Validation
Prototype-led discovery — use the prototype as the conversation to validate problem, concept, and usability, then decide: iterate, pivot, pause, or invest.
Purpose — prototype-led discovery
Most validation literature assumes one of two worlds: no prototype yet (The Mom Test) or finished design (traditional UX usability testing). Neither matches how we work. An Accelerator goes from executive idea to a working prototype in a day or two, then uses that prototype to learn. So we don't call this user research. We call it prototype-led discovery — and the prototype isn't the thing being validated. The prototype is the conversation. It's an instrument for learning what to invest in next.
- Entry
- A working prototype and at least one riskiest assumption written down.
- Artifact
- Evidence pack + decision recommendation.
- Gate
- Iterate, pivot, pause, or invest.
Four questions, in this order
Validation is not one activity. Each session answers a different question with a different output. Run them in this sequence — problem before concept, concept before usability, and always end with a decision.
Session shape
Activities
- Name the riskiest assumption before the session. One sentence. If the session doesn't touch this assumption, redesign the session.
- Pick the smallest audience that answers the question. 5–8 real users. Not executives, not your partners — people who have the problem today.
- Run problem before concept, every time. Ten minutes on their current world before the prototype comes out. If you reveal early you'll spend the rest of the session hearing about pixels.
- Capture behavior, not opinions. What they did, where they hesitated, what they tried that the prototype didn't support. Direct quotes beat paraphrases — that's why you record.
- Debrief the same day. Assumptions validated / disproven / new. Recommendation. Do it while it's fresh; memory rots faster than you think.
- Iterate once, at most twice. If two rounds don't move the signal, that is the evidence. Graceful stop, not a longer engagement.
Decide — the missing stage
Most UX playbooks stop at "share findings." The Accelerator's job isn't findings — it's helping the organization decide what to do next. Every session ends with a structured debrief that produces a recommendation, not a report.
- Riskiest assumption going in: [one sentence]
- What we saw: 3–5 behavioral observations, with a quote or clip per point.
- Validated: which assumptions held up, and the evidence that held them up.
- Disproven: which assumptions broke, and what actually happened instead.
- New questions: what this session surfaced that we didn't know to ask before.
- Recommendation: one of iterate (specific change), pivot (specific reframe), pause (what would restart it), or invest (what handoff needs).
- What would flip this recommendation: the single piece of evidence that would change your mind.
What "enough evidence" looks like. Greenfield work (new capability, new audience) decides on directional signal: did 4 of 6 users complete the core task; did 3 of 5 say "when can I have this"; did the same hesitation appear across sessions. Iterative work on an existing product needs quantitative lift over the current experience. Don't hold a greenfield bet to an iterative bar — you'll kill things that deserved another round. And don't hold an iterative change to a greenfield bar — you'll ship on vibes.
Questions worth asking (and ones to avoid)
- "Walk me through the last time you dealt with [problem]."
- "What's the hardest part of your current process?"
- "What were you expecting to happen when you clicked that?"
- "What surprised you?"
- "Where does this not fit your workflow?"
- "What would you do next if I wasn't here?"
- "Do you like it?" — everyone says yes to your face.
- "Would you use this?" — self-prediction is unreliable.
- "Would you pay for it?" — cheap talk unless money moves.
- "What do you think?" — invites polite fiction.
- Anything that starts with "Wouldn't it be cool if…"
Common failure modes
Prompt library
Paste these into a Lovable prototype chat to generate a first draft of research questions, tasks, or a debrief. Fill in the bracketed context first, then edit the output down — these are starting points, not finished scripts. One prompt per stage.
You are helping an Innovation Accelerator design the first 10-15 minutes of a moderated session — BEFORE the prototype is revealed. Goal: understand the participant's current world, not gather requirements. Context: - Problem we think we're solving: [problem] - Target user (role, context, current workflow): [user] - The riskiest assumption about the problem itself: [assumption] Produce: 1. Five behavioral warm-up questions grounded in the LAST time the user encountered this problem — no hypotheticals, no "would you", no "do you like". 2. Three "walk me through" prompts that get the user narrating their current process. 3. Two questions that surface cost of the current workaround (time, money, frustration, missed opportunity). 4. One question that reveals who else has this problem and how they solve it today. Flag any question that is hypothetical, leading, or asks the user to predict future behavior — rewrite each into a behavioral form.
You are helping an Innovation Accelerator design the concept-reveal portion of a 45-minute session. The prototype is shown as "one possible way of approaching the problem" — the goal is to learn whether the approach makes sense, not whether the pixels are pretty. Context: - Problem: [problem] - Prototype approach in one sentence: [approach] - The riskiest assumption about the approach: [assumption] - What we're specifically NOT trying to validate yet: [out of scope] Produce: 1. A 2-sentence framing to say when revealing the prototype (non-defensive, invites critique). 2. Six open reaction questions covering: first impression, what surprised them, what feels unrealistic, what's missing, where it doesn't fit their workflow, and what a colleague would say. 3. Two probes for behavioral commitment — would they show a teammate, would they give up something they use today. 4. Two "kill" questions — what would make this NOT worth doing for them. Do not include "do you like it", "would you use it", or any question that invites polite agreement.
Help me design the usability portion of a session (or an unmoderated Maze test) for the prototype below. Tasks only — no interviewing during this phase.
Prototype: [link or short description of flows available]
Primary user goal: [goal]
Success criteria: [e.g. completes intake without help, finds the report in <60s]
Produce:
1. 3-5 task prompts written as user GOALS ("You need to…"), not instructions ("Click the blue button"). Each task maps to a success criterion.
2. For each task: one post-task open follow-up ("what were you expecting there?") and one 1-5 confidence rating.
3. Three post-test questions: what was confusing, what was missing, what would they do next in real life.
4. A short list of observation cues — what the moderator should silently watch for (hesitation, misclicks, verbal confusion, workarounds).
Call out any task where the prototype likely cannot support the path, or where success is ambiguous.You are helping an Innovation Accelerator turn a completed prototype-led discovery session into a one-page decision recommendation. The output is NOT a research report — it's a recommendation the sponsor can act on. Inputs I'll paste: - Riskiest assumption we set out to test: [assumption] - What we saw (raw notes, quotes, observations): [notes] - Behavioral signals (completed tasks, shared it, asked when they could have it, refused, hesitated): [signals] - Whether this is greenfield or iterative work: [greenfield / iterative] Produce a one-page recommendation with these sections: 1. Riskiest assumption going in. 2. What we saw — 3-5 behavioral observations, each with a quote or clip reference. 3. Validated — which assumptions held up, and the evidence. 4. Disproven — which assumptions broke, and what happened instead. 5. New questions — what this session surfaced that we didn't know to ask. 6. Recommendation — exactly one of: ITERATE (specific change), PIVOT (specific reframe), PAUSE (what would restart it), INVEST (what handoff needs). 7. What would flip this recommendation — the single piece of evidence that would change your mind. 8. Feasibility / viability flags for the Accelerator to chase separately — do NOT treat as user-answered. Be blunt. If evidence is thin, say thin. If the recommendation is PAUSE, don't soften it.
Supporting tools
- Maze for structured unmoderated concept + usability tests at scale.
- Microsoft Clarity dropped into every shared prototype for session replay and rage-click detection.
- markup.io for external users who need to comment without an account.
- NPS panel on Metrics to capture willingness signals in the moment.