Nexus Digital Solutions currently spends $213,429 annually ($17,786/month) across six AI projects — approximately 25.4% of its declared $840,000 AI budget. This utilization level reflects an organization in active scaling rather than mature deployment. While the portfolio demonstrates meaningful capability in three high-performing initiatives, it also carries concentrated waste and one critically failing project that demands immediate executive action.
A structured 10-week optimization program targeting redundant subscriptions and zero-utilization tools can recover $77,820 in annual savings — 36.5% of current AI spend — with no disruption to high-performing initiatives. Immediate cancellation of four non-adopted subscriptions alone delivers $4,950 in monthly savings within two weeks, requiring only vendor portal access and leadership authorization.
| Metric | Current Value | Benchmark |
|---|---|---|
| Cost per Employee (Monthly) | $29.64 | — |
| AI Tools per Department (Avg) | 2.4 | — |
| Average Seat Utilization Rate | 58% | ≥ 75% |
| ROI-Positive Projects | 50% (3 of 6) | ≥ 75% |
| Shadow AI Risk Level | Medium | Low |
| Monthly LLM API Cost | $1,446 (8.1%) | — |
| Monthly SaaS Subscription Cost | $16,340 (91.9%) | — |
| Estimated Monthly Waste | $6,580 | $0 |
Nexus has deployed AI across five departments with a dual-track architecture: 19 SaaS subscriptions running alongside custom-built LLM pipelines using OpenAI, Anthropic, and AWS Bedrock. Three projects are demonstrably performing, but capability has outpaced procurement controls and lifecycle management.
The core problem is structural: engineering teams have successfully built custom AI pipelines that have made several legacy SaaS tools redundant, but the offboarding cycle has not been completed. With 1.31 billion monthly tokens consumed, model usage is substantial, yet the cost structure is almost entirely subscription-driven — meaning cost optimization does not require reducing AI capability, only eliminating vendor overhead.
| Project | Dept | Monthly Cost | ROI | KPI | Risk | Status |
|---|---|---|---|---|---|---|
| Customer Support AI Copilot | CS | $4,255 | 92 | 104% | Med | Active |
| Content Generation Engine | Mktg | $1,843 | 88 | 117% | Low | Active |
| Sales Intelligence Pipeline | Sales | $5,367 | 85 | 112% | Low | Active |
| Code Review Automation | Eng | $2,421 | 63 | 71% | Med | Underperf. |
| Internal Knowledge Assistant | Eng | $2,400 | 58 | 62% | High | Underperf. |
| Financial Forecasting Asst. | Fin | $1,500 | 12 | 0% | Crit | Failing |
| Portfolio Total | $17,786 | 66 avg | 78% |
The strongest overall performer, with 104% KPI achievement and measurable improvements in CSAT and autonomous ticket resolution rates approaching the 60% target. Primary optimization opportunity is over-provisioned infrastructure, which carries medium risk but does not impair current performance.
The portfolio's standout cost-efficiency story. At $1,843/month — the second-lowest project cost — this initiative delivers the highest KPI over-achievement at 117%. Marketing's successful migration from Jasper AI to a custom Claude Sonnet pipeline is a clear example of in-house capability surpassing vendor tooling. Jasper AI ($1,200/month) should be cancelled immediately.
Strong performance at 112% KPI achievement. At $5,367/month, this is the portfolio's highest-cost project. Gong AI ($3,200/month at 70% seat utilization) represents the primary optimization lever — seat-level right-sizing and renegotiation are actionable without disrupting pipeline performance.
A structurally conflicted project: the custom RAG chatbot and Notion AI are competing for the same use case. RAG accuracy at 78% is insufficient to justify deprecating Notion AI, yet Notion AI utilization has collapsed to 30% across 80 seats. Immediate action: right-size Notion AI from 80 to 24 seats, recovering ~$640/month.
Moderate value delivery hampered by dual-tool redundancy. Codacy and a Claude Sonnet-based review pipeline are running in parallel without a defined winner. A structured 4-week pilot is required before any cancellation decision.
Financial Forecasting Assistant — ROI Score: 12
This project represents a critical governance failure. The Runway FP&A subscription ($1,500/month) has not been activated, no KPIs have been achieved, and there is no evidence of deployment or adoption. This is not an underperforming project — it is an unstarted one. Cancel in Week 1. Conduct a structured viability review before any future Finance AI investment.
| Category | Annual Savings | Effort | Timeline |
|---|---|---|---|
| Cancel four zero-value subscriptions | $59,400 | Low | Weeks 1–2 |
| Right-size Gong AI & GitHub Copilot seats | $9,120 | Medium | Weeks 2–5 |
| Codacy pilot + Notion AI seat reduction | $7,680 | Medium | Weeks 3–8 |
| Governance framework (recurrence prevention) | Structural | Medium | Weeks 4–10 |
| Total Recoverable | $77,820 | 10 weeks |
Savings represent 36.5% of current annual AI spend. No high-performing initiative is impacted.
| Subscription | Monthly Cost | Reason |
|---|---|---|
| Runway FP&A | $1,500 | Unstarted; zero adoption; project in critical failure |
| Jasper AI | $1,200 | Team fully migrated to Claude Sonnet pipeline |
| Zendesk AI Add-on | $1,800 | Full overlap with Intercom AI + custom GPT-4o pipeline |
| Lavender AI | $450 | Confirmed non-adoption across all assigned seats |
| Total | $4,950/mo |
No technical migration required. Execution requires vendor portal access and leadership authorization — estimated at 2–4 hours of administrative time total.
Audit and remove inactive Gong AI seats (currently 70% utilized at $3,200/month) through direct vendor negotiation targeting a 15–25% rate reduction. Reduce GitHub Copilot from 50 to 38 seats based on confirmed active developer usage. Combined estimated monthly savings of $640–$960.
Pause Codacy ($330/month) and run a structured 4-week defect-category comparison against Claude-only code review. Retain Codacy only if it demonstrates unique coverage in security or compliance categories. Simultaneously reduce Notion AI from 80 to 24 engineering seats — this action is independent of the RAG sprint and recovers ~$640/month in guaranteed savings.
Three governance mechanisms required: (1) monthly utilization monitoring dashboard with a 60% seat utilization renewal threshold; (2) mandatory dual CTO/CFO approval for any new AI subscription above $500/month; (3) documented model routing and tiering standards. This converts a one-time recovery into permanent portfolio management capability.
Following Runway FP&A cancellation, convene a Finance and Engineering leadership review to assess whether a viable AI forecasting use case exists. Any future investment must be conditioned on documented success criteria, a named project owner, and a 90-day evaluation window with measurable KPIs defined before spend is approved.
| Phase | Timeline | Focus | Savings |
|---|---|---|---|
| Phase 1: Immediate Cancellations | Wk 1–2 | Zero-risk administrative actions | $59,400/yr |
| Phase 2: Consolidation & Right-Sizing | Wk 3–8 | Structured pilots and seat audits | $15,420/yr |
| Phase 3: Governance | Wk 8–10 | Policy deployment and monitoring | $3,960/yr |
| Program Total | 10 weeks | $77,820/yr |
Conservative floor: Even if Codacy is retained and Gong AI negotiation yields minimal results, Phase 1 cancellations alone deliver $59,400 annually — providing guaranteed program value regardless of Phase 2 outcomes.
| Risk | Severity | Prob. | Primary Mitigation |
|---|---|---|---|
| Zendesk AI cancellation creates ticket routing gap | Med | Low | 5-day parallel monitoring; rollback trigger at >15% resolution time increase or >3pt CSAT decline |
| RAG sprint extends beyond 6 weeks | Med | Med | Week-4 hard decision gate; Notion seat reduction proceeds independently |
| Codacy pilot reveals unique security coverage | Med | Med | Evidence-based pilot; retain at reduced count if unique value confirmed |
| New redundant subscriptions added during optimization | High | Med | Governance policy enforced by Week 6; dual CTO/CFO gate above $500/mo |
| Financial Forecasting Asst. re-funded without review | High | Med | Formal use-case review required; executive sign-off mandated |
Systemic governance risk: The current $6,580/month in waste accumulated across only six projects in the absence of procurement controls. As Nexus scales toward its $840,000 AI budget capacity, the same dynamic will compound proportionally. The AI Tool Procurement Policy and utilization monitoring dashboard are not optional enhancements — they are the mechanism by which this optimization program delivers durable, rather than temporary, value.
Cost attribution was performed by mapping all AI-related subscription and API costs to the projects and departments that consume them. ROI scores (0–100) are composite metrics incorporating KPI achievement, value alignment, utilization efficiency, and risk level. The Portfolio Health Score (62/100) is a portfolio-level composite weighting ROI distribution, utilization efficiency, redundancy index, and governance maturity. All annualized figures are current monthly run-rate × 12 without growth adjustment.
Cost data reflects current subscription invoices and LLM API billing records (high confidence). Utilization rates are based on seat activation and feature engagement data. KPI figures are sourced from departmental reporting and may reflect partial-quarter data. Savings projections for seat reductions assume pro-rata billing; vendor renegotiation outcomes will vary.
| Provider | Role | Monthly Exposure |
|---|---|---|
| OpenAI (GPT-4o, GPT-4o-mini) | Primary LLM and embedding | Incl. in $1,446 API total |
| Anthropic (Claude Sonnet 4.5) | Content and code review LLM | Incl. in $1,446 API total |
| AWS Bedrock | Sales LLM pipeline | Incl. in $1,446 API total |
| SaaS Subscriptions (19 tools) | Productivity, analytics, workflow | $16,340/month |
| Timeline | Action | Owner |
|---|---|---|
| Week 1 | CTO and CFO authorize cancellation of Runway FP&A, Jasper AI, Zendesk AI Add-on, and Lavender AI; execute before end of current billing cycle | CTO, CFO |
| Week 2 | Convene Finance and Engineering leadership for Financial Forecasting Assistant post-mortem; produce written viability assessment with go/no-go by Week 4 | CFO, VP Eng |
| Week 3 | Launch Codacy pilot, Notion AI seat right-sizing, and RAG accuracy sprint simultaneously; assign named workstream owners; schedule Week 10 governance sign-off | VP Eng, Program Lead |