Chapter 11 — AI Roadmapping & Budget Planning

Overview

Create a multi-horizon roadmap that sequences initiatives, budgets resources, and honors constraints. A well-crafted roadmap makes tradeoffs explicit, aligns stakeholders on priorities, and ensures resources flow to the highest-value opportunities.

AI transformation isn't a single project—it's a portfolio of initiatives that must be carefully sequenced to balance quick wins with foundational capabilities. This chapter provides frameworks for building multi-horizon AI roadmaps and comprehensive budget models that keep initiatives funded and on track.

The Strategic Imperative

Why Roadmapping Matters:

• 67% of AI initiatives fail to reach production due to poor planning (Gartner)
• Organizations with structured roadmaps deliver 3.2x more value per dollar invested
• Clear sequencing reduces time-to-value by 40% through parallelization
• Explicit dependencies prevent 60% of common bottlenecks
• Budget transparency reduces mid-flight funding crises by 75%

Consequences of Poor Planning:

• Teams work on low-value projects while high-value ones wait
• Platform dependencies block multiple initiatives simultaneously
• Talent sits idle waiting for blockers to clear
• Sponsors lose faith when promises slip repeatedly
• Budget overruns force project cancellations mid-stream

Three-Horizon Framework

graph TB
    subgraph H1["Horizon 1: 0-6 Months<br/>Prove Value & Learn"]
        H1A[Quick Win 1<br/>AI Chatbot Pilot]
        H1B[Foundation POC<br/>Data Pipeline]
        H1C[Team Building<br/>Hire 3-5 people]
    end

    subgraph H2["Horizon 2: 6-18 Months<br/>Scale & Build Platforms"]
        H2A[Platform Deploy<br/>MLOps Infrastructure]
        H2B[Production Systems<br/>3-5 Use Cases]
        H2C[Capability Build<br/>Team 15-30 people]
    end

    subgraph H3["Horizon 3: 18+ Months<br/>Transform & Optimize"]
        H3A[Enterprise Platform<br/>Self-Service AI]
        H3B[Scaled Portfolio<br/>10+ Use Cases]
        H3C[AI Culture<br/>Centers of Excellence]
    end

    H1A --> H2B
    H1B --> H2A
    H2A --> H2B
    H2A --> H3A
    H2B --> H3B
    H2C --> H3C

    style H1A fill:#d4edda
    style H2A fill:#fff3cd
    style H3A fill:#e1f5ff

Horizon Characteristics

Initiative Prioritization Framework

graph TD
    A[Evaluate Initiative] --> B{Business Value >7/10?}
    B -->|No| C[Backlog<br/>Revisit Quarterly]
    B -->|Yes| D{Technical Feasibility >6/10?}

    D -->|No| E{Can Improve<br/>Feasibility?}
    E -->|Yes| F[H2/H3 After<br/>Dependency Resolution]
    E -->|No| C

    D -->|Yes| G{Dependencies<br/>Met?}
    G -->|Yes| H[H1 Quick Win<br/>Start Immediately]
    G -->|No| I{Dependency<br/>Timeline?}

    I -->|<6 months| J[H1 Sequence<br/>After Dependency]
    I -->|6-18 months| F
    I -->|>18 months| K[H3 Long-Term<br/>Strategic Option]

    style H fill:#d4edda
    style F fill:#fff3cd
    style K fill:#e1f5ff

Prioritization Matrix

Priority Score = (Value + Feasibility + Strategic Fit + Effort) / 4

Dependency Mapping

graph LR
    A[Data Platform] -->|Enables| B[Fraud Detection]
    A -->|Enables| C[Personalization]
    A -->|Enables| D[Forecasting]

    E[Customer Data Pipeline] -->|Required for| C
    E -->|Required for| F[Churn Prediction]

    G[AI Chatbot Pilot] -->|Learns from| H[Voice AI]
    G -->|Pattern reuse| I[Email Assistant]

    J[MLOps Platform] -->|Required for| B
    J -->|Required for| C
    J -->|Required for| D
    J -->|Required for| F

    K[Team Hiring] -->|Required for| A
    K -->|Required for| J

    style A fill:#f8d7da
    style J fill:#f8d7da
    style K fill:#f8d7da

Dependency Categories

Dependency Resolution Matrix

Roadmap Sequencing

gantt
    title AI Initiative Roadmap (18 Months)
    dateFormat  YYYY-MM-DD

    section H1 Quick Wins
    AI Chatbot Pilot           :active, h1a, 2024-01-01, 90d
    Document AI POC            :h1b, 2024-02-01, 60d
    Data Pipeline POC          :h1c, 2024-01-15, 60d
    Hire Wave 1 (3-5 people)   :h1d, 2024-01-01, 90d

    section Foundations
    MLOps Platform Deploy      :h2a, 2024-03-15, 120d
    Data Platform Build        :h2b, 2024-03-01, 150d

    section H2 Scale
    Fraud Detection            :h2c, 2024-07-15, 120d
    Personalization            :h2d, 2024-08-01, 90d
    Forecasting                :h2e, 2024-09-01, 90d
    Hire Wave 2 (8-12 people)  :h2f, 2024-06-01, 120d

    section H3 Transform
    Voice AI                   :h3a, 2024-11-01, 120d
    Self-Service Platform      :h3b, 2025-01-01, 150d
    Center of Excellence       :h3c, 2025-02-01, 90d

Sequencing Principles

1. Front-Load Quick Wins: H1 should include at least one high-visibility success

• Example: Chatbot pilot showing 30% deflection in 3 months builds credibility

2. Parallel Foundation + Value: Build platforms while delivering pilot value

• Data platform builds while chatbot delivers wins
• Avoids "no value until platform ready" trap

3. Respect Dependencies: Don't promise what can't be delivered

• Fraud detection waits for data platform
• Voice AI waits for chatbot learnings

4. Maintain Momentum: No gaps >2 months without visible progress

• Continuous stream of demos and wins
• Prevents sponsor disengagement

5. Balance Portfolio: Mix of revenue, cost savings, and risk reduction

• Not all cost-cutting or all revenue-generating
• Diversified value story

6. Capacity Reality: Don't oversubscribe teams; leave 20% buffer

• Account for unplanned work, learning curves
• Prevents burnout and quality issues

Phase Gates & Milestones

graph LR
    A[Ideation] -->|Gate 0<br/>Strategic Fit| B[Discovery]
    B -->|Gate 1<br/>Feasibility| C[Build]
    C -->|Gate 2<br/>Quality| D[Pilot]
    D -->|Gate 3<br/>Value Proven| E[Scale]
    E -->|Gate 4<br/>ROI Realized| F[Optimize]

    style A fill:#e1f5ff
    style B fill:#fff4e1
    style C fill:#ffe1e1
    style D fill:#e1ffe1
    style E fill:#f0e1ff
    style F fill:#d4edda

Gate Criteria

Capacity Planning

Team Ramp Model

graph LR
    A[Month 0:<br/>3 Core Team] --> B[Month 3:<br/>7 People<br/>Wave 1 Complete]
    B --> C[Month 6:<br/>12 People<br/>+Contractors]
    C --> D[Month 9:<br/>18 People<br/>Wave 2 Hiring]
    D --> E[Month 12:<br/>22 People<br/>H2 Full Team]
    E --> F[Month 18:<br/>28 People<br/>Wave 3 Started]
    F --> G[Month 24:<br/>34 People<br/>H3 Full Team]

    style A fill:#f8d7da
    style C fill:#fff3cd
    style E fill:#e1f5ff
    style G fill:#d4edda

Capacity Utilization

Analysis: ML Engineers over-subscribed by 1 person-month. Mitigation: Hire 1 additional MLE in Wave 1 or push MLOps platform start by 2 weeks.

Budget Model

Infrastructure Costs

AI Services & Licenses

Vendor & Consulting

Build vs. Buy vs. Partner

graph TD
    A[Component Decision] --> B{Strategic<br/>Differentiation?}
    B -->|High| C{Internal<br/>Expertise?}
    B -->|Low| D[Buy SaaS<br/>$100K/yr]

    C -->|Yes| E[Build In-House<br/>$400K + $50K/yr]
    C -->|No| F{Time<br/>Constraint?}

    F -->|Urgent| G[Partner Co-Develop<br/>$250K + $30K/yr]
    F -->|Not Urgent| H{Can Hire<br/>Talent?}

    H -->|Yes| E
    H -->|No| G

    style D fill:#d4edda
    style E fill:#fff3cd
    style G fill:#e1f5ff

Build/Buy/Partner Analysis

3-Year TCO Comparison:

Savings: Hybrid approach saves $1.02M vs. all-build, provides more control than all-buy.

Complete Budget Summary

Multi-Year Budget

Budget Allocation by Horizon

Quarterly Review Process

graph LR
    A[Month 1-2:<br/>Execute Plan] --> B[Month 3:<br/>Quarterly Review]
    B --> C[Month 4-5:<br/>Execute Updates]
    C --> D[Month 6:<br/>Quarterly Review]
    D --> E[Month 7-8:<br/>Execute Updates]
    E --> F[Month 9:<br/>Quarterly Review]
    F --> G[Month 10-11:<br/>Execute Updates]
    G --> H[Month 12:<br/>Annual Review]
    H --> A

    B -->|Adjustments| A
    D -->|Adjustments| C
    F -->|Adjustments| E
    H -->|Major Refresh| A

    style B fill:#fff3cd
    style D fill:#fff3cd
    style F fill:#fff3cd
    style H fill:#f8d7da

Review Agenda

Case Study: Financial Services AI Roadmap

Context

Mid-sized bank, $5B assets, 500 branches, starting from zero AI capability. Goal: Deploy AI across customer experience, fraud prevention, and operations.

Initial Inventory

25 potential initiatives identified across 4 categories:

• Customer Experience: 8 ideas (chatbot, personalized offers, voice banking, sentiment analysis)
• Fraud & Risk: 6 ideas (transaction fraud, KYC automation, credit risk scoring)
• Operations: 7 ideas (document processing, call routing, forecasting, compliance)
• Infrastructure: 4 ideas (data platform, MLOps, governance, self-service)

Prioritization Results

H1 Candidates (Top 5):

Selected H1 Portfolio

graph TD
    A[Month 1-2:<br/>Data Platform POC] --> B[Month 2-4:<br/>Chatbot Pilot]
    A --> C[Month 3-6:<br/>Fraud Detection Build]

    B --> D[Month 5-6:<br/>Chatbot Production]
    C --> E[Month 6-9:<br/>Fraud Production]

    F[Parallel:<br/>Hire 3 People<br/>Month 1-3]
    G[Parallel:<br/>MLOps Vendor Selection<br/>Month 2-4]

    style A fill:#f8d7da
    style B fill:#fff3cd
    style C fill:#e1f5ff
    style D fill:#d4edda
    style E fill:#d4edda

Budget Allocation

Results After Year 1

✅ Delivered:

• Fraud detection live, catching **$2.4M fraud annually** (vs.$2M projected)
• Chatbot handling 35% of tier-1 support calls ($850K annual savings)
• Data platform POC validated, production build underway
• Team grown to 18 people (vs. 19 planned, 1 open req)

⚠️ Challenges:

• Document AI delayed 6 weeks due to integration complexity
• Budget overrun of 8% due to higher LLM costs than estimated
• Fraud detection took 10 weeks vs. 8 weeks planned (still acceptable)

🚫 Descoped:

• Credit risk model pushed to Year 2 due to regulatory delays (expected)
• Voice banking removed from roadmap due to low ROI re-assessment

Key Lessons

Financial Results

Implementation Checklist

Phase 1: Inventory & Prioritization (Weeks 1-3)

• Conduct stakeholder workshops to identify all potential AI initiatives
• Research industry benchmarks and use cases
• Score each initiative on value (1-10), feasibility (1-10), strategic fit (1-10), effort (1-10 inverted)
• Create long-list (20-30 ideas) and short-list (8-12 candidates)
• Validate prioritization with executive team

Phase 2: Dependency Mapping (Weeks 2-4)

• Identify data dependencies for each initiative
• Map platform/infrastructure requirements
• Document compliance and legal dependencies
• Identify vendor/partner dependencies
• Create dependency matrix showing blockers and unblock dates

Phase 3: Sequencing & Roadmap (Weeks 4-6)

• Apply sequencing logic (quick wins + foundations in parallel)
• Respect dependency constraints
• Balance portfolio across value types (revenue, savings, risk)
• Create wave plan with milestones
• Define phase gates and criteria
• Build visual roadmap (Gantt, swim lane, or horizon view)

Phase 4: Capacity Planning (Weeks 5-7)

• Estimate effort for each initiative (person-months by role)
• Create hiring plan with waves
• Model capacity utilization by quarter
• Identify over/under-subscription
• Develop contractor/vendor strategy
• Define onboarding and ramp assumptions

Phase 5: Budget Model (Weeks 6-8)

• Estimate infrastructure costs (compute, storage, network)
• Price out licenses and AI services
• Calculate team costs (salaries, benefits, recruiting)
• Include vendor/consulting costs
• Add change management budget
• Apply contingency buffers (15-20%)
• Create multi-year budget summary

Phase 6: Build vs. Buy vs. Partner (Weeks 7-8)

• List all major components/capabilities needed
• Evaluate strategic differentiation for each
• Assess internal capability and speed needs
• Run 3-year TCO analysis for each option
• Make build/buy/partner decisions
• Document rationale

Phase 7: Socialization & Approval (Weeks 9-10)

• Create executive roadmap presentation
• Build detailed backup materials (dependency maps, budget workbook)
• Pre-brief sponsor and key stakeholders
• Present to steering committee or executive team
• Incorporate feedback and revise
• Obtain formal approval and budget commitment

Phase 8: Operationalize (Weeks 11-12)

• Set up project tracking (Jira, Asana, etc.)
• Create RAID log (Risks, Assumptions, Issues, Dependencies)
• Schedule recurring roadmap reviews (quarterly)
• Kick off first wave of hiring
• Begin first H1 initiatives
• Establish metrics and reporting dashboards

Ongoing: Review & Update

• Quarterly roadmap review (actuals vs. plan, lessons, adjustments)
• Monthly budget tracking (actuals vs. forecast)
• Bi-weekly capacity utilization review
• Trigger-based re-planning when major changes occur
• Annual strategic refresh of H2 and H3

Key Takeaways

1. Three-Horizon Framework balances quick wins (H1), foundational capabilities (H2), and transformation (H3). Don't skip to H3 without H1/H2 groundwork.

2. Explicit Dependencies prevent bottlenecks. Map data, platform, compliance, vendor, and people dependencies upfront. Sequence initiatives to respect constraints.

3. Capacity Planning prevents over-subscription. Model person-months by role, leave 20% buffer for unknowns. Hiring takes 2-3 months—plan accordingly.

4. Build/Buy/Partner decisions drive 40-60% of budget. Buy commodities, build differentiators, partner for speed + expertise.

5. Phase Gates enable course correction. Define clear go/no-go criteria at each milestone. Killing a project early is success, not failure.

6. Quarterly Reviews keep roadmap aligned with reality. Update based on actual progress, learnings, and changing business priorities.

7. Budget Transparency builds trust. Show all costs (team, infrastructure, vendors, change management) with contingency buffers. No surprises.

8. Roadmaps Are Living Documents. Plan in detail for H1 (6 months), medium detail for H2 (18 months), light detail for H3 (24+ months). Refresh quarterly.

Further Reading

• "The Roadmap Book" by C. Todd Lombardo
• "Project Portfolio Management" by PMI
• "Team Topologies" by Matthew Skelton (for capacity planning)
• McKinsey on AI Roadmapping: https://www.mckinsey.com/capabilities/quantumblack/our-insights