AI product strategy is about making good bets: which capabilities matter to your users, how to deliver them reliably, how to price them and how to avoid building features the next model release makes obsolete.
Questions a strategy must answer
Which user jobs does AI genuinely improve? What is our proprietary advantage — data, workflow, distribution, trust? What quality bar is acceptable and how will we measure it? What does it cost per use and how do we price it? What happens when the underlying models improve?
Feature patterns that hold up
Features embedded in an existing workflow; features that use data only you have; features that make experts faster rather than replacing them; features with transparent, correctable outputs.
Feature patterns that erode
Thin wrappers over a model with generic prompts; features competing directly with a model provider's own product; anything whose only moat is being first.
Roadmapping under uncertainty
Plan in short cycles, keep the architecture provider-agnostic, invest in evaluation data (it compounds), and revisit assumptions quarterly.
My perspective
I encourage founders to think of AI as a capability layer, not a category. The question is never "should we add AI" but "which user problem becomes dramatically easier if the software can read, reason or generate".
Business use cases
- AI feature roadmaps for existing software products
- Evaluating build vs buy vs API for a capability
- Pricing and packaging AI features
- Defining quality bars and evaluation sets
- Positioning an AI product against fast-moving competitors
Limitations to be honest about
- Strategy is only as good as the user research behind it.
- Model capabilities shift; long fixed roadmaps age quickly.
- Competitive analysis in AI is noisy — many products are demos.
How I approach it
- A structured review of users, workflows, data assets and competitors, followed by a ranked list of AI opportunities with cost, quality and defensibility estimates and a 90-day plan.
Frequently asked questions
How do I decide which AI features to build first?
Rank by user value, data availability, quality risk and defensibility. Start where value is high and quality is measurable.
Should we build our own models?
Rarely at the start. Use APIs, invest in evaluation and data, and revisit once you have measured gaps.
How do we avoid being made obsolete by the next model release?
Build on workflow, data and integration advantages rather than raw model capability.
Last updated 11 September 2026