← Back to Templates
Download Markdown

AI Architecture Gate – Decision Framework

Use Case Name: _________________________________

Date: _________________________________

Submitted By: _________________________________

AI Architect/Platform Owner: _________________________________

Objective

The AI Architecture Gate ensures only justified, feasible, and safe AI use cases move into implementation.

This gate is mandatory before:

  • Budget approval
  • Model selection
  • Vendor onboarding

Gate 1: Problem Validation

Every AI initiative MUST start with a problem statement WITHOUT mentioning AI.

Problem Statement (Non-AI Framing)

Business Pain:

Current Workflow:

Who Owns the Problem:

  • Name: _________________________________
  • Role: _________________________________
  • Department: _________________________________

Cost of Doing Nothing:

  • Time cost: _________________________________
  • Money cost: _________________________________
  • Error cost: _________________________________

Success Metrics

Measurable Success Metric:

  • Metric: _________________________________
  • Baseline: _________________________________
  • Target: _________________________________
  • How measured: _________________________________

Non-AI Solution Considered:

  • Rules/heuristics tested
  • Manual process improvement considered
  • Basic automation evaluated
  • Other: _________________________________

Why Non-AI Solutions Are Insufficient:

Gate 1 Decision

  • ✅ PASS - Clear business owner, measurable success metric, non-AI solution considered
  • ❌ FAIL - Missing required elements

Notes:

Gate 2: AI Necessity

Answer YES to at least 3 of the following for AI to be justified:

  • Problem is ambiguous or probabilistic
  • Rules cannot fully capture logic
  • Some error is acceptable
  • Pattern exists in historical data
  • Humans struggle to scale decisions

Count: _____ out of 5

Justification

Why rules are insufficient:

Why heuristics fail:

Why AI adds value:

Gate 2 Decision

  • ✅ PASS - At least 3 criteria met, clear justification
  • ❌ FAIL - Less than 3 criteria met, or insufficient justification

If FAIL: Use automation or rules instead of AI.

Notes:

Gate 3: AI Level Approval

Select the appropriate AI capability level:

Level Description Selected
AI-Assisted Human decides, AI suggests [ ]
AI-Integrated AI embedded in workflow [ ]
AI-Driven AI makes primary decision [ ]
Agentic AI AI plans + executes steps [ ]

Default Recommendation: AI-Integrated

Selected Level: _________________________________

Justification for Selected Level:

Agentic AI Requirements (If Level = Agentic AI)

Agentic AI requires additional safeguards:

  • Human-in-the-loop (HITL) defined for critical decisions
  • Comprehensive audit logging implemented
  • Kill switch for emergency shutdown
  • Extensive testing and validation plan
  • Risk assessment completed
  • Executive approval obtained

HITL Implementation:

Audit Logging Plan:

Kill Switch Mechanism:

Gate 3 Decision

  • ✅ PASS - Appropriate level selected, requirements met (if Agentic)
  • ❌ FAIL - Level not justified, or Agentic requirements not met

Notes:

Gate 4: Data & Compliance

Data Readiness Check

Data Source Identified:

  • Source: _________________________________
  • Location: _________________________________
  • Format: _________________________________

Data Ownership:

  • Owner: _________________________________
  • Contact: _________________________________
  • Access granted: [ ] Yes [ ] Pending [ ] No

Historical Depth:

  • Time period: _________________________________
  • Volume: _____ records
  • Update frequency: _________________________________

Data Quality:

  • Completeness: _____% (target: >85%)
  • Accuracy: _____% (target: >85%)
  • Freshness: _____ (how recent?)

Labels Available:

  • Yes, fully labeled
  • Partial labels
  • No labels (unsupervised learning)

Compliance & Security

GDPR / Regulatory Review:

  • Review completed
  • Compliance officer: _________________________________
  • Date: _________________________________
  • Status: [ ] Approved [ ] Pending [ ] Issues

Security Sign-Off:

  • Security review completed
  • Security officer: _________________________________
  • Date: _________________________________
  • Status: [ ] Approved [ ] Pending [ ] Issues

Privacy Considerations:

  • PII handling reviewed
  • Data retention policy defined
  • Access controls defined
  • Encryption requirements met

Regulatory Compliance:

  • Industry regulations reviewed (specify): _________________________________
  • Data residency requirements met
  • Audit trail requirements defined

Gate 4 Decision

  • ✅ PASS - Data source approved, compliance review done, security sign-off obtained
  • ❌ FAIL - Data issues, compliance blockers, or security concerns

Notes:

Gate 5: Risk Assessment

Error Tolerance

Error Tolerance Defined:

  • Acceptable error rate: _____%
  • Critical errors: _____% (zero tolerance)
  • Non-critical errors: _____% (acceptable)

Impact of Errors:

  • Low impact: _________________________________
  • Medium impact: _________________________________
  • High impact: _________________________________

Error Handling Strategy:

Fallback Strategy

Fallback Plan Defined:

  • Manual process fallback
  • Rule-based fallback
  • Previous system fallback
  • Other: _________________________________

Fallback Trigger Conditions:

Fallback Activation Process:

Rollback Strategy:

  • Rollback plan documented
  • Rollback tested
  • Rollback timeline: _____ hours/days

Monitoring & KPIs

Monitoring KPIs Agreed:

  • KPI 1: _________________________________ = _____
  • KPI 2: _________________________________ = _____
  • KPI 3: _________________________________ = _____

Monitoring Tools:

  • Observability platform selected
  • Dashboards defined
  • Alerting rules configured
  • Team trained on monitoring

Human-in-the-Loop (HITL) Defined:

  • HITL required: [ ] Yes [ ] No
  • HITL trigger conditions: _________________________________
  • HITL review process: _________________________________
  • HITL team assigned: _________________________________

Monitoring Frequency:

  • Real-time
  • Hourly
  • Daily
  • Weekly

Risk Assessment Summary

Risk Level:

  • Low - Minimal impact if fails
  • Medium - Moderate impact, manageable
  • High - Significant impact, requires extra safeguards

Risk Mitigation:

Gate 5 Decision

  • ✅ PASS - Error tolerance defined, fallback strategy exists, monitoring KPIs agreed
  • ❌ FAIL - Risk assessment incomplete or risks too high

Notes:

Gate Summary

Gate Status Notes
Gate 1: Problem Validation ✅ / ❌
Gate 2: AI Necessity ✅ / ❌
Gate 3: AI Level Approval ✅ / ❌
Gate 4: Data & Compliance ✅ / ❌
Gate 5: Risk Assessment ✅ / ❌

Total Gates Passed: _____ out of 5

Final Decision

Decision:

  • ✅ APPROVED TO BUILD - All gates passed, proceed to implementation
  • 🔄 REVISE & RESUBMIT - Some gates need work, resubmit after addressing issues
  • ❌ REJECTED - Non-AI solution recommended, or too many blockers

Reasoning:

Recommended Next Steps:

If APPROVED:

  • Budget approval: [ ] Pending [ ] Approved
  • Team assignment: [ ] Pending [ ] Assigned
  • Timeline: Start date: _____ | Target completion: _____

If REVISED:

  • Issues to address: _________________________________
  • Resubmission deadline: _____
  • Owner: _________________________________

If REJECTED:

  • Alternative solution recommended: _________________________________
  • Reason for rejection: _________________________________

Mandatory Outputs (If Approved)

Every approved use case must submit:

  • AI Use Case One-Pager - 1-page summary
  • Chosen AI Level - Documented in Gate 3
  • 30-Day Pilot Plan - Initial validation plan

One-Pager Summary:

  • Problem: _________________________________
  • Solution: _________________________________
  • Expected impact: _________________________________
  • Timeline: _________________________________

30-Day Pilot Plan:

  • Week 1: _________________________________
  • Week 2: _________________________________
  • Week 3: _________________________________
  • Week 4: _________________________________
  • Success criteria: _________________________________

Stop Conditions

Stop the initiative if ANY of the following occur:

  • Business value is unclear
  • Data quality is insufficient
  • Risk exceeds tolerance
  • Compliance blockers cannot be resolved
  • ROI becomes negative
  • Team cannot operationalize

If stopped, document reason:

Sign-Off

Submitted By:

  • Name: _________________ Date: _______
  • Role: _________________
  • Signature: _________________

AI Architect / Platform Owner:

  • Name: _________________ Date: _______
  • Decision: [ ] Approved [ ] Revise [ ] Rejected
  • Signature: _________________

Business Owner:

  • Name: _________________ Date: _______
  • Signature: _________________

Security/Compliance:

  • Name: _________________ Date: _______
  • Signature: _________________

Finance (If budget >€100K):

  • Name: _________________ Date: _______
  • Signature: _________________

Internal Standard Reference

This gate implements the AI Internal Standard principles:

  1. ✅ Problem First (Non-AI Framing) - Gate 1
  2. ✅ AI Suitability Gate - Gate 2
  3. ✅ AI Capability Level Selection - Gate 3
  4. ✅ Data Readiness Check - Gate 4
  5. ✅ Risk & Control - Gate 5
  6. ✅ Mandatory Output - Documented above
  7. ✅ Stop Conditions - Documented above

Need help? This template is part of the framework described in "Before You Build: A Realistic Framework for Evaluating AI Use Cases".

Download the Markdown file to fill it out, or copy the content to your preferred document editor.