AI Opportunity Canvas

Project identity

Project name — short descriptive name

Owner / team — person or team responsible

Stakeholders — primary business owner(s), users, compliance/security contacts

Problem statement

Describe the specific problem you intend to solve and for whom. Focus on observable pain or measurable waste (customer friction, manual effort, error rates, throughput, cost, safety incidents).

Prompt: Who experiences the problem, what happens today, and what is the negative impact?

Example: Customer support spends 30% of time summarizing call notes manually, causing delays and inconsistent records.

User value & business impact

Explain how solving the problem will benefit users and the business. Be concrete: time saved, revenue preserved, conversion lift, error reduction, compliance improvement, or risk avoided.

Include target impact and how you will measure it.

Example metric: reduce average handling time by 20%, saving X hours/week and improving NPS by Y points.

Proposed AI approach

Give a short description of the AI method or capability you expect to use (classification, extraction, recommendation, generative assistant, forecasting, anomaly detection). Note whether you plan to use off-the-shelf models, fine-tuning, custom models, or an internal rules + ML hybrid.

Data sources & quality

List primary data sources, approximate volume, recency, and known quality issues. Note access constraints (privacy, PII, licensing) and whether labeled training data already exists.

Quick checklist: accessible storage, sample size, labeling status, schema stability, legal/consent flags.

Feasibility risks & dependencies

Describe technical, operational, and organizational risks: integration complexity, latency requirements, model explainability needs, data pipeline maturity, or need for retraining infrastructure. List major dependencies and estimated effort to remove each blocker.

Safety, compliance & ethical considerations

Record potential harms, fairness concerns, privacy issues, regulatory constraints, and required sign-offs. Identify mitigation ideas such as human-in-the-loop, guardrails, explainability features, or withholding sensitive fields.

Success metrics (leading & lagging)

Define 2–4 measurable indicators that will determine whether the project is worth scaling. Include both quality/accuracy metrics and business KPIs.

Examples: precision/recall for extraction tasks, % automation achieved, downstream conversion lift, cycle time reduction, customer satisfaction change.

Quick prioritization scores

Score each on a 1–5 scale (1 = low, 5 = high). Use these to prioritize and to highlight areas needing attention.

• Estimated user value — impact on users/business
• Data readiness — available labeled data, quality, and access
• Technical feasibility — likelihood of a working prototype within a short sprint
• Safety / compliance risk — potential for harm or regulatory barriers (higher score = greater risk)

Interpretation tip: Favor ideas with high user value, reasonable data readiness, and manageable safety risk. High value with very low data readiness may be a candidate for a data investment project rather than immediate modeling.

Prototyping plan

Write a focused plan that produces evidence quickly and minimizes exposure. Include a short timeline, deliverables, who will build it, what data and compute are needed, what success looks like, and acceptance criteria.

Suggested prototype parts:

• Minimum viable experiment (demo, sample predictions, or human-assisted workflow)
• Evaluation dataset and metrics to measure performance
• Integration scope for the prototype (sandbox only vs limited live testing)
• Timebox (e.g., a few sprints or X weeks) and budget constraints

Next steps & decision gates

List possible outcomes and the criteria for each (stop, iterate, scale). Define who decides and what artifacts they will review (prototype results, cost estimate, compliance review).