Converge on concepts with a desirability-feasibility-responsibility review
Turn an AI-expanded concept pool into a human decision by independently assessing desirability, feasibility, viability, and responsibility.
Why
The workflow problem
AI-supported divergence leaves too many plausible concepts and can hide delivery constraints or harms behind polished descriptions.
A structured review makes trade-offs and missing evidence visible before resources are committed. AI can reduce administrative effort, but scoring cannot turn unknowns into facts or make responsibility an averageable afterthought.
Concrete output
What you will produce
A scored concept decision matrix with evidence links, criterion definitions, individual scores, rationale, dissent, and a next-test recommendation.
How
Run the workflow
Define decision criteria and thresholds
Humans define desirability, feasibility, viability, and responsibility for the decision at hand, including non-negotiable thresholds and required evidence. An accountable sponsor approves weights before concepts are scored.
Prepare evidence-linked concept cards
A human owner supplies each concept's evidence IDs, assumptions, operational dependencies, and known harms. AI may format the cards and list absent fields, but may not fill missing evidence.
Find missing decision fields
Review the concept cards below for completeness against the stated criteria. Criteria and non-negotiable thresholds: [insert] Concept cards: [paste] Return only: concept | missing evidence | ambiguous claim | operational dependency not stated | responsibility risk requiring human review. Do not score, infer facts, select a winner, or rewrite uncertainty as confidence.Score independently
Relevant human decision-makers score each concept independently using the approved definitions and cite their reasoning. Keep blank cells blank when evidence is missing.
Discuss variance and dissent
Compare score differences, inspect the underlying evidence, and record dissent or unresolved risk. Do not collapse a high-consequence responsibility concern into an average score.
Choose a bounded next move
Humans select concepts to test, revise, hold, or stop. The decision record names the owner, learning question, required safeguard, and review date.
Human–AI partnership
Who contributes what
AI contribution
Format evidence-linked cards, flag missing fields, and summarise score variance. It does not provide expertise, assign values, or select a winner.
Human responsibility
Define criteria and thresholds, supply expertise and evidence, assess trade-offs, protect non-negotiable concerns, resolve disagreement, and own the decision.
Stop rule
Stop AI use when missing evidence becomes a confident score, criterion weights are not human-approved, responsibility is treated as averageable, or the tool is asked to choose the winner.
Check before use
Watch out
Decision matrices can launder subjective assumptions through numbers. Publish definitions, distinguish evidence from forecast, retain dissent, and check who bears consequences that decision-makers may not experience.
Evidence base
What supports this workflow
A named one-day company workshop and an AI-opportunity card-deck practice source support structured concept review, but neither independently demonstrates better outcomes. This is a conservative convergence workflow that should be evaluated in context.
- Jarek and Jurkiewicz · AI Strategy as the Foundation of Service Design
Harvested named-company case for pressure-testing AI concepts in a one-day canvas workshop; productivity claims are self-reported.
- IDEO.org · Determine What to Prototype
Canonical source for selecting concepts and deciding what merits a prototype or test.