| name | product-ai-fit |
| description | Decide whether a product problem should use AI and define the appropriate human-AI workflow. Use when a team starts with a model or agent idea, must compare AI with deterministic software or process change, needs an AI use-case brief, or must assess data, uncertainty, value, feasibility, cost, and fallback fit. |
Product AI Fit
Start with a user decision or workflow that benefits from interpretation, generation, prediction, ranking, or adaptive behavior. Recommend AI only when it creates enough value to justify uncertainty and operating cost.
Inputs
- User, job, workflow, and current alternative
- Required output or decision
- Tolerance for error, latency, variability, and explanation
- Available data, feedback, and permissions
- Expected frequency, value, and cost
- Human expertise and escalation capacity
Workflow
- Define the user outcome and the part of the workflow that is currently constrained.
- Compare four approaches: no change, process or policy change, deterministic software, and AI-assisted or AI-automated behavior.
- Identify the AI capability required: classify, predict, retrieve, rank, generate, perceive, or plan.
- Determine whether examples, labels, context, feedback, and evaluation data exist or can be obtained responsibly.
- Analyze the cost and consequence of wrong, delayed, inconsistent, or manipulated outputs.
- Choose the human role: author, reviewer, approver, exception handler, or recipient.
- Define confidence handling, fallback, explanation, correction, and user control.
- Estimate value per successful task, cost per task, latency budget, and operational burden.
- Recommend AI, deterministic software, hybrid workflow, or no build, with the smallest validating prototype.
Output contract
Return workflow and problem, alternatives, AI capability, value mechanism, data requirements, error-cost analysis, human role, fallback, economics, recommendation, prototype, and decision rule.
Quality gate
- The problem exists without mentioning AI.
- AI provides a specific advantage over deterministic alternatives.
- Uncertainty is visible in the user experience and operating model.
- Evaluation and feedback data are plausible.
- Cost, latency, privacy, and failure recovery are included.
Avoid
- Choosing a model before defining the task
- Using AI where rules are adequate and safer
- Assuming natural interaction means reliable behavior
- Automating high-impact decisions without meaningful oversight
- Ignoring the cost of review and exceptions
Source grounding
Operational synthesis informed by AI use-case, workflow, uncertainty, and lifecycle concepts in Building AI-Powered Products and production trade-offs in Designing Machine Learning Systems.
