This analysis compares the 10-year total cost of ownership for two approaches to AI deployment at ACME Highland Health: Option A (unified platform via Project Catalyst, the adopted approach) vs. Option B (piecemeal point solutions purchased independently by business units). The analysis covers build costs (Years 1–3) and operating costs (Years 4–10) to provide a complete lifecycle view. Read alongside the Cost-Benefit Analysis, which covers the benefits side of the equation.
Table of Contents
1. Executive Summary
| Option A: Unified Platform | Option B: Piecemeal | Delta | |
|---|---|---|---|
| 10-Year Total Cost | $196,400,000 | $136,500,000 | +$59,900,000 |
| Build Cost (Years 1–3) | $99,000,000 | $71,400,000 | +$27,600,000 |
| Operating Cost (Years 4–10) | $97,400,000 | $65,100,000 | +$32,300,000 |
Option A costs $59.9M more over 10 years. The question is not whether it costs more — it does. The question is whether the additional $59.9M buys capabilities, risk reduction, and organizational maturity that Option B cannot deliver. Section 7 analyzes the delta. The CBA demonstrates that the benefits generated by Option A (NPV +$2.97M at 8% discount) exceed the cost premium when measured against the operational improvements that only a governed, unified platform can achieve.
2. Methodology & Assumptions
2.1 Time Horizon
10 years (2026–2036): Years 1–3 are build phase (Project Catalyst program lifecycle). Years 4–10 are operations phase (ACME operates the platform independently after Pulaski roll-off). This horizon captures the full lifecycle cost including the transition from consulting delivery to internal operations.
2.2 Key Assumptions
- Inflation: 3% annual escalation for labor costs; 0% for cloud infrastructure (historically declining, but modeled flat for conservatism); 3% for SaaS/tooling licenses.
- Staffing: Option A transitions from 262 program-phase personnel to approximately 45 permanent ACME AI operations staff (Years 4–10). Option B requires approximately 25 permanent staff across fragmented, independent AI initiatives.
- Cloud costs: Based on current hyperscaler pricing with 3-year reserved-instance commitments. Multi-cloud architecture assumed for Option A (per Change Control CN-003).
- Governance overhead: Option A includes permanent AI Governance & CoE function (12 FTEs during program, transitioning to 8 permanent ACME FTEs). Option B does not include enterprise governance — each point solution has its own vendor-managed oversight.
- Discount rate: Costs are presented in nominal dollars (not discounted) for TCO comparison. NPV analysis is in the CBA.
3. Option A: Unified AI Platform (Project Catalyst)
3.1 Build Phase (Years 1–3): $99,000,000
| Category | Year 1 | Year 2 | Year 3 | Total |
|---|---|---|---|---|
| Internal FTE Labor | $5,600,000 | $8,400,000 | $5,600,000 | $19,600,000 |
| Onshore Consultant Labor | $7,700,000 | $11,600,000 | $8,100,000 | $27,400,000 |
| Offshore Consultant Labor | $4,000,000 | $6,000,000 | $4,200,000 | $14,200,000 |
| Cloud & AI Platform Infrastructure | $8,100,000 | $10,800,000 | $8,000,000 | $26,900,000 |
| Tooling (Governance, Compliance, QA) | $600,000 | $700,000 | $600,000 | $1,900,000 |
| Contingency Reserve | $2,520,000 | $3,780,000 | $2,700,000 | $9,000,000 |
| Total Build | $28,520,000 | $41,280,000 | $29,200,000 | $99,000,000 |
Note: Year totals above reflect cost-category allocation; SOW milestone payments ($27.72M / $41.58M / $29.70M) differ slightly due to payment timing vs. cost incurrence.
3.2 Operations Phase (Years 4–10): $97,400,000
| Category | Annual (Year 4) | Annual (Year 10, w/ inflation) | 7-Year Total |
|---|---|---|---|
| AI Operations Staff (45 ACME FTEs) | $6,750,000 | $8,060,000 | $51,800,000 |
| Cloud Infrastructure & Compute | $4,200,000 | $4,200,000 | $29,400,000 |
| AI Governance & CoE (8 ACME FTEs) | $1,200,000 | $1,430,000 | $9,200,000 |
| Tooling & License Renewals | $500,000 | $600,000 | $3,800,000 |
| Model Retraining & Validation (annual) | $400,000 | $480,000 | $3,200,000 |
| Total Annual Operations | $13,050,000 | $14,770,000 | $97,400,000 |
4. Option B: Piecemeal Point Solutions
4.1 Build Phase (Years 1–3): $71,400,000
Under Option B, ACME purchases three separate AI solutions from different vendors — a PA automation vendor, a conversational AI vendor, and an underwriting analytics vendor — each with its own implementation, infrastructure, and support contract.
| Category | PA Automation | Conversational AI | Underwriting Analytics | Total |
|---|---|---|---|---|
| Vendor License & Implementation | $18,000,000 | $12,000,000 | $8,000,000 | $38,000,000 |
| Integration & Customization (ACME) | $5,500,000 | $4,000,000 | $3,200,000 | $12,700,000 |
| Separate Cloud/Infra per Vendor | $5,000,000 | $3,500,000 | $2,800,000 | $11,300,000 |
| Internal FTE (PM, BA, IT per solution) | $3,600,000 | $2,800,000 | $3,000,000 | $9,400,000 |
| Total Build | $32,100,000 | $22,300,000 | $17,000,000 | $71,400,000 |
4.2 Operations Phase (Years 4–10): $65,100,000
| Category | Annual (Year 4) | 7-Year Total |
|---|---|---|
| Vendor License Renewals (3 vendors) | $3,800,000 | $28,900,000 |
| Cloud Infrastructure (3 separate environments) | $2,600,000 | $18,200,000 |
| Internal Support Staff (25 FTEs across 3 solutions) | $2,100,000 | $16,100,000 |
| No Centralized Governance Function | $0 | $0 |
| Vendor-Managed Model Updates (per contract) | $300,000 | $1,900,000 |
| Total Annual Operations | $8,800,000 | $65,100,000 |
5. Side-by-Side Comparison
| Dimension | Option A: Unified | Option B: Piecemeal |
|---|---|---|
| 10-Year Total Cost | $196,400,000 | $136,500,000 |
| Build Cost | $99,000,000 | $71,400,000 |
| Operations Cost (7 years) | $97,400,000 | $65,100,000 |
| Governance Function | Permanent AI CoE (8 FTEs) | None — vendor-dependent |
| Model Validation | Independent (ACME-owned) | Vendor-managed (conflict of interest) |
| Fairness Testing | ACME-controlled, NAIC-aligned | Vendor's standard (may not meet NAIC) |
| Data Platform | Unified, ACME-owned, multi-cloud | 3 separate vendor environments |
| Vendor Lock-In Risk | Low (ACME owns platform) | High (3 vendor dependencies) |
| Regulatory Audit Readiness | Single governance framework | 3 separate vendor audit packages |
| Cross-BRD Data Sharing | Native (shared platform) | Requires custom integration |
| Organizational Capability at Year 10 | ACME has permanent AI operations team + CoE | ACME depends on 3 vendors indefinitely |
6. Cost Category Detail
6.1 Consulting Labor ($61.2M — Option A Build Phase)
Pulaski Advisory Group provides 122 consultants (75 onshore, 47 offshore) across the 36.5-month program. Labor is the largest cost category (61.8% of build budget). The consulting investment is front-loaded — it peaks in Year 2 (when BRD-02 and BRD-03 run in parallel) and declines in Year 3 as Pulaski consultants transfer knowledge and roll off. By closeout, ACME operates independently.
6.2 Cloud & AI Platform Infrastructure ($26.9M build + $29.4M ops)
The cloud platform serves all three BRDs from a shared foundation — a critical cost advantage over Option B, which requires three separate cloud environments. Build-phase costs cover: hyperscaler compute (GPU instances for model training), storage, networking, managed AI/ML services, and development/staging/production environments. Operations-phase costs ($4.2M/year) cover production compute, storage growth, and managed service fees. Multi-cloud architecture (CN-003) enables cost optimization across providers.
6.3 AI Governance & CoE ($9.2M — Option A Operations Phase)
This is the cost category that Option B does not have. The permanent AI CoE (8 ACME FTEs) provides: enterprise AI standards maintenance, model lifecycle governance, regulatory tracking, shadow-AI prevention, and coordination with model validation. The $1.2M–$1.4M annual cost is the price of governance — the alternative (no governance, as in Option B) carries regulatory and liability costs that are harder to quantify but potentially far larger (see Section 9).
6.4 Model Retraining & Validation ($3.2M — Option A Operations Phase)
Production AI models require periodic retraining as data distributions shift, medical policies change, and regulatory standards evolve. Option A's annual retraining cost ($400K–$480K/year) covers: retraining compute, independent validation per cycle, fairness re-testing, and regression testing. Option B's vendor-managed model updates ($300K/year) are cheaper but offer less control — ACME depends on the vendor's retraining schedule and methodology.
7. The $28M Delta: What Option A Buys
Option A's build phase costs $27.6M more than Option B's. This delta buys four capabilities that Option B cannot deliver at any price:
7.1 Unified Data Platform ($8.2M of the delta)
Option A builds a shared data platform that all three AI capabilities use. Option B requires three separate data environments with custom cross-system integration. The unified platform enables: cross-BRD data sharing (PA decisions inform member service context; member interaction patterns inform underwriting), consistent data governance (one PHI handling standard vs. three vendor approaches), and lower marginal cost for future AI capabilities (the platform already exists).
7.2 Permanent AI Governance & CoE ($6.4M of the delta)
Option A establishes an enterprise AI governance function that persists after the program ends. Option B does not include governance — each vendor manages its own model risk, with no enterprise coordination, no independent validation, and no shadow-AI prevention. The governance delta buys: regulatory defensibility (single NIST AI RMF-aligned framework vs. three vendor approaches that may not satisfy NAIC requirements), liability protection (independent validation vs. vendor self-certification), and organizational AI maturity (ACME builds internal capability vs. perpetual vendor dependency).
7.3 Coordinated Change Management ($4.8M of the delta)
Option A invests in a dedicated Change Management & Training team (8 people, G. Okonkwo) that manages adoption across all three AI capabilities with a consistent approach. Option B leaves adoption to each vendor's standard implementation playbook. The coordinated approach produces higher adoption rates (which drive the operational benefits documented in the CBA) and lower organizational disruption.
7.4 Independent Model Validation ($4.1M of the delta)
Option A's two-line-of-defense structure includes a dedicated Independent Model Validation team (P. Okafor, 6 people) that tests every model against accuracy, fairness, explainability, security, and regulatory compliance criteria before production. Option B relies on vendor self-certification — the same team that builds the model certifies its quality. The independence delta buys: structural elimination of conflict-of-interest in model validation, regulatory defensibility when (not if) regulators examine ACME's AI governance, and liability protection against deploying models that fail fairness or accuracy standards.
7.5 Remaining Delta ($4.1M)
The remaining build-phase delta covers: higher-quality consulting talent (Pulaski's healthcare AI specialization vs. general SaaS implementation partners), more comprehensive testing (adversarial testing, hallucination testing, accessibility testing that point-solution vendors may not include), and program-level integration that ensures all three capabilities work together rather than operating as isolated tools.
8. Sensitivity Analysis
Five key assumptions drive the cost difference between Options A and B. Each assumption is varied ±20% to assess the impact on the 10-year TCO delta:
| Assumption | Base Value | -20% Impact on Delta | +20% Impact on Delta |
|---|---|---|---|
| Consulting labor rates | $175–$275/hr (onshore) | Delta narrows by $12.2M | Delta widens by $12.2M |
| Cloud infrastructure costs | $26.9M build | Delta narrows by $5.4M | Delta widens by $5.4M |
| Option B vendor license costs | $38M total | Delta widens by $7.6M (Option B cheaper) | Delta narrows by $7.6M (Option B more expensive) |
| Operations staff count (Option A) | 45 FTEs | Delta narrows by $10.4M | Delta widens by $10.4M |
| Labor cost inflation rate | 3% annual | Delta narrows by $4.8M | Delta widens by $4.8M |
Key finding: Even in the most favorable scenario for Option A (consulting rates -20%, vendor costs +20%), Option A still costs more than Option B. The cost premium is structural — it reflects the additional investment in governance, platform unification, and organizational capability. The justification for that premium comes from the benefits side (CBA), not from cost parity.
9. Risk Scenarios
9.1 Regulatory Enforcement Scenario
9.2 Vendor Discontinuation Scenario
If one of Option B's three AI vendors discontinues their product, raises prices significantly, or is acquired, ACME must replace that capability with minimal leverage. Under Option A, ACME owns the platform and the models — vendor changes affect infrastructure providers (replaceable via multi-cloud) but not the AI capabilities themselves. Estimated migration cost if an Option B vendor exits: $5M–$12M per vendor (depending on data portability and replacement timeline).
9.3 AI Liability Scenario
If ACME faces litigation related to AI-generated decisions (precedent: Moffatt v. Air Canada), the legal defense depends heavily on governance documentation. Under Option A, ACME demonstrates: independent model validation, documented fairness testing, mandatory human-in-the-loop for adverse determinations, and a complete audit trail. Under Option B, ACME's defense depends on vendor-provided documentation of unknown quality and completeness. Estimated incremental legal exposure under Option B vs. Option A: unquantifiable but significant — potentially tens of millions in a class-action scenario.
10. Recommendation & Cross-References
Recommendation: Option A is the adopted approach. It costs more, and the cost premium is real. But the premium buys capabilities — unified governance, independent validation, platform ownership, and organizational AI maturity — that Option B cannot deliver. The Cost-Benefit Analysis demonstrates that the operational benefits generated by Option A (NPV +$2.97M at 8% discount) exceed the total program investment when measured over a 10-year horizon. The risk scenarios in Section 9 illustrate that the cost of not having governance (regulatory fines, vendor lock-in, legal liability) can rapidly exceed the governance premium.
Read alongside:
- Cost-Benefit Analysis — benefits framework and NPV analysis
- Program Charter §13–16 — budget authorization and financial controls
- Program Management Plan §8 — cost management procedures
- SOW-01, SOW-02, SOW-03 — contractual fee schedules