Fragmented validation
Research, audience thinking, competitor evidence, and planning lived in separate workflows.
Validate AI · B2B AI product design
Helping founders make decisions with AI they can question.
Validate AI brings market research, competitor analysis, audience thinking, and idea validation into one workspace. My focus was not only generating a verdict. It was making the path to that verdict visible, correctable, and useful.
- Role
- Product Designer
- Ownership
- Research, UX, UI, system design
- Audience
- Early stage founders and product teams
Original project artwork from Behance01 / Original problem
Founders were making expensive decisions with incomplete evidence.
The existing project statement identified a common early stage problem: assumptions are often treated as proof. Research is spread across multiple tools, competitor checks are inconsistent, and a confident looking AI answer can hide how little context supports it.
Confidence without false certainty
The product needed to communicate uncertainty without making the AI feel vague or unhelpful.
From evidence to interface
I connected the research model, product flow, information hierarchy, screen system, and interaction states.
02 / Complete workflow
One continuous path from assumption to action.
The workflow gives every AI step a visible purpose. Founders can see what is happening, what information is missing, and what decision the output is meant to support.
Frame the idea
Capture the problem, audience, assumptions, and the decision the founder needs to make.
Add evidence
Ask for the information the AI needs instead of pretending it already has enough context.
Run analysis
Show progress across market, competitor, audience, pain point, and feasibility signals.
Review confidence
Separate evidence from inference and explain why a result is strong, weak, or incomplete.
Choose a next step
Turn the result into specific actions the founder can accept, correct, or investigate further.
03 / Research evidence
Research changed the product from a verdict generator into a decision workspace.
The original case study documents qualitative founder interviews, interview questions, competitor review, personas, an empathy map, and a task flow. I translated the recurring needs into four design requirements.
Evidence is scattered
Founders move between search, trend, competitor, and planning tools before they can form one view.
A score alone is not enough
People need to understand what influenced a verdict before they can trust it.
Confidence depends on context
The quality of an AI recommendation rises or falls with the completeness of the founder's inputs.
Results must lead somewhere
A useful validation report should make the next product decision clearer, not simply generate more reading.
04 / Early explorations
The options I moved away from were the ones that removed user control.
Recording rejected directions makes the reasoning visible. Each decision reduced ambiguity while preserving the founder's ability to inspect and correct the system.
One shot AI verdict
A single pass felt fast but concealed missing evidence and made the result look more certain than it was.
Why it failedNo reasoning, no correction path, no visible evidence quality.Fully automated next step
Automatically choosing a direction reduced effort but removed the founder from an important business decision.
Why it failedLow agency and no room to challenge assumptions.Progressive analysis
Break the verdict into visible stages, surface evidence gaps, and end with actions the founder can review.
Why it workedClear status, correction, confidence, and decision support.05 / Designing trust between users and AI
Trust is not a friendly robot icon. It is a sequence of understandable choices.
The interface communicates what the AI is doing, what it needs, when the founder can correct it, and how the result becomes an actionable decision.
AI analysisValidating the evidence behind your idea
Market signals complete · competitor context in progress
→
Needs your inputWho is the first customer you can reach?
This answer improves the audience confidence score.
→
ResultDirectional opportunity82Confidence is limited by two evidence gaps.
Show the work
Name the analysis in progress and reveal the signals behind the result.
Ask, do not assume
Make missing information visible and explain why the AI needs it.
Keep correction close
Let founders edit inputs, challenge assumptions, and rerun an analysis without losing context.
Turn insight into action
Pair confidence with practical next steps, evidence gaps, and decisions to make.
06 / Edge cases and system states
The experience also works when the AI has no answer yet.
Empty, loading, incomplete, low confidence, error, and complete states use the same structure so the interface remains predictable under uncertainty.
Start with one idea
A focused prompt and example reduce blank page anxiety.
Analyzing five signal groups
Progress labels explain what the AI is doing instead of showing an indefinite spinner.
Two evidence gaps
The interface asks for missing audience and competitor context before raising confidence.
Result is directional
A caution state separates weak evidence from a negative verdict.
Analysis paused
The user's draft stays safe and the retry action explains what will happen next.
Ready to decide
The result ends with an evidence summary and prioritised actions.
07 / Variables, tokens and reusable components
A system for consistent confidence, not just consistent colour.
Tokens connect visual language to product meaning. Component variants cover default, hover, loading, caution, error, success, and disabled behaviour for handoff.
Colour and spacing
- space.2
- 8 px
- space.4
- 16 px
- space.6
- 24 px
- radius.card
- 16 px
Confidence card
Directional48
Needs evidence62
Evidence aligned82
Action states
08 / Engineering collaboration
Handoff describes behaviour, not only dimensions.
The design system gives engineering a shared language for data states, component variants, breakpoints, and AI responses. The handoff model keeps implementation questions tied to the user's decision.
Align
Review feasibility, data availability, and the confidence model before high fidelity work.
Specify
Document component anatomy, responsive rules, content limits, and every system state.
Review
Compare builds against behaviour and accessibility criteria, then resolve exceptions in the system.
09 / Final interface design
The real product design, shown in context.
The final workspace uses a focused blue accent, generous white space, and a clear dashboard hierarchy. Each view helps founders move from an idea to evidence, analysis, pain points, and a report they can act on.
Original Figma home page design
Dashboard overview and live analysis engine
My Ideas and detailed validation report
Analysis runs, progress, and result states
Pain point evidence and complaint patterns
Reports, scores, downloads, and summaries
10 / Result and validation
The result is a decision workspace with an honest testing plan.
The designed experience unifies the founder journey across ideas, analysis, trends, competitors, pain points, reports, and settings. Because launch metrics were not supplied, the outcome is stated without invented conversion numbers.
Explain the verdict
Can a founder describe why the AI reached this confidence level?
Correct the AI
Can a founder find and update an assumption that changed the result?
Act on the result
Can a founder choose the next evidence gathering action without guidance?
Trust and task clarity
Track comprehension, correction success, time to next action, and confidence before and after explanation.
The hardest AI design problem was not producing an answer. It was helping someone understand, correct, and act on that answer.
Validate AI shows how I approach AI product design as a trust system: transparent status, evidence aware confidence, meaningful correction, reusable states, and a clear path from insight to action.
Back to selected work