AI Level Decision Matrix
Use Case Name: _________________________________
Date: _________________________________
Understanding AI Levels
Use this matrix to select the appropriate AI level for your use case. Start simple. Upgrade later if needed.
Level 0: This Doesn't Need AI
What It Is:
- Rule-based logic
- Simple automation
- Heuristics
- If-then-else statements
Examples:
- "If transaction amount > €5,000, flag for review"
- "Route ticket to team based on category"
- "Send email when status changes"
- "Calculate discount based on customer tier"
When to Use:
- ✅ Problem is deterministic (same input → same output)
- ✅ Rules can capture all cases
- ✅ No pattern recognition needed
- ✅ Logic is straightforward
When NOT to Use:
- ❌ Problem is probabilistic
- ❌ Patterns are complex
- ❌ Rules can't capture all cases
Cost: €5K-20K
Time: 2 weeks
Complexity: Low
Decision Criteria:
- Problem is deterministic
- Rules can solve 100% of cases
- No ML/AI expertise needed
→ If all checked: Use Level 0. Stop here.
Level 1: Traditional Data Analytics / BI
What It Is:
- Statistical models
- Regression analysis
- Basic analytics
- Time series forecasting
- Clustering
Examples:
- Sales forecasting with linear regression
- Customer segmentation with k-means
- Trend analysis with time series
- A/B test analysis
- Correlation analysis
When to Use:
- ✅ Linear relationships in data
- ✅ Historical patterns predict future
- ✅ No real-time decisions needed
- ✅ Statistical analysis sufficient
When NOT to Use:
- ❌ Non-linear relationships
- ❌ Real-time decisions required
- ❌ Complex pattern recognition needed
Cost: €20K-50K
Time: 3 weeks
Complexity: Low-Medium
Decision Criteria:
- Linear relationships exist
- Historical data predicts future
- Statistical models sufficient
- No real-time requirements
→ If all checked: Use Level 1.
Level 2: AI-Supported (Human Decision-Making Enhanced)
What It Is:
- AI suggests, human decides
- Human-in-the-loop (HITL)
- AI improves efficiency but doesn't replace humans
- Augmented intelligence
Examples:
- AI recommends products, human approves
- AI flags anomalies, human investigates
- AI generates content, human edits
- AI suggests code changes, developer reviews
- AI identifies at-risk customers, retention team decides offers
When to Use:
- ✅ Human judgment is critical
- ✅ AI improves efficiency but can't replace humans
- ✅ Error tolerance is low
- ✅ Decisions need human oversight
When NOT to Use:
- ❌ Decisions are routine and well-defined
- ❌ High volume requires full automation
- ❌ Human oversight is not feasible
Cost: €50K-150K
Time: 4-6 weeks
Complexity: Medium
Decision Criteria:
- Human judgment is essential
- AI supports but doesn't replace humans
- Error tolerance is low
- Decisions need human review
→ If all checked: Use Level 2. (Default recommendation for most use cases)
Level 3: AI-Integrated (Automated Decision-Making in Workflow)
What It Is:
- AI makes decisions, embedded in business process
- Automated decision-making
- High automation rate (90%+)
- Human review only for edge cases
Examples:
- Automated fraud detection (approve/decline)
- Automated document classification
- Automated content moderation
- Automated ticket routing
- Automated code review (with human override)
When to Use:
- ✅ Decisions are routine and well-defined
- ✅ Error tolerance is acceptable
- ✅ High volume requires automation
- ✅ Human review only for exceptions
When NOT to Use:
- ❌ Decisions are complex and require judgment
- ❌ Error tolerance is very low
- ❌ Human oversight is required for all decisions
Cost: €150K-400K
Time: 8-12 weeks
Complexity: Medium-High
Decision Criteria:
- Decisions are routine
- Error tolerance is acceptable
- High volume requires automation
- Human review only for edge cases (5-10%)
→ If all checked: Use Level 3.
Level 4: Advanced ML (Real-Time, Complex Patterns)
What It Is:
- Deep learning
- Ensemble models
- Real-time inference
- Complex pattern recognition
- Adaptive systems
Examples:
- Real-time fraud detection with ensemble models
- Computer vision for quality control
- Natural language understanding for complex queries
- Recommendation systems with deep learning
- Predictive maintenance with neural networks
When to Use:
- ✅ Patterns are complex and evolving
- ✅ Real-time decisions required
- ✅ Traditional ML insufficient
- ✅ Large-scale data available
When NOT to Use:
- ❌ Simple patterns can be captured with Level 2-3
- ❌ Real-time not required
- ❌ Limited data available
- ❌ Cost/complexity not justified
Cost: €400K-1M+
Time: 12-24 weeks
Complexity: High
Decision Criteria:
- Complex patterns that evolve over time
- Real-time inference required (<100ms latency)
- Traditional ML insufficient
- Large-scale labeled data available (100K+ records)
- ROI justifies complexity and cost
→ If all checked: Use Level 4. (Requires strong justification)
Level 5: Agentic AI (Autonomous Task Execution)
What It Is:
- AI agents that plan, execute, and adapt autonomously
- Multi-agent systems
- Self-optimizing systems
- Autonomous decision-making with minimal human oversight
Examples:
- Multi-agent systems for complex workflows
- Autonomous research agents
- Self-optimizing trading systems
- Autonomous customer service agents
- Multi-agent orchestration for complex tasks
When to Use:
- ✅ Tasks require planning and multi-step execution
- ✅ System needs to adapt to new situations
- ✅ Human oversight acceptable but not required for each decision
- ✅ Complex workflows with multiple agents
When NOT to Use:
- ❌ Single-step decisions
- ❌ Human oversight required for all decisions
- ❌ Cost/complexity not justified
- ❌ Risk tolerance is low
Cost: €1M+
Time: 6-24 months
Complexity: Very High
Requirements for Level 5:
- Human-in-the-loop (HITL) for critical decisions
- Comprehensive audit logging
- Kill switch for emergency shutdown
- Extensive testing and validation
- Risk assessment completed
- Executive approval obtained
Decision Criteria:
- Multi-step planning and execution required
- System must adapt autonomously
- Human oversight acceptable but not required per decision
- ROI justifies €1M+ investment
- All requirements above are met
→ If all checked: Use Level 5. (Requires extensive justification and approval)
Decision Flow
Start Here:
Is the problem deterministic? (Same input → same output)
- YES → Level 0 (No AI needed)
- NO → Continue
Are linear relationships sufficient? (Statistical models work)
- YES → Level 1 (Analytics/BI)
- NO → Continue
Is human judgment critical? (AI supports, doesn't replace)
- YES → Level 2 (AI-Supported) ← Default recommendation
- NO → Continue
Are decisions routine and well-defined? (High automation acceptable)
- YES → Level 3 (AI-Integrated)
- NO → Continue
Are patterns complex and evolving? (Real-time, deep learning needed)
- YES → Level 4 (Advanced ML)
- NO → Continue
Does it require autonomous planning and execution? (Multi-agent systems)
- YES → Level 5 (Agentic AI)
- NO → Re-evaluate
Your Use Case Assessment
Problem Description:
Selected Level: Level _____ (0 / 1 / 2 / 3 / 4 / 5)
Why this level?
Estimated Cost: €_____
Estimated Timeline: _____ weeks/months
Can we start simpler and upgrade later?
- Yes, start at Level _____
- No, need this level from day one
Justification:
Key Principles
- Start Simple - Most problems don't need Level 4 or 5
- Upgrade Later - You can always increase complexity if needed
- Match Complexity - Don't over-engineer
- Default to Level 2-3 - AI-Supported or AI-Integrated for most use cases
- Justify Higher Levels - Level 4-5 require strong justification and ROI
Approval
Recommended by: _________________ Date: _______
Approved by: _________________ Date: _______
AI Architect: _________________ Date: _______