Skip to content

Lean UX Canvas

beta  ·  Family discovery-docs  ·  Phase discover  ·  Sizes lean  ·  ~5,200 tokens

The one-page canvas that frames one initiative’s business problem, the outcomes the business wants, the users affected, the outcomes those users want, and candidate solutions, then turns those links into testable hypotheses and names what to learn first and the least work that will learn it, before anyone commits to building.

An operator card. Read it before a first pass and again every time the team loops back through it. The reasoning behind every rule here lives in lean-ux-canvas_companion.md.

Open a canvas when an initiative is about to start and nobody has yet written down the business problem, the outcomes, the users, or a candidate solution for it. Gothelf’s own walkthrough starts here, before any other box is touched, and the book calls the canvas “a great way to start an initiative,” filled by the whole team with stakeholders and clients joining “as much as possible.” A first pass rarely takes less than half a day, and some teams spread it across a series of meetings over a couple of weeks; either is normal, and the spread reflects the initiative’s own scope, not disagreement about the method.

Treat the canvas as a facilitation tool first. Its originators call it exactly that, “a facilitation tool for cross-functional teams designed to create a customer-centric conversation,” and the filled page is the record of that conversation, not the thing of value on its own. If no cross-functional conversation is happening, a page of boxes filled in alone by one person has not done this document’s job, whatever it says.

If the real question is Write this instead
Which problem is worth solving, before anyone names a solution? problem-statement
We have chosen one concept; who will design it? design-brief
We need to plan one test of one hypothesis, with variants, allocation, and a decision rule experiment-design-doc
Where is the whole product going, over years? product-vision
Who exactly are we building for, researched rather than guessed? user-persona
What does the whole business look like: channels, revenue, cost structure? lean-canvas, Ash Maurya’s Lean Canvas; this canvas asks nothing about revenue or cost (2026-10-30: this cell said the bundle was planned and not yet built; it is built now)
  • The problem is already chosen and so is the concept. A canvas names candidate solutions in Box 5 and turns them into hypotheses in Box 6; it has not chosen one. Once a concept is chosen, hand it to the people who will design it in a design brief instead.
  • The team already knows exactly which test it is running. Box 8 only sketches the shape of a test: the smallest thing that would learn whether the riskiest assumption holds. It does not plan one, with variants, an allocation, metrics, and a decision rule. If the team already knows those, it needs an experiment-design-doc, not another pass of this canvas.
  • Nobody from outside the writer will read or discuss it. The canvas earns its name as “a facilitation tool,” and its own book lists communicating the work “broadly beyond the team” among what a filled canvas is for. A canvas written and filed by one person, with no conversation and no wider reader, has skipped the half of its job a document alone cannot do.
  • The initiative is actually the whole product’s direction for years, not one initiative for weeks. That scale is a product vision’s job.

There is nothing to pick. This bundle ships one size, lean, with no full variant: Gothelf publishes the canvas as a single one-page shape in both the 2019 and 2024 versions, and this research found no source that splits a short form of the canvas from a longer one. Do not invent a full-size version by adding sections the originators never wrote; a team that needs more structure than the canvas itself offers is better served by experiment-design-doc once Box 8’s smallest experiment becomes a real test, or by design-brief once the team has chosen a concept, than by an invented longer canvas. See the companion’s Variants and sizing section.

Below 12 out of 18, the canvas reads as filled-in homework rather than the record of a conversation. A reader cannot tell what the team actually believes about any box, and the facilitator ends up walking every reader through it live instead of letting the page carry the conversation on its own.

Each row’s test is the same: could someone satisfy this cell without actually improving the box? If a cell can be cleared by adding words rather than evidence, it is written wrong, and no row here is scored by counting items.

# Criterion 0 1 2
1 Problem, not a solution The box names a feature or a solution, or no evidence is given for what changed A problem is named, but you cannot point to the evidence behind it, or it never closes on a how-might-we question You can point at the sentence naming what changed, the evidence that it is real, and the closing how-might-we question
2 Outcome is a behavior A revenue figure, an output count, or a feeling, with no stated magnitude A behavior is named, but with no stated magnitude or timeframe, or it reads as a feeling rather than a measured change You can state who does what, by how much, and by when, as a measured behavior change, not a number on a report or a feeling
3 Users named specifically Only “users” or a single buyer is named A group is named, but a stranger to the project could not picture a member of it, or only the buyer appears You can name a group a stranger could picture, and it includes anyone who configures the product or is affected by it beyond the buyer
4 A distinct user outcome The user outcome restates the business outcome’s measured behavior in different words A user’s own reason is attempted, but it still reads as a measurable, business-shaped target You can point to a reason stated in the user’s own language, distinct from Box 2’s measured behavior, even if it is not itself measurable
5 Genuinely different candidates A single solution appears, with no non-technical option Several candidates appear, but they are variations of one software feature You can point to candidates that take different approaches to the problem, one of them a non-technical option that is not a thinner version of a software one
6 One feature per hypothesis A row spans more than one feature, or a slot (outcome, user, benefit, feature) is missing One hypothesis per row, but a slot is vague enough that you could not design a test from it alone Each row names the business outcome, the user, the benefit, and exactly one feature, specific enough to test on its own
7 One named riskiest assumption A list of risks appears, with no single one chosen An assumption is implied, but several are named with no stated priority among them You can point to one named assumption, and the sentence states it as the one most important to learn right now
8 Learning, not a release The box describes building a smaller version of the eventual product An activity is named, but you cannot tell whether it tests the named assumption or only previews the feature You can point to a specific test that would prove or disprove the named assumption, with no feature actually built
9 Dated revision outcomes No dated row exists, or a row states only that something was “tested,” with no outcome A dated row exists, but you cannot tell whether the assumption held or which boxes changed as a result You can point to a dated row naming the assumption tested, whether it held, and which boxes changed; once the canvas stops changing, a successor document is named

Starting from a solution rather than a problem. Box 1 asks what changed in the world, not what to build. A community pitfalls list puts this first among six recurring failures, and the canvas’s own structure makes it visible immediately: nothing in Box 1 should read like Box 5.

Writing a vague business outcome. A revenue figure, an output count, or a sentiment in Box 2 fails Gothelf’s own test that an outcome is a measured change in human behavior. A vague outcome in Box 2 also breaks Box 6, since the hypotheses table pulls its business-outcome slot straight from this box.

Naming a user segment too broadly to act on. “Users” or “fleet managers” cannot be pictured by a stranger to the project, and Gothelf tightened Box 3’s wording in 2019 specifically because teams were taking this shortcut. Naming only the buyer and skipping everyone who configures or is affected by the product is the same failure in a different shape.

Confusing the business outcomes box with the user outcomes box. This is the canvas’s best-documented point of confusion: Gothelf names it directly, and an independent podcast discussion of the same revision reports that Box 4 was the part of the canvas with the most confusion. Business outcomes are measured behavior; user outcomes are why users seek the product out in the first place, and they are often not measurable at all. Writing one box’s content into the other is the confusion Gothelf rewrote Box 4 to remove, and a community pitfalls list names it as one of its six recurring failures.

Filling the solutions box with only one idea. Box 5 was explicitly widened in 2019 to admit new systems, business models, and non-technical solutions, not only software features. A single idea here undercuts that widening before the team ever reaches Box 6.

Bundling more than one feature into a single hypothesis. The canvas’s own rule is one feature per hypothesis; a sentence spanning several features cannot be tested cleanly, because a result would not say which feature caused it.

Skipping the experiment step entirely. A canvas that reaches Box 6 and stops, with no riskiest assumption named in Box 7 and no smallest test designed in Box 8, has stopped short of the reason the canvas exists: to find out whether the hypotheses are true before anyone commits to building.

Leaving the revisions section blank because a first pass feels finished. The canvas is revised while its experiments run: after Box 8 returns a result, the team asks Box 7’s question again, and evidence can even send the team back to Box 1. A canvas revised with no record of what changed cannot show what the team actually learned, which is exactly why this section exists.

Self-grade against the rubric above, then delete every HTML comment; they are guidance, not content. Remember that a canvas is not finished when you first fill it. It stays open while the initiative’s experiments run, and it is finished only when the team commits to building, changes course, or stops the initiative, at which point the final row of the revisions section should name what takes over.

lean-ux-canvas_template-lean.md · ~5,200 tokens

---
title: "{{initiative_name}} Lean UX Canvas"
initiative_name: "{{initiative_name}}"
facilitator: "{{facilitator}}"
status: "{{status}}"
date: "{{date}}"
iteration: "{{iteration}}"
doc_type: lean-ux-canvas
size: lean
source_template: lean-ux-canvas
source_template_version: 0.1.0
---
<!--
LEAN UX CANVAS. This bundle ships one size, deliberately: Gothelf publishes the canvas as a single
one-page shape in both the 2019 and 2024 versions, and this research found no source that splits a
short form of the canvas from a longer one. See lean-ux-canvas_companion.md section 4 (Variants and
sizing).
WHAT THIS DOCUMENT IS, AND IS NOT. A Lean UX Canvas frames one initiative's business problem, the
outcomes the business wants, the users affected, the outcomes those users want, and candidate
solutions, then turns those links into testable hypotheses. It closes on the single most important
thing the team needs to learn right now and the smallest piece of work that would learn it. It is a
facilitation tool first: its originators call it a facilitation tool for cross-functional teams
designed to create a customer-centric conversation, and the filled page is the record of that
conversation, not the thing of value by itself. See lean-ux-canvas_companion.md section 1
(Orientation).
It is NOT a problem statement (that names which problem is worth solving, before anyone names a
solution; this canvas names candidate solutions in Box 5 and links them to outcomes in Box 6 - a
problem statement, where one exists, can feed Box 1, but nothing this research read makes that a
rule). It is NOT an experiment-design-doc (that plans one test of one hypothesis, with variants,
allocation, metrics and a decision rule; this canvas only picks which hypothesis to test next and
sketches the least work in a single box, Box 8). It is NOT a design-brief (that hands one chosen
concept to the people who will design it; this canvas has not chosen a concept, only named
candidates in Box 5). It is NOT a product-vision (that states a whole product's direction for
years; this canvas frames one initiative for weeks). It is NOT a user-persona (Box 3 names the
users in a line and may point to a persona; it is not the researched portrait itself). It is NOT
Ash Maurya's lean-canvas (that models a whole business - channels, revenue, cost structure - while
this canvas frames one initiative's hypotheses and asks nothing about revenue or cost). Where this
canvas sits relative to each of these neighbors is this library's own POSITION in places, not a
sourced sequence; see lean-ux-canvas_companion.md section 8 (Relationships to other artifacts) for
which lines are sourced and which are not.
THIS DOCUMENT IS REVISED WHILE THE INITIATIVE'S EXPERIMENTS RUN. After the smallest experiment in
Box 8 returns a result, the team asks Box 7's question again, and the evidence can even send the
team back to Box 1 if it shows the wrong problem, the wrong audience, or the wrong measure of
success was in play. It is finished when the team commits to building, changes course, or stops the
initiative, not on a calendar. Box 9 below exists to keep that history visible, rather than letting
the canvas go stale with no record of what changed. See lean-ux-canvas_companion.md section 3 (Box
9: Revisions and What Takes Over).
HOW TO FILL THIS IN
1. Read the comment under each heading: WHAT it wants, WHY it matters (with a pointer into
lean-ux-canvas_companion.md), guiding questions to ASK, a GOOD and a WEAK example, and the TRAP
to avoid. For the two table sections, PRIORITY explains how to use the table and ROW HINT says
what a good row contains.
2. Fill the boxes roughly in the order below on a first pass: Gothelf's own walkthrough starts at
Box 1, and Box 6 cannot be written before Boxes 2 through 5 exist. Once the canvas is underway,
any box can be revisited on its own.
3. If a section does not apply to this initiative, write "N/A" and one line of why, rather than
deleting it.
4. This canvas is not finished when you first fill it; it is revised as experiments return results
(see above). Before you first share a pass: self-grade against lean-ux-canvas_guide.md, then
DELETE every HTML comment. They are guidance, not content.
-->
# {{initiative_name}} Lean UX Canvas
## Business Problem
<!-- WHAT What changed in the world and why it makes this a problem worth the organization's
attention right now, closing on a "how might we" question. Not a solution or a feature
idea.
WHY Gothelf's own walkthrough of how to use the canvas starts here, naming the business
problem statement before touching any other box. The box's closing "how might we"
question also supplies the objective half of an OKR built from the canvas, with Box 2
supplying the key results half. Deep dive: lean-ux-canvas_companion.md section 3 (Box 1)
and section 8 (Relationships to other artifacts).
ASK What changed in the world that makes this worth the organization's attention now? What
evidence shows this is a real problem rather than an assumption? If you ended this
statement with "how might we," what question would it ask?
GOOD "Dispatchers at single-site depots phone every driver each morning to confirm status,
after three depots reported drivers marked absent in error last quarter. How might we
confirm a driver's status without a phone call?"
(names what changed, cites evidence, ends on a how-might-we question)
WEAK "We should build an automated check-in feature."
(a solution, not a problem, and no evidence of what changed)
TRAP Writing a solution or a feature idea into this box instead of the problem it would
solve. That is Box 5's job, not Box 1's, and starting from a solution rather than a
problem is the first of a community skill file's named pitfalls. -->
{{business_problem_statement}}
## Business Outcomes
<!-- WHAT The measurable change in human behavior the business wants, stated as who does what, by
how much. Not a revenue figure or an output count.
WHY Gothelf defines a business outcome as a measure of human behavior, not a number on a
report, and has separately called this box perhaps the most critical one in the whole
canvas, because its outcomes become the success criteria Box 6's hypotheses are later
tested against. The 2024 prompt made that discipline more explicit by asking for the
same who-does-what-by-how-much format his and Seiden's OKR book uses. Deep dive:
lean-ux-canvas_companion.md section 3 (Box 2) and section 6 (Business outcomes against
user outcomes).
ASK What will people do differently if this business problem is solved? Can you state it as
who does what, by how much, and by when? Is this a behavior you can actually measure,
not a sentiment or a revenue figure?
GOOD "Dispatchers who use the check-in feature make 80% fewer manual confirmation calls
within 60 days of rollout."
(a measured behavior, a magnitude, a timeframe)
WEAK "Dispatchers feel more confident in their fleet's status."
(a feeling, not a measured change in behavior)
TRAP Writing a revenue figure or an output count here instead of a behavior change. Confusing
this box with Box 4 is the canvas's best-documented point of confusion, and Gothelf's
own revision of both boxes exists to draw that line. -->
{{business_outcome}}
## Users
<!-- WHAT Who the team is solving for, named specifically enough that someone unfamiliar with the
initiative could picture them, including anyone who configures the product or is
affected by it, not only the person who buys or approves it.
WHY The 2024 wording widens this box past paying customers to anyone who configures the
product or is affected by its use, and Gothelf tightened the box in 2019 because teams
were taking a shortcut on it, treating it too casually. The book's own practice fills
this box with proto-personas where the team has them. Deep dive:
lean-ux-canvas_companion.md section 3 (Box 3).
ASK Who specifically uses or is affected by this, beyond whoever buys or approves it? Could
someone outside the team picture this user from your description? Does a researched
persona already exist that this box should point to, rather than inventing a thinner
sketch for the canvas?
GOOD "Dispatchers at single-site depots who coordinate 15-40 drivers each morning, and the
drivers themselves, who are affected by the check-in process even though they do not
choose it."
(names more than one affected group, specific enough to picture)
WEAK "Fleet managers."
(too broad to act on, and skips the affected drivers entirely)
TRAP Naming only the buyer and skipping everyone else who configures or is affected by the
product. Treating this box as a formality is exactly the shortcut Gothelf tightened the
wording to stop. -->
{{users}}
## User Outcomes and Benefits
<!-- WHAT What users want to achieve and how they want to feel, in language a user would
recognize, not a restatement of the business outcome in different words. The 2024 prompt
adds jobs-to-be-done language.
WHY Gothelf names this box as the part of the canvas with the most confusion, and resolves
it directly against Box 2: business outcomes are measured behavior, while user outcomes
are the reasons users seek out a product in the first place, often not measurable goals
at all but intangible states of being. Deep dive: lean-ux-canvas_companion.md section 3
(Box 4) and section 6, which also records a minority "ethics box" reading that Gothelf's
own pages do not use.
ASK Why does this user actually seek this out, independent of what the business wants from
them? What job are they trying to get done? How do they want to feel afterward, in
language they would recognize?
GOOD "Dispatchers want to trust that a driver's status is accurate without double-checking it
themselves, so they can start the day's routing without delay."
(a reason to seek the product out, in the user's own language, distinct from the
business's measured behavior)
WEAK "Dispatchers make 80% fewer calls."
(this restates Box 2's business outcome, not a user outcome)
TRAP Writing the business outcome again in different words. Box 2 against Box 4 is the
canvas's best-documented point of confusion, and filling either box with language that
belongs in the other is, by Gothelf's own account, the single most common way a team's
canvas becomes hard for anyone else to read. -->
{{user_outcomes_and_benefits}}
## Solutions
<!-- WHAT Candidate solutions, listed as widely as the box invites, including at least one
non-technical option, before narrowing to anything specific.
WHY The 2019 widening explicitly admits new systems, business models, and non-technical
solutions, not only software features, and Gothelf's own guidance calls for ideas that
can be big, small, innovative, or even a little unconventional. Naming a solution here
does not commit the team to building it. Deep dive: lean-ux-canvas_companion.md section
3 (Box 5).
ASK What candidate solutions, technical and non-technical, could address the user and
business outcomes above? Have you listed more than one, or stopped at the first idea?
Does at least one candidate not involve building new software?
GOOD "An automated SMS check-in; a geofence that logs arrival automatically; or,
non-technically, a shared paper log the depot's shift lead photographs each morning."
(more than one candidate, including a non-technical option)
WEAK "Build an automated check-in app."
(a single idea; the box's own widening to several, non-technical candidates goes unused)
TRAP Stopping at one solution. A single idea in this box undercuts the box's own 2019
widening and is one of six pitfalls a community skill file names. -->
{{candidate_solutions}}
## Hypotheses
<!-- WHAT A table, one row per candidate hypothesis, each combining the business outcome, the
user, the benefit, and the feature into a single sentence.
WHY This box combines Boxes 2 through 5 into one sentence carrying four slots: the business
outcome, the user, the benefit the user attains, and the feature or solution that
attains it. Each hypothesis should address exactly one feature; a hypothesis spanning
several cannot be tested cleanly. Deep dive: lean-ux-canvas_companion.md section 3 (Box
6) and section 7 (Anti-patterns and failure modes).
ASK For each candidate solution, what single hypothesis sentence links it to one business
outcome and one user benefit? Does each row name exactly one feature, or has more than
one been bundled into the same sentence?
PRIORITY One row per feature, never one row bundling several features into a single sentence;
a bundled hypothesis cannot be tested cleanly. Rows are candidates, not ranked by this
table; Box 7 is where the team picks one to test first.
ROW HINT A good row names the business outcome, the user, the benefit, and exactly one
feature. A weak row omits a slot, or bundles two features into one sentence.
GOOD | H-01 | We believe 80% fewer manual confirmation calls will follow if dispatchers get confidence in driver status from an automated SMS check-in. |
WEAK | H-01 | We believe check-ins will improve if we build the SMS feature and the geofence feature. |
(two features bundled into one sentence that cannot be tested cleanly)
TRAP Bundling more than one feature into a single hypothesis. The canvas's own rule is one
feature per hypothesis, and a sentence spanning several features cannot be tested
cleanly. -->
| ID | Hypothesis (business outcome, user, benefit, one feature) |
|---|---|
| {{hypothesis_id}} | We believe {{hypothesis_business_outcome}} will follow if {{hypothesis_user}} gets {{hypothesis_benefit}} from {{hypothesis_feature}}. |
## What We Need to Learn First
<!-- WHAT One hypothesis chosen from the table above, the ways it might fail, and the single
riskiest assumption behind it. Not a list of everything that could go wrong.
WHY Gothelf is pointed about scope here: the job is not to list every risk a hypothesis
carries, but to identify the one thing that matters most to learn right now. In a
hypothesis's early stages, the guidance favors risks to value over risks to feasibility.
This box is revisited, not answered once: after an experiment from Box 8 returns, the
team asks this box's question again. Deep dive: lean-ux-canvas_companion.md section 3
(Box 7).
ASK Of all the ways this hypothesis could fail, which one matters most to learn about right
now? Is this risk about whether users value the outcome, or only about whether the team
can build it? After the last experiment returned, has this answer changed?
GOOD "The riskiest assumption behind the SMS check-in hypothesis is that dispatchers would
actually trust a text reply over a phone call, not whether the SMS platform is
technically feasible."
(one named assumption, about value rather than feasibility)
WEAK "There are a lot of risks: cost, timeline, driver adoption, SMS reliability, and whether
dispatchers like it."
(a list, not a single named priority)
TRAP Listing every risk a hypothesis carries instead of naming the one thing most important
to learn right now. That is exactly the shortcut Gothelf's own guidance warns against. -->
{{riskiest_assumption}}
## The Least Work to Learn It
<!-- WHAT The smallest experiment that would show whether the riskiest assumption named above is
true or false. Not a smaller version of the finished product.
WHY "MVP" in this context means a small and fast way of learning something, not a
stripped-down release meant to ship. This box sketches the shape of a test; it does not
plan one in the detail an experiment-design-doc would. Deep dive:
lean-ux-canvas_companion.md section 3 (Box 8) and section 8 (Relationships to other
artifacts).
ASK What is the cheapest, fastest way to learn whether the assumption above holds? Does this
design actually test the assumption, or does it quietly become a smaller version of the
final feature? If this needs a full test plan with variants and a decision rule, does
experiment-design-doc need to take over from here?
GOOD "Ask five dispatchers to try replying to one manual SMS check-in for a week, with no app
built, and track how many replies come back and whether dispatchers say they would
trust it."
(a fast way of learning, not a smaller release)
WEAK "Build a stripped-down version of the SMS app with only the core check-in feature."
(a smaller release, not a fast way of learning)
TRAP Designing a smaller version of the eventual product instead of the fastest way to learn
the answer. The originators are explicit that MVP here means a way of learning, not a
release. -->
{{smallest_experiment}}
## Revisions and What Takes Over
<!-- WHAT A dated row for each pass through the two boxes above, recording what was learned,
whether the assumption held, and which boxes changed as a result. Then, once the canvas
stops changing, the document that takes over.
WHY This section is this library's own addition. Its dated, iterative half is sourced: the
canvas's own header carries date and iteration fields, and a community skill file treats
the canvas explicitly as not a one-time exercise, to be revisited as the team learns.
The successor half is this library's POSITION, not sourced: no source this research read
names which document takes over once a canvas has served its purpose. This library names
one, informed by one named consultancy's account of the canvas being abandoned in favor
of other tools once a project reached enough definition, together with this family's own
successor obligation. Deep dive: lean-ux-canvas_companion.md section 3 (Box 9) and
section 8 (Relationships to other artifacts).
ASK What did the last experiment tell you, and did the assumption it tested hold?
Which boxes did you change because of it, and what does Box 7 name now?
Has the canvas stopped changing, and if so, did the team commit, change course or stop?
PRIORITY Add one row every time the team loops back through the two boxes above; do not
overwrite an earlier row. Fill "What takes over" only once the canvas stops changing, not
on a first pass.
ROW HINT A good row names the date, which assumption was tested, whether it held, and which
boxes changed as a result. A weak row says only "tested" with no stated outcome.
GOOD | 2026-04-10 | Tested whether dispatchers would trust an SMS reply over a phone call | Held: 5 of 5 dispatchers replied within the pilot week and said they trusted it more than expected | Box 7 now names geofence reliability as the next riskiest assumption |
WEAK | | Tested the SMS idea | Went fine | |
(no date, no clearly stated outcome, nothing named as having changed)
TRAP Leaving this section blank because a first pass feels finished. A canvas revised without
a record cannot show what the team actually learned, which is exactly why this family's
cadence exception requires one. -->
| Date | Assumption tested | What was learned | Boxes changed |
|---|---|---|---|
| {{revision_date}} | {{assumption_tested}} | {{outcome_learned}} | {{boxes_changed}} |
**What takes over when this canvas is finished**, because the team committed to building, changed
course, or stopped the initiative: {{successor_document}}.

The reasoning, the history and every source, in the repository:

  • Companion - the long-form argument: why these sections, where the sources disagree, and what the bundle refuses to claim
  • History - what changed in this bundle, and when
  • Research log - every source consulted, with what each one actually supports
  • Catalog metadata - the machine-readable record this page is generated from

Catalog record: 9 sections across 1 format(s), methodology Lean UX, typically owned by Product team (cross-functional).