Every AI project needs to justify its cost. Not with hand-waving about strategic positioning, but with a specific, defensible calculation of what the project will cost and what it will return — in dollars, over a defined time period.
Here's the framework we use when building the business case for AI projects, and the calculation structure that CFOs and boards find credible.
Step 1: Quantify the Current Cost of the Workflow
Start with what you're automating. Document the current workflow cost precisely:
- Time per task — how many minutes does the workflow take per instance? Measure it, don't estimate.
- Volume per period — how many times per week/month does it occur?
- Fully loaded cost per hour — salary + benefits + overhead for the people who do this work (typically 1.4–1.6× base salary)
- Current total annual cost = (minutes per task ÷ 60) × volume per year × hourly cost
Example: A customer service team processes 2,000 order status inquiries per month. Each takes 8 minutes. The fully loaded cost per hour is $35. Annual cost = (8/60) × 24,000 × $35 = $112,000/year.
Step 2: Calculate the AI Automation Savings
AI agents rarely automate 100% of a workflow. A realistic estimate accounts for the portion the agent handles autonomously vs. the portion that still requires human involvement:
- Automation rate — what percentage of instances can the agent handle end-to-end? For well-scoped, high-volume workflows, 70–85% is achievable. Be conservative.
- Time reduction on remaining instances — even for instances that still need human involvement, AI-prepared context reduces handling time. Estimate this separately.
- Annual savings = (current annual cost × automation rate) + (remaining human cost × time reduction ×[1 - automation rate])
Using the example above: 75% automation rate, 40% time reduction on remainder.
Annual savings = ($112,000 × 0.75) + ($112,000 × 0.25 × 0.40) = $84,000 + $11,200 = $95,200/year.
Step 3: Calculate the Full Cost of the AI System
Use all four cost components (not just build cost):
| Cost Component | Year 1 | Year 2 | Year 3 |
|---|---|---|---|
| Build cost (one-time) | $45,000 | — | — |
| LLM API costs | $6,000 | $8,000 | $10,000 |
| Infrastructure | $3,600 | $3,600 | $3,600 |
| Maintenance (15% of build) | $6,750 | $6,750 | $6,750 |
| Total cost | $61,350 | $18,350 | $20,350 |
Step 4: Build the 3-Year ROI Model
| Metric | Year 1 | Year 2 | Year 3 |
|---|---|---|---|
| Annual savings | $71,400* | $95,200 | $95,200 |
| Annual cost | $61,350 | $18,350 | $20,350 |
| Net annual benefit | $10,050 | $76,850 | $74,850 |
| Cumulative net benefit | $10,050 | $86,900 | $161,750 |
*Year 1 savings discounted 25% to account for ramp-up period.
Payback period: approximately 8 months after go-live. 3-year ROI: 263%.
The Three Objections to Prepare For
"Your automation rate is too optimistic."
Counter: document how you arrived at the rate. If comparable deployments at other companies achieved X%, cite them. If you're using conservative estimates, show your work. Offering to use a lower rate as a sensitivity analysis defuses this objection — if it's 60% instead of 75%, the project still pays back in 13 months.
"Staff savings don't actually reduce headcount."
Counter: frame it as capacity creation, not headcount reduction. "This frees 2,800 hours per year currently spent on order status inquiries — capacity that can be redirected to higher-value interactions." Many AI projects are approved on capacity creation even when headcount reduction isn't the goal.
"What if it doesn't work as promised?"
Counter: define the production success criteria and the decision gate. "If the system doesn't achieve X% automation rate on 500 live transactions within 60 days, we evaluate whether to continue or redeploy." A defined off-ramp reduces perceived risk significantly.
The presentation principle: Lead with the business outcome, not the technology. "This project will save $95,000 per year by automating 75% of order status inquiries" lands differently than "this project will deploy an AI agent using RAG architecture and LLM orchestration." Decision-makers approve business outcomes. Finance approves payback periods. Your job is to translate the technology into those terms.
Need Help Building the Business Case for Your AI Project?
We help clients build CFO-ready ROI models for AI initiatives — with documented assumptions, sensitivity analysis, and realistic projections based on comparable deployments.
Talk to the Team