Not every problem needs AI, and not every AI application is a good product. The best AI products solve real problems where AI provides a step-change improvement over existing solutions.

The AI Product Opportunity Framework

Ask these questions to evaluate an AI product idea:

1. Is there a real problem?

• Who has this problem? How often? How painful is it?
• What do they currently do about it? (The "status quo" is your real competitor)
• Would they pay to solve it? How much?

2. Does AI meaningfully improve the solution?

• Is the task repetitive, data-heavy, or requiring pattern recognition?
• Does AI provide 10x improvement, not just 10% improvement?
• Could this be solved with traditional software? (If so, AI adds complexity without benefit)

3. Is the problem bounded enough?

• Can you define clear success criteria?
• Is the domain narrow enough for current AI capabilities?
• Are errors acceptable or catastrophic? (Healthcare diagnosis vs. email drafting)

Finding Opportunities

Workflow automation: Look for tasks that knowledge workers do repeatedly:

• Summarizing documents, emails, or meetings
• Categorizing and routing requests
• Drafting standard communications
• Extracting data from unstructured sources

Augmentation over automation: The best AI products enhance human capability rather than replace it:

• Writing assistants that suggest, not write
• Analysis tools that surface patterns for human decision-making
• Creative tools that generate options for human curation

The "Last Mile" problem: Many industries have manual "last mile" processes that AI can address:

• A lawyer reviews contracts → AI highlights risky clauses for human review
• A doctor reads medical images → AI flags anomalies for doctor confirmation
• A recruiter screens resumes → AI surfaces top candidates for human interview

Common Pitfalls

• Solution looking for a problem — "We have AI, what should we build?" (backwards)
• Demo magic — Looks amazing in a demo but fails on real-world data
• Accuracy assumptions — Assuming AI will be 99% accurate when it's actually 85%
• Ignoring the human — Building fully autonomous systems when human oversight is needed