CFO AI Budget Defense: ROI Models That Survive Q1

```yaml
brief:
  title: "Q1 AI Budget Defense: IRR/NPV & Controls"
  owner: "CFO – Alicia Moran"
  co_owners:
    - "VP FP&A – Daniel Cho"
    - "Controller – Priya Singh"
    - "CISO – Marta Reyes"
    - "Head of Data – Lila Patel"
  meeting_date: "2025-02-06"
  program_scope:
    - name: "Close Prep Automation"
      systems: ["Snowflake", "Workday", "Slack"]
    - name: "Contract & Redlines Assistant"
      systems: ["Salesforce", "Box", "DocuSign"]
    - name: "Support-to-Billing Triage"
      systems: ["Zendesk", "ServiceNow", "Netsuite"]
  regions:
    - region: "us-east-1"
      data_residency: "US-only"
    - region: "eu-central-1"
      data_residency: "EU-only (GDPR)"
  budget_ask_fy25: 1200000
  hurdle_rate: 0.19     # 19% IRR target
  payback_threshold_months: 9
baseline:
  finance_kpis:
    - name: "CloseDays"
      current: 7.2
      target: 6.0
    - name: "ReworkRate_AP"
      current: 11.4
      target: 7.5
    - name: "CycleTime_Contracts_days"
      current: 9.8
      target: 7.0
  unit_costs:
    analyst_hour_usd: 78
    counsel_hour_usd: 210
  cost_baseline_monthly_usd: 314000
experiment_design:
  control_groups:
    - workflow: "AP validation"
      population: 40
      treatment: 20
      control: 20
    - workflow: "MSA redlines"
      population: 60
      treatment: 30
      control: 30
  success_criteria:
    - metric: "CloseDays"
      delta: -1.0
      slo: 6.0
    - metric: "ReworkRate_AP"
      delta: -3.0
      slo: 8.0
  statistical_method: "diff-in-diff with CUPED"
  confidence_level: 0.9
telemetry_governance:
  data_sources:
    - "snowflake.fin_ops.timesheets"
    - "snowflake.ap.invoices"
    - "salesforce.opportunity"
    - "servicenow.incident"
    - "zendesk.tickets"
  prompt_logging: true
  prompt_store: "s3://audit-prompts/finance/retain-7y"
  rbac_roles:
    - role: "FPnA_Reviewer"
      access: ["derived_telemetry", "roi_models"]
    - role: "Controller_Approver"
      access: ["approvals_log", "evidence_pack"]
    - role: "Legal_Viewer"
      access: ["prompt_logs", "privacy_assessments"]
  human_in_the_loop:
    required_for: ["journal_entry_posting", "contract_acceptance"]
  model_inventory:
    providers: ["OpenAI via Azure", "Cohere", "Anthropic"]
    never_train_on_client_data: true
roi_model:
  horizon_months: 24
  discount_rate: 0.11
  irr_target: 0.19
  npv_usd_base: 4200000
  payback_months_base: 7
  sensitivity:
    downside: { npv_usd: 2100000, payback_months: 10 }
    upside:   { npv_usd: 6400000, payback_months: 6 }
approvals:
  steps:
    - step: "FP&A review"
      owner: "VP FP&A"
      due: "2025-01-22"
    - step: "Security & Privacy sign-off"
      owner: "CISO"
      due: "2025-01-24"
    - step: "Controller sign-off"
      owner: "Controller"
      due: "2025-01-27"
  last_signoff: "2025-01-27"
risks_and_mitigations:
  - risk: "PII exposure in prompts"
    control: "RBAC + data masking + residency enforcement"
  - risk: "Hallucination in contract summaries"
    control: "Human approval + source citations"
  - risk: "Model drift"
    control: "Monthly evals + rollback toggles"
board_decision:
  recommendation: "Proceed to scale Close Prep & Contract Assistant; hold triage until SLO met"
  next_review: "2025-03-31"
```

{
  "title": "CFO AI Budget Defense: ROI Models That Survive Q1",
  "published_date": "2025-11-23",
  "author": {
    "name": "Rebecca Stein",
    "role": "Executive Advisor",
    "entity": "DeepSpeed AI"
  },
  "core_concept": "Board Pressure and Budget Defense",
  "key_takeaways": [
    "Tie AI spend to finance-grade IRR/NPV in 30 days using baselines and control groups.",
    "Govern telemetry with RBAC, data residency, and prompt logging to clear audit.",
    "Start with 2-3 measurable workflows; prove payback in months, not years.",
    "Use a repeatable board brief to make funding decisions fast and defensible."
  ],
  "faq": [
    {
      "question": "What if we don’t have clean baselines?",
      "answer": "We create them in week one using historicals in Snowflake/BigQuery and define control groups. If variance is too high, we pause funding until we can attribute impact."
    },
    {
      "question": "Can we use our preferred cloud and BI?",
      "answer": "Yes. We deploy on AWS/Azure/GCP, write to Snowflake/BigQuery/Databricks, and surface in Power BI/Looker. Finance owns the tables and the model."
    },
    {
      "question": "How do you avoid vendor lock-in?",
      "answer": "We build a provider-agnostic trust layer; prompts, evals, and telemetry are portable. You can swap LLMs without changing governance or ROI measurement."
    }
  ],
  "business_impact_evidence": {
    "organization_profile": "Mid-market SaaS company (2,300 employees, $450M ARR) operating in US/EU with Snowflake and Salesforce core systems.",
    "before_state": "Three AI pilots with anecdotal benefits, no finance-owned baselines, and security pause due to missing prompt logs.",
    "after_state": "Governed telemetry in Snowflake, control groups for AP and contracts, and a board brief with IRR/NPV and 7-month payback.",
    "metrics": [
      "40% analyst hours returned in close prep within 30 days.",
      "NPV modeled at $4.2M over 24 months; base-case payback in 7 months.",
      "Error-driven rework in AP reduced from 11.4% to 7.9% during pilot."
    ],
    "governance": "Legal and Security approved due to RBAC by role, region-specific data residency, complete prompt/output logging with 7-year retention, and human-in-the-loop for journal entries and contract acceptance. No training on client data."
  },
  "summary": "Under Q1 scrutiny, defend AI budgets with IRR/NPV in 30 days. Baselines, control groups, and governed telemetry—board-ready and audit-proof."
}