A structured template for building a defensible investment proposal for AI agent deployment. Complete each section with your organisation's specific data, estimates, and context. This is a practical tool: work through it sequentially and you will have a document ready for executive review.
State the business problem in one to two sentences using business terms, not technology terms. Include current cost, error rate, processing time, or capacity constraint.
[Business problem statement]
Describe what the agent does, at what scope, and with what expected performance in one to two sentences.
[Solution summary]
| Metric | Low Estimate | Expected | High Estimate |
|---|---|---|---|
| Three-Year Total Cost of Ownership | $ | $ | $ |
| Three-Year Value Capture | $ | $ | $ |
| ROI Ratio (Value / Cost) | |||
| Payback Period (Quarter) |
Identify the top three risks by DAMAGE severity score and the governance investment required to manage each.
| Risk | DAMAGE Score | Governance Investment Required |
|---|---|---|
| 1. | $ | |
| 2. | $ | |
| 3. | $ |
State the specific approval, funding amount, resource allocation, or decision being requested.
[Specific request]
Describe who does the work today, how much it costs, how long it takes, what the error rates are, and what the capacity constraints are. This is the measured baseline. Without it, ROI claims are indefensible.
Why does this problem matter now? Connect to organisational strategy: competitive pressure, regulatory change, capacity constraints, cost reduction targets, and customer experience imperatives.
[Strategic context]
Show the specific economic value available if the problem is solved. Separate expense reduction from revenue enhancement (the dual tailwind framework).
Expense reduction opportunity:
| Category | Annual Value | Basis of Estimate |
|---|---|---|
| Labour cost reduction / throughput increase | $ | |
| Error / rework cost reduction | $ | |
| Other direct savings | $ | |
| Total expense reduction | $ |
Revenue enhancement opportunity:
| Category | Annual Value | Basis of Estimate |
|---|---|---|
| [e.g., Product attachment rate improvement] | $ | |
| [e.g., Customer retention impact] | $ | |
| [e.g., Time-to-value improvement] | $ | |
| Total revenue enhancement | $ |
What the agent does, what it does not do, what decisions it makes autonomously, and what it escalates to humans.
Custom-built, marketplace-licensed, SaaS, managed, or hybrid. Include rationale for the choice and exit criteria if the choice proves wrong.
Sufficient for technical reviewers to assess feasibility.
How humans and agents work together, as this directly affects coordination tax estimates.
Populate each of the nine TCO layers with specific estimates for this deployment. Present low / expected / high estimates for each layer.
| TCO Layer | Year 1 | Year 2 | Year 3 | Three-Year Total |
|---|---|---|---|---|
| 1. Discovery and architecture | $ | $ | $ | $ |
| 2. Build and integration | $ | $ | $ | $ |
| 3. Infrastructure and compute | $ | $ | $ | $ |
| 4. Model consumption | $ | $ | $ | $ |
| 5. Cost of failures | $ | $ | $ | $ |
| 6. Human costs | $ | $ | $ | $ |
| 7. Coordination tax | $ | $ | $ | $ |
| 8. Governance and security infrastructure | $ | $ | $ | $ |
| 9. Opportunity cost and lock-in avoidance | $ | $ | $ | $ |
| Total Cost of Ownership | $ | $ | $ | $ |
State the assumptions behind each estimate: volume, complexity, exception rate, model pricing, labour rates, and COLA adjustment rate.
| Assumption | Value | Source / Basis |
|---|---|---|
| Transaction volume (annual) | ||
| Automation rate (steady state) | % | |
| Exception rate | % | |
| Model cost per query (standard) | $ | |
| Model cost per query (grounded search) | $ | |
| Labour rate (fully loaded) | $ | |
| COLA adjustment | % annually | |
Show what a traditional three-layer TCO model (build, infrastructure, and consumption) would have estimated, compared to the percentage by which it understates true costs.
| Model | Three-Year Estimate | Difference |
|---|---|---|
| Naive TCO (Layers 1–3 only) | $ | |
| Full Nine-Layer TCO | $ | +___% |
| Category | Year 1 | Year 2 | Year 3 | Three-Year Total |
|---|---|---|---|---|
| Labour cost reduction / throughput increase | $ | $ | $ | $ |
| Error reduction | $ | $ | $ | $ |
| Other direct savings | $ | $ | $ | $ |
| Total direct savings | $ | $ | $ | $ |
Distinguish between headcount elimination and capacity expansion, as the economic models are different.
Value type: ☐ Headcount reduction ☐ Capacity expansion ☐ Both
| Category | Year 1 | Year 2 | Year 3 | Three-Year Total |
|---|---|---|---|---|
| [Specific mechanism] | $ | $ | $ | $ |
| [Specific mechanism] | $ | $ | $ | $ |
| Total revenue enhancement | $ | $ | $ | $ |
Quantify where possible. Qualitatively describe where not.
| Category | Quantified Value | Qualitative Description |
|---|---|---|
| Customer experience improvement | ||
| Regulatory posture improvement | ||
| Competitive positioning |
Value capture does not start at 100% on day one. Show the ramp from pilot to steady state.
| Period | Automation Rate | Value Capture (% of Steady State) |
|---|---|---|
| Month 1–3 (Pilot) | % | % |
| Month 4–6 (Early ramp) | % | % |
| Month 7–9 (Mid ramp) | % | % |
| Month 10–12 (Approaching steady state) | % | % |
| Year 2 (Steady state) | % | % |
| Year 3 (Optimised) | % | % |
| Metric | Low | Expected | High |
|---|---|---|---|
| Cumulative three-year value | $ | $ | $ |
| Cumulative three-year TCO | $ | $ | $ |
| Net value (Value minus TCO) | $ | $ | $ |
| ROI ratio (Value / Cost) | |||
| Payback period (quarter) |
How ROI changes if key assumptions vary by plus or minus 20%.
| Variable | -20% | Base Case | +20% |
|---|---|---|---|
| Automation rate | ROI: | ROI: | ROI: |
| Exception rate | ROI: | ROI: | ROI: |
| Model consumption cost | ROI: | ROI: | ROI: |
| Transaction volume | ROI: | ROI: | ROI: |
Map the deployment against the Agentic AI Risk Catalog. Identify which risks are relevant based on the institutional touchpoints with which the agent interacts.
Institutional touchpoints engaged:
☐ Customer-facing systems ☐ Internal workflows ☐ Financial systems ☐ Regulatory reporting ☐ Data infrastructure ☐ Third-party integrations ☐ Decision-making processes
For each relevant risk, assess across six dimensions (1–5 scale). Prioritise risks scoring 3.5 or above on any dimension.
| Risk | D | A | M | A | G | E | Average | Priority |
|---|---|---|---|---|---|---|---|---|
| ☐ High ☐ Med ☐ Low | ||||||||
| ☐ High ☐ Med ☐ Low | ||||||||
| ☐ High ☐ Med ☐ Low | ||||||||
| ☐ High ☐ Med ☐ Low | ||||||||
| ☐ High ☐ Med ☐ Low |
D: Detectability, A: Autonomy Sensitivity, M: Multiplicative Potential, A: Attack Surface, G: Governance Gap, E: Enterprise Impact
| Risk | Prevention Cost | Detection Cost | Expected Loss (Scenario) | Remediation Cost |
|---|---|---|---|---|
| $ | $ | $ | $ | |
| $ | $ | $ | $ | |
| $ | $ | $ | $ |
The specific governance infrastructure required to manage identified risks to acceptable levels, which feeds directly into TCO Layer 8.
| Governance Measure | Annual Cost | Risk(s) Addressed |
|---|---|---|
| $ | ||
| $ | ||
| $ | ||
| Total governance investment | $ |
After governance investment, what risk remains accepted? Be specific.
[Residual risk statement]
What regulatory and industry frameworks govern this deployment?
| Framework | Applicable | Key Requirements |
|---|---|---|
| EU AI Act | ☐ Yes ☐ No | |
| GDPR | ☐ Yes ☐ No | |
| MAS AIRG | ☐ Yes ☐ No | |
| SR 11-7 | ☐ Yes ☐ No | |
| DORA | ☐ Yes ☐ No | |
| [Other] | ☐ Yes ☐ No |
For each applicable framework, describe specific obligations and how the deployment addresses them.
| Framework | Obligation | How Addressed |
|---|---|---|
Risks that fall outside current regulatory coverage, with self-governance approach.
| Gap Area | Self-Governance Approach |
|---|---|
Anticipated regulatory changes within the planning window and architectural accommodations.
[Regulatory trajectory assessment]
| Phase | Duration | Start | End | Key Deliverables |
|---|---|---|---|---|
| Discovery | ||||
| MVP | ||||
| Production build | ||||
| Pilot | ||||
| Ramp | ||||
| Steady state |
| Dependency | Owner | Status | Required By |
|---|---|---|---|
| Data readiness | ☐ Ready ☐ In progress ☐ Not started | ||
| Integration availability | ☐ Ready ☐ In progress ☐ Not started | ||
| Governance infrastructure | ☐ Ready ☐ In progress ☐ Not started | ||
| Stakeholder availability | ☐ Ready ☐ In progress ☐ Not started | ||
| Budget release | ☐ Ready ☐ In progress ☐ Not started |
| Role | Phase(s) | FTE Allocation | Internal / External |
|---|---|---|---|
| Gate | Criteria | Decision Maker |
|---|---|---|
| Discovery to MVP | ||
| MVP to Production build | ||
| Production build to Pilot | ||
| Pilot to Ramp | ||
| Ramp to Steady state |
| Metric | Target | Cadence | Owner | Connects to |
|---|---|---|---|---|
| [Value capture line item] | ||||
| [Value capture line item] | ||||
| [Value capture line item] |
| Role | Responsible Party | Scope |
|---|---|---|
| Agent owner | ||
| Performance monitoring | ||
| Change approval | ||
| Incident response | ||
| Audit and compliance |
| Indicator | Threshold | Alert Action |
|---|---|---|
| Token consumption | ||
| Quality / accuracy | ||
| Exception rate | ||
| Latency | ||
| Error rate |
| Severity | Definition | Response Time | Escalation Path |
|---|---|---|---|
| Critical | |||
| High | |||
| Medium | |||
| Low |
Under what conditions would the organisation retire this agent?
| Trigger | Threshold | Decision Process |
|---|---|---|
| Performance degradation | ||
| Better alternative available | ||
| Changed business requirements | ||
| Regulatory change | ||
| Cost exceeds value |
☐ Proceed ☐ Proceed with conditions ☐ Defer
Rationale:
[Rationale tied to economic and risk analysis above]
State the exact approval, budget, resource allocation, or decision being requested.
[e.g., “We request approval of $_____ in year-one funding for [agent] deployment, with year-two funding of $_____ contingent on achieving [specific milestone] by [date].”]
| Timeframe | Action |
|---|---|
| Week 1 | |
| Month 1 | |
| First milestone |
Include as needed based on organisational requirements.
Template source: Corvair.ai (corvair.ai). Based on the frameworks described in the Economics of Agentic AI article series.
This template is part of a seven-article series on the economics of agentic AI in financial services.
Working through this template is the first step. Corvair advisors can guide you through the economic modelling, risk assessment, and governance planning to produce a document your board will trust.
Schedule a Briefing Read the Business Case Article