Executive Summary
Project Catalyst is a $99.0M, 36.5-month investment to build enterprise AI capability at ACME Highland Health across claims & prior authorization, member & provider experience, and underwriting & risk. This analysis models the program over a 10-year horizon at an 8% discount rate. The business case is positive but disciplined: it clears the 8% hurdle with a modest margin, and its return depends on delivering the benefit ramp on schedule. It is presented on conservative assumptions rather than optimistic ones — the intent is a case that holds up under scrutiny, not one that flatters the program.
1 · Approach & Methodology
The analysis compares the fully-loaded program investment against quantified, risk-adjusted operational benefits over a 10-year horizon (Years 1–3 build; Years 4–10 steady-state operation). Key modeling conventions:
- Discount rate 8% — ACME's weighted cost of capital; all cash flows discounted to present value at end-of-year timing.
- Conservative benefit recognition — no benefit during the build beyond partial ramp as capabilities go live; full run-rate not assumed until Year 5.
- Run-state cost included — ongoing cloud/compute, model monitoring and retraining, and CoE operations of $9.5M/year netted against gross benefit from Year 4 onward.
- Benefits are operational, not headcount-elimination — modeled as productivity, deflection, automation, and loss-avoidance, consistent with the program's human-in-the-loop governance posture.
- Excludes the separate CMS-0057-F baseline compliance project (out of scope) and any speculative revenue upside.
2 · Investment (Cost) Model
The $99.0M program investment is drawn against three annual SOW envelopes. Steady-state run cost of $9.5M/year begins once capabilities are in production (Year 4+).
| Program Year | Scope | Build Investment | Run-State Cost |
|---|---|---|---|
| Year 1 | Foundation, CoE, Platform, BRD-01 | $27,720,000 | $0 |
| Year 2 | BRD-02, BRD-03 (peak delivery) | $41,580,000 | $0 |
| Year 3 | Optimization & sustain, closeout | $29,700,000 | $0 |
| Years 4–10 | Steady-state operation (per year) | $0 | $9,500,000 |
| Build total | $99,000,000 | $9.5M/yr |
Build investment reconciles to the Program Budget and Charter §13.1. Run-state cost reconciles to the ongoing-operations line in the TCO model.
3 · Benefit Model — Full Run-Rate (Year 5+)
Four quantified benefit streams total $27.2M annually at full run-rate, each modeled conservatively from ACME's operating baselines.
| Benefit Stream | Basis of Estimate | Annual Value |
|---|---|---|
| Claims & PA Automation & Labor | ~800k annual prior-auth requests; AI auto-adjudication of routine cases under clinical oversight reduces manual review labor and cycle time against a ~$30/request baseline cost. | $12,800,000 |
| PA & Claims Fraud / Waste Detection | Improved detection of inappropriate or anomalous prior-auth and claims patterns via model scoring. | $2,400,000 |
| Member & Provider Experience & Deflection | Virtual-assistant contact deflection plus agent-assist productivity across member and provider service channels. | $8,800,000 |
| Underwriting Risk & Productivity | Improved risk selection and underwriter throughput from decision-support scoring (post independent validation). | $3,200,000 |
| Total Full Run-Rate Annual Benefit | $27,200,000 |
Benefit Realization Ramp (Years 1–5)
Benefits ramp as capabilities reach production and adoption matures — a deliberately conservative curve given the phased BRD go-lives.
| Year | Phase | Realization | Gross Benefit | Drivers |
|---|---|---|---|---|
| Year 1 | Build & Pilot | 0% | $0 | BRD-01 pilot only; no production volume |
| Year 2 | BRD-01 ramp + BRD-02/03 pilot | 25% | $6,800,000 | BRD-01 in production (PA automation ramping); BRD-02/03 in pilot |
| Year 3 | Expansion & optimization | 55% | $14,960,000 | BRD-02 to production; BRD-03 ramping; PA approaching steady-state |
| Year 4 | Sustained operations | 85% | $23,120,000 | All BRDs in steady-state on trailing edge of ramp |
| Year 5+ | Full run-rate | 100% | $27,200,000 | All systems optimized; full benefit capture |
4 · 10-Year Cash Flow, NPV, IRR & Payback
Net cash flow combines build investment (Years 1–3), realized gross benefit, and run-state cost (Years 4+), discounted at 8%.
| Yr | Investment | Gross Benefit | Run Cost | Net Cash Flow | Cumulative | PV @ 8% | Cum. PV |
|---|---|---|---|---|---|---|---|
| Y1 | $27.72M | $0 | $0 | -$27,720,000 | -$27.72M | -$25,666,667 | -$25.67M |
| Y2 | $41.58M | $6.80M | $0 | -$34,780,000 | -$62.50M | -$29,818,244 | -$55.48M |
| Y3 | $29.70M | $14.96M | $0 | -$14,740,000 | -$77.24M | -$11,701,087 | -$67.19M |
| Y4 | $0 | $23.12M | $9.50M | +$13,620,000 | -$63.62M | +$10,011,107 | -$57.17M |
| Y5 | $0 | $27.20M | $9.50M | +$17,700,000 | -$45.92M | +$12,046,323 | -$45.13M |
| Y6 | $0 | $27.20M | $9.50M | +$17,700,000 | -$28.22M | +$11,154,002 | -$33.97M |
| Y7 | $0 | $27.20M | $9.50M | +$17,700,000 | -$10.52M | +$10,327,780 | -$23.65M |
| Y8 | $0 | $27.20M | $9.50M | +$17,700,000 | +$7.18M | +$9,562,759 | -$14.08M |
| Y9 | $0 | $27.20M | $9.50M | +$17,700,000 | +$24.88M | +$8,854,407 | -$5.23M |
| Y10 | $0 | $27.20M | $9.50M | +$17,700,000 | +$42.58M | +$8,198,525 | +$2.97M |
| NPV | +$2,968,904 |
- Simple payback (from program start): ~7.6 years — cumulative undiscounted cash flow turns positive during Year 8.
- Discounted payback (from program start): ~9.6 years — cumulative present value turns positive in Year 10, within the evaluation window.
- Operational payback (from steady-state operations, after the 3-year build): ~4.6 years — run-state net cash flow recovers the accumulated build outlay within roughly four-and-a-half years of go-live.
5 · Sensitivity Analysis
NPV sensitivity to a ±20% swing in each key assumption, holding others at base case. Prior-auth volume and baseline cost are the dominant value drivers; delivery-schedule and incident risks are downside-only.
| Assumption | Base Case | Downside | Base NPV | Upside |
|---|---|---|---|---|
| PA automation volume (annual requests) | 800,000 | -$8.4M | +$2.97M | +$8.4M |
| Baseline PA processing cost | $30/req | -$6.2M | +$2.97M | +$6.2M |
| Cloud/compute cost escalation | 0.8%/yr | -$1.8M | +$2.97M | — |
| Benefit-ramp delay (6 months per BRD) | On time | -$5.1M | +$2.97M | — |
| Model incident (pause PA 3 months, Yr 2) | No incident | -$2.3M | +$2.97M | — |
6 · Scenario Analysis
| Scenario | Assumptions | 10-Yr NPV | Payback (simple) |
|---|---|---|---|
| Conservative | Benefits -20%, one 6-month ramp delay | ~-$6M | > 10 years (Year 11) |
| Base | Modeled assumptions as above | +$2.97M | ~7.6 years |
| Optimistic | Benefits +20%, on-time ramp, lower compute escalation | ~+$12M | ~6 years |
The spread is deliberately wide: this is a program whose return is earned through execution discipline. The base case is defensible; the conservative case is a genuine downside that governance is designed to prevent.
7 · Indirect & Qualitative Benefits
Not monetized in the model above, but material to the decision:
- Regulatory posture — a governed AI capability positions ACME ahead of tightening AI-in-insurance oversight (NAIC, state laws) and supports the separate CMS-0057-F compliance effort.
- Reusable platform & CoE — the Data & Cloud AI Platform and Center of Excellence lower the cost of every future AI use case beyond the three in scope.
- Member & provider experience — faster prior-auth decisions and better self-service improve satisfaction and retention, with second-order revenue effects excluded from the model.
- Talent & capability — the program builds durable in-house AI/MLOps capability that outlasts the engagement.
8 · Risks to Benefit Realization
The benefit case is directly exposed to the program's anticipated risks — each has a mitigation owner on the RAIDD Log:
- Data quality (RSK-01) — poor legacy claims data delays PA automation, the largest benefit stream.
- Compute cost (RSK-02) — unmanaged GenAI spend erodes net benefit; FinOps guardrails protect the run-state cost line.
- Adoption (RSK-07) — deflection and productivity benefits depend on frontline adoption; phased rollout protects them.
- Model incident (RSK-03) — a member-facing incident forcing a pause directly reduces realized benefit (see sensitivity).
Read Alongside
This analysis pairs with the Total Cost of Ownership (10-year run-and-maintain view), the Program Budget (baseline detail), and the Program Charter §2.3 financial summary.