed portal does not answer whether they can obtain its required inputs. Match the investigation to the assumption, rather than selecting a convenient activity first.
ENoLL's UNaLab assumption-mapper guidance adds the practical question of how easy or difficult a critical assumption is to test. It offers a workshop approach for identifying assumptions and choosing an investigation; its suggested group size and timings are not requirements for every team. UNaLab: assumption mapper.
Keep the question and the planned observation distinct from the result. If the next action is a proof of concept or a pilot, define its boundary separately. A map chooses a learning priority. A proof of concept may examine a feasibility question, while a limited pilot may investigate use in an operating setting. Choose the test around the uncertainty and define its boundaries before starting.

Genuine Weeve interface with a demonstration question and Local mode selected. This is a source-review starting point, not an automatically validated assumption map.
Use recorded planning discussions as source material
Assumptions can be scattered across kickoff meetings, reviews and follow-up calls. One person may describe a dependency as uncertain while a later summary presents it as settled. Revisiting the relevant discussions can help you find where that change occurred.
With permission to record, Weeve can retain conversations and transcripts on your Mac. Chat with recordings can help review selected discussions and return source references for the statements it finds.
A useful question is: “Across these selected planning recordings, identify conditions the proposal depends on. Cite the passages, distinguish stated evidence from expectations, and show conflicting accounts or unresolved details.” Check the source passages and the answer's coverage before moving anything into your map.
That is preparation for a human judgement. A spoken assertion is evidence that the assertion was made, not independent confirmation that the underlying condition holds. Weeve does not automatically validate assumptions, run the proposed tests or maintain an approved map for you.
Local AI workflows are available with downloaded models. Selecting an optional external provider for chat changes where relevant text is processed. Check the selected mode and permissions for the material before using it.
Keep the map useful as the proposal changes
Revisit the important entries before the decisions that depend on them. Update the source, the scope of the conclusion and the remaining uncertainty when new material arrives. Record who will investigate an unresolved point and when it needs to inform the decision.
An assumption can move because the evidence improves, because its importance changes or because the statement is rewritten more precisely. Preserve that reason. Moving an item to a more reassuring position without an explanation hides the learning the map was meant to support.
The useful result is a clearer account of what your plan depends on, what you currently know and what would be worth learning next.



