Identifying AI Opportunities

Not every feature needs AI. The best PMs know where AI creates genuine value vs. where it's just hype.

The AI Opportunity Framework

Ask these questions to evaluate potential AI features:

1. Is There a Pattern to Learn?

AI excels at finding patterns humans can't see or can't process at scale.

Good AI candidates:

• Fraud detection (patterns in transaction behavior)
• Content recommendations (patterns in user preferences)
• Demand forecasting (patterns in historical sales)

Poor AI candidates:

• Static configuration settings
• Simple if/then business rules
• One-time data migrations

2. Do You Have Enough Data?

AI needs data to learn. Rough minimums:

AI Task	Minimum Data
Text classification	1,000+ labeled examples
Image classification	5,000+ labeled images
Recommendation system	100,000+ user interactions
Custom LLM fine-tuning	10,000+ examples

No data yet? Consider:

• Starting with rule-based logic
• Using pre-trained models (GPT-4, Claude)
• Collecting data before building AI

3. What's the Cost of Errors?

AI makes mistakes. Can your use case tolerate them?

Error Tolerance	Examples	AI Approach
High tolerance	Content recommendations, search ranking	Deploy freely, iterate
Medium tolerance	Customer support, content moderation	Human review for edge cases
Low tolerance	Medical diagnosis, legal advice	Human-in-the-loop required

4. Is Scale a Problem?

AI shines when human review doesn't scale.

Scale indicators:

• Reviewing 10,000+ items/day manually
• Personalizing for millions of users
• Real-time decisions needed

No scale problem? Human review might be better.

The Value vs. Effort Matrix

Plot potential AI features on this matrix:

High Value │  Quick Wins    │   Strategic Bets
           │  (Do First)    │   (Plan Carefully)
           │                │
───────────┼────────────────┼───────────────────
           │                │
Low Value  │  Avoid         │   Avoid
           │                │
           └────────────────┴───────────────────
                Low Effort     High Effort

Common AI Opportunity Categories

Category 1: Content Intelligence

Opportunity	Business Value	Complexity
Summarization	Save reading time	Low
Classification	Organize at scale	Low
Search enhancement	Better discovery	Medium
Content generation	Scale content creation	Medium

Category 2: User Understanding

Opportunity	Business Value	Complexity
Segmentation	Better targeting	Low
Churn prediction	Retention focus	Medium
Recommendation	Increased engagement	Medium
Personalization	Improved conversion	High

Category 3: Process Automation

Opportunity	Business Value	Complexity
Data extraction	Reduce manual entry	Low
Routing/Triage	Faster response	Medium
Quality assurance	Consistent standards	Medium
Decision support	Better outcomes	High

Opportunity Scoring Template

Rate each potential AI feature (1-5):

Criteria	Score	Notes
User impact	_/5	How much does this help users?
Data availability	_/5	Do we have the data?
Technical feasibility	_/5	Can we build this?
Error tolerance	_/5	Can we accept AI mistakes?
Scale benefit	_/5	Does AI scale better than alternatives?
Total	_/25	Prioritize features scoring 18+

Red Flags: When to Say No

Decline AI projects when:

• "We want AI because competitors have it" (no clear problem)
• "Let's see what AI can do with our data" (solution looking for problem)
• "We need 100% accuracy" (AI can't guarantee this)
• "We have no data but want to start anyway" (build data collection first)

Key Takeaway

AI opportunities exist where patterns in data can solve user problems at scale. Use this framework to separate genuine opportunities from AI hype.

---

Next: Once you've identified an opportunity, how do you write requirements for AI features? :::