Chapter 9 — Readiness & Feasibility Assessment

Overview

Assess data, technology, people/process, and regulatory readiness to reduce delivery risk and avoid costly dead-ends. This chapter provides frameworks for evaluating whether your organization is ready to execute on an AI opportunity.

Before committing significant resources to an AI initiative, organizations must honestly evaluate their readiness across multiple dimensions. A thorough feasibility assessment identifies gaps, constraints, and risks early—when they're cheapest to address.

The Four Dimensions of Readiness

A comprehensive readiness assessment examines four critical dimensions that must align for successful AI delivery:

graph TB
    A[Readiness Assessment] --> B[Data Readiness]
    A --> C[Technology Readiness]
    A --> D[People & Process]
    A --> E[Legal & Compliance]

    B --> B1[Quality: >80%]
    B --> B2[Volume: Sufficient]
    B --> B3[Access: Available]
    B --> B4[Privacy: Compliant]

    C --> C1[Infrastructure: Scalable]
    C --> C2[Platform: Deployed]
    C --> C3[Integration: Tested]
    C --> C4[Monitoring: Ready]

    D --> D1[Skills: Adequate]
    D --> D2[Process: Defined]
    D --> D3[Governance: Established]
    D --> D4[Change Mgmt: Planned]

    E --> E1[Data Rights: Secured]
    E --> E2[Regulations: Mapped]
    E --> E3[Audit Trail: Enabled]
    E --> E4[Risk: Assessed]

    style B fill:#e1f5ff
    style C fill:#fff3cd
    style D fill:#d4edda
    style E fill:#ffe1e1

1. Data Readiness

Assessment Framework:

Data Assessment Process:

flowchart TD
    A[Start Data Assessment] --> B[Catalog Data Sources]
    B --> C[Sample Data Quality]
    C --> D{Quality >80%?}
    D -->|Yes| E[Check Volume]
    D -->|No| F[Estimate Cleaning Effort]
    E --> G{Sufficient Volume?}
    G -->|Yes| H[Verify Access Rights]
    G -->|No| I[Explore Augmentation]
    F --> J[Build Quality Plan]
    H --> K{Privacy Compliant?}
    K -->|Yes| L[Data Ready - Green]
    K -->|No| M[Address Privacy Issues]
    I --> N[Data Volume - Yellow]
    J --> O[Data Quality - Yellow/Red]
    M --> P[Privacy - Red]

    style L fill:#d4edda
    style N fill:#fff3cd
    style O fill:#ffe1e1
    style P fill:#ffe1e1

Practical Steps:

1. Data Inventory: List all potential sources (internal databases, external APIs, unstructured documents)
2. Sample Analysis: Pull 1,000-10,000 records, check completeness, consistency, accuracy
3. Labeling Test: Have SMEs label 100 examples, measure inter-rater agreement (kappa >0.7)
4. Privacy Audit: Map PII fields, verify consent, identify anonymization requirements
5. Integration Test: Attempt sample joins, measure latency, test pipeline reliability

2. Technology Readiness

Infrastructure Assessment:

Technology Spike Protocol:

sequenceDiagram
    participant PM as Product Manager
    participant Eng as Engineer
    participant Infra as Infrastructure
    participant Vendor as Vendor/Platform

    PM->>Eng: Define spike objectives
    Eng->>Eng: Build minimal POC
    Eng->>Infra: Request test environment
    Infra->>Eng: Provision resources
    Eng->>Vendor: Test API integration
    Vendor-->>Eng: Return performance data
    Eng->>Eng: Measure latency/cost
    Eng->>PM: Report findings + recommendation
    PM->>PM: Update feasibility assessment

Critical Tech Spikes:

3. People & Process Readiness

Operating Model Maturity:

Process Readiness Checklist:

• Decision Rights: Clear RACI for model approval, data access, deployment
• SLAs Defined: Response time, uptime, accuracy thresholds documented
• Change Control: Process for model updates, rollback procedures in place
• Vendor Management: Contract terms, SLAs, exit clauses reviewed
• Knowledge Transfer: Documentation standards, onboarding process established
• Stakeholder Engagement: Communication plan, feedback loops defined

4. Legal & Compliance Readiness

Regulatory Landscape:

Compliance Assessment Flow:

graph TD
    A[Identify Regulations] --> B[Map Data Flows]
    B --> C[Classify Risk Level]
    C --> D{High Risk?}
    D -->|Yes| E[Enhanced Requirements]
    D -->|No| F[Standard Requirements]
    E --> G[Impact Assessment]
    G --> H[External Audit]
    F --> I[Internal Review]
    H --> J[Implement Controls]
    I --> J
    J --> K[Documentation Package]
    K --> L[Legal Sign-off]
    L --> M{Approved?}
    M -->|Yes| N[Compliant - Green]
    M -->|No| O[Remediation - Red]

    style N fill:#d4edda
    style O fill:#ffe1e1

Critical Compliance Questions:

1. Data Rights: Do we have legal basis to use this data for AI? (Consent, contract, legitimate interest)
2. Model Governance: What level of explainability is required? Are there bias/fairness requirements?
3. Auditability: Can we reconstruct model decisions for regulatory review? Retention requirements?
4. Safety & Security: What are acceptable failure modes? How do we handle adversarial inputs?

Readiness Heat Map

Create a visual dashboard showing readiness across all dimensions:

Confidence Scoring:

• High: Evidence-based assessment, verified through testing
• Medium: Reasoned estimate, some validation performed
• Low: Assumption-based, requires further investigation

Mitigation Planning

For each identified gap, develop a specific mitigation plan:

flowchart LR
    A[Identify Gap] --> B{Severity?}
    B -->|Critical| C[Block Project]
    B -->|High| D[Mitigation Required]
    B -->|Medium| E[Monitor & Plan]
    B -->|Low| F[Accept Risk]

    C --> G[Escalate to Sponsors]
    D --> H[Develop Mitigation]
    H --> I{Can Close Gap?}
    I -->|Yes| J[Execute Plan]
    I -->|No| K[Adjust Scope]
    E --> L[Set Review Milestone]
    F --> M[Document in Risk Register]

    J --> N[Track Progress]
    K --> O[Redefine Success]
    L --> N
    M --> N

Mitigation Strategies by Gap Type:

Case Study: Real-Time Personalization Assessment

Context: E-commerce company wants real-time product recommendations using LLMs to personalize descriptions based on user behavior.

Initial Vision:

• User browses → System generates personalized descriptions in real-time
• Expected: <1s page load, cost <$0.05/page view
• Scale: 100K daily users, 10 page views/session = 1M requests/day

Readiness Assessment:

Data Readiness: 🟡 Yellow

• ✅ Availability: User behavior data in clickstream (Green)
• ⚠️ Quality: Product catalog 15% missing attributes (Yellow)
• ✅ Volume: 50K products, 1M users, sufficient (Green)
• 🔴 Real-time: 2-second lag in event stream (Red for <1s target)

Technology Readiness: 🔴 Red

Latency Spike Results:

LLM inference (1K token context): 3.2s (p95)
Product catalog lookup: 0.15s (p95)
Total end-to-end: 3.5s vs. 1s target ❌

Cost Spike Results:

Average tokens: 1,200 input + 400 output
Cost: $0.002/1K input, $0.006/1K output = $0.0048/request
At 1M requests/day: $4,800/day = $144K/month ❌ (Budget: $15K/month)

People Readiness: 🟡 Yellow

• Team has 1 ML engineer, needs 2 more for real-time systems
• No experience with real-time LLM serving
• Product team trained on AI concepts (Green)

Compliance Readiness: 🟢 Green

• Personalization consent already obtained
• No new PII collected
• Existing privacy controls sufficient

Mitigation Analysis:

graph TD
    A[Real-time Latency Issue] --> B{Solutions?}
    B --> C[Faster Model]
    B --> D[Edge Deployment]
    B --> E[Architecture Change]

    C --> F[Test GPT-3.5 vs GPT-4]
    F --> G[Still 2.1s - Not Enough]

    D --> H[Edge Inference POC]
    H --> I[Complexity Too High]

    E --> J[Shift to Batch Processing]
    J --> K{Acceptable?}
    K -->|Yes| L[Pre-generate for Top 1000]

    A2[Cost Issue] --> B2{Solutions?}
    B2 --> C2[Reduce Tokens]
    B2 --> D2[Cheaper Model]
    B2 --> E2[Batch Processing]

    C2 --> F2[Optimize Prompts]
    F2 --> G2[Still $90K/month]

    E2 --> J
    J --> M[Daily Batch: $3K/month ✅]

    style M fill:#d4edda
    style G fill:#ffe1e1
    style I fill:#ffe1e1
    style G2 fill:#fff3cd

Revised Approach (Feasible):

1. Batch Recommendations: Pre-generate personalized descriptions daily for top 1000 products per user segment

   • Latency: Instantaneous (pre-computed)
   • Cost: ~$3K/month (48x cheaper)
   • Coverage: 80% of traffic (top products)

2. Phase 2 (Future): Add real-time for remaining 20% after faster models available

Key Decisions:

• Scope Adjusted: Real-time → Batch processing
• Timeline: Launch in 6 weeks vs. 12+ weeks for real-time
• Risk Reduced: Avoid expensive failure, prove value first
• Sponsor Approval: Gained with honest assessment

Outcome:

• ✅ Batch system launched successfully
• ✅ 18% increase in conversion for personalized products
• ✅ Cost stayed within budget
• ✅ Built credibility for Phase 2 expansion

Lessons Learned:

1. Early spikes prevented mistakes: Latency/cost issues found in week 2, not month 6
2. Scope flexibility is power: Adjusting ambition to capability delivered faster value
3. Evidence builds trust: Showing spike data convinced sponsors
4. Mitigation isn't failure: Finding viable alternative is success

Deliverables

1. Readiness Assessment Report

Structure:

# AI Readiness Assessment: [Project Name]

## Executive Summary
- Overall readiness: [Red/Yellow/Green]
- Key blockers: [List critical issues]
- Recommended path: [Go/Modify Scope/No-Go]

## Assessment by Dimension

### Data Readiness: [Status]
[Findings, evidence, metrics]

### Technology Readiness: [Status]
[Findings, evidence, metrics]

### People & Process Readiness: [Status]
[Findings, evidence, metrics]

### Legal & Compliance Readiness: [Status]
[Findings, evidence, metrics]

## Mitigation Plans
- [Gap 1]: [Plan, owner, timeline]
- [Gap 2]: [Plan, owner, timeline]

## Gating Criteria
Proceed when:
- [ ] All red items mitigated to yellow/green
- [ ] Critical spikes completed successfully
- [ ] Sponsor approval obtained

2. Gating Criteria Document

Gate Template:

## Phase Gate: Discovery → Build

### Must-Have (Red Gate)
- [ ] Data quality >85% in production sources
- [ ] Legal sign-off on data usage and privacy
- [ ] Key tech spike successful (latency/cost in budget)
- [ ] Core team hired (ML engineer, data scientist)

### Should-Have (Yellow Gate)
- [ ] MLOps platform deployed or approved
- [ ] Model governance process defined
- [ ] Integration architecture validated

### Decision
- **Green**: Proceed to build
- **Yellow**: Proceed with risk acknowledgment
- **Red**: Do not proceed until must-haves met

**Sponsor Sign-off**: _________________ Date: _______

Implementation Checklist

Phase 1: Planning (Week 1)

• Identify assessment team (data, tech, process, legal leads)
• Define scope: Which AI initiative to assess?
• Gather baseline artifacts: architecture, data dictionaries, org charts
• Schedule stakeholder interviews

Phase 2: Data Assessment (Weeks 2-3)

• Complete data inventory across all sources
• Run data quality analysis on samples (1K-10K records)
• Test data joins and integrations
• Conduct labeling feasibility test
• Privacy audit: map PII, verify consent
• Document findings in heat map

Phase 3: Technology Assessment (Weeks 2-4)

• Catalog current infrastructure and platforms
• Execute critical tech spikes (latency, cost, integration)
• Evaluate vendor/platform options
• Document findings and recommendations

Phase 4: People & Process (Weeks 3-4)

• Skills inventory: current vs. required
• Operating model maturity assessment
• RACI definition and validation
• Change management readiness
• Document findings and mitigation plans

Phase 5: Legal & Compliance (Weeks 3-4)

• Identify applicable regulations
• Map data flows and risk levels
• Conduct compliance gap analysis
• Engage legal counsel for review
• Develop remediation plan for gaps

Phase 6: Synthesis & Mitigation (Week 5)

• Aggregate heat map across all dimensions
• Prioritize gaps by severity and impact
• Develop mitigation plans with owners
• Assess overall readiness: Green/Yellow/Red
• Create scenario analysis

Phase 7: Gating & Decision (Week 6)

• Define phase gate criteria
• Draft readiness assessment report
• Present findings to sponsors
• Obtain go/no-go decision
• If go: Confirm mitigation commitments

Ongoing: Track & Update

• Monitor mitigation progress weekly
• Update heat map as gaps close
• Escalate new risks immediately
• Re-assess readiness at each phase gate
• Incorporate learnings into next initiative

Key Takeaway: Readiness assessment is not a one-time gate but a continuous discipline. Honest evaluation early, backed by evidence from spikes and probes, prevents costly failures and builds organizational trust. The goal is not perfection but informed decision-making—knowing what's ready, what needs work, and what must be mitigated before proceeding.