AI Product Strategy

Find the AI opportunities worth pursuing, and the ones that will still be defensible in two years.

What this skill is

A structured workflow for deciding where Artificial Intelligence (AI) belongs in a product and how to deliver it. It moves from opportunity discovery through build-buy-partner analysis, data moat assessment, and trust design to a phased roadmap with explicit risks. It draws on Marty Cagan's four product risks (value, usability, feasibility, viability), Andrew Ng's AI Transformation Playbook, Benedict Evans's writing on AI moats, and Julie Zhuo's approach to product strategy.

What it solves

• AI features chosen because competitors announced them, not because they solve a user problem
• Build decisions made without comparing the cost and speed of buying or partnering
• Strategies that assume a data advantage without testing whether the data is proprietary or replicable
• AI features that fail user trust because errors, confidence, and control were never designed
• Roadmaps that jump to differentiation before a minimum viable version has proven value
• Risk reviews that cover model accuracy but ignore viability, cost, and regulatory exposure

When to invoke

• Leadership asks "what is our AI strategy?" and the team needs a structured answer
• A team is deciding whether to build an AI capability in-house, buy a vendor solution, or partner
• A product manager needs to judge whether the company's data creates a lasting advantage
• Planning the next two to four quarters of AI features
• Evaluating AI vendors or model providers for a specific use case

Phase 1: Product and market baseline

Capture the context the strategy depends on. Ask for anything missing.

Phase 2: Identify AI opportunity areas

Generate candidates across six categories, then score each.

Score each opportunity 1-5 on user value, frequency of the job, data availability, technical feasibility, and tolerance for error. Shortlist the top three to five.

Phase 3: Build, buy, or partner

For each shortlisted opportunity:

Record the recommendation and the condition that would flip it.

Phase 4: Data moat assessment

Test whether the data advantage is real.

Rate the moat as strong, moderate, or weak and say what would strengthen it.

Phase 5: User experience and trust design

Specify how each AI feature earns and keeps trust:

• Transparency - show why the system produced a result and how confident it is
• User control - let users edit, override, undo, or opt out
• Graceful failure - define what happens when the model is wrong or unavailable
• Progressive disclosure - start with assistive suggestions before autonomous actions
• Trust building - expand autonomy only as measured accuracy and user acceptance rise

Phase 6: Implementation roadmap

Phase 7: Risk and mitigation planning

Assess each opportunity against Cagan's four risks plus AI-specific exposure.

Output

• Product and market baseline
• Scored AI opportunity map with a shortlist
• Build, buy, or partner recommendation per shortlisted opportunity, with flip conditions
• Data moat rating and strengthening actions
• Trust and user experience requirements per feature
• Three-stage roadmap with exit criteria
• Risk register with mitigations and owners

Operating rules

Always

• Start from a user job and pain point, not from a model capability
• Compare build, buy, and partner explicitly for every shortlisted opportunity
• Test the data moat against replicability, not just volume
• Design failure modes and user control before launch
• Include inference cost in the viability assessment

Never

• Recommend an AI feature because a competitor shipped one
• Claim a data moat without evidence that the data is proprietary
• Skip the minimum viable stage and jump straight to differentiation
• Give an AI feature autonomy before its accuracy is measured
• Treat responsible AI review as optional