How to build an AI API budget your CFO will approve

Why AI budgets fail

Most AI API budgets fail because they start from the wrong direction. Teams estimate a monthly number, get approval, and then watch the actual spend diverge within weeks.

The problem: AI spend scales with usage, not headcount. A team of 10 can spend $500/month or $5,000/month depending on model selection, prompt length, and caching habits. Traditional per-seat budgeting does not work here.

The bottom-up framework

Build your budget from actual usage data, not guesses.

Step 1: Measure current baseline. Track API calls for two weeks across all providers. Count calls per developer per day, average tokens per call, and model distribution.

Step 2: Calculate unit economics. What does one API call cost on average? What does it cost per task type? This gives you a cost-per-unit that finance can model against.

Step 3: Project by growth. Multiply your unit cost by projected call volume. Factor in new hires, new use cases, and expected efficiency gains from optimization.

Step 4: Build in optimization targets. Show the CFO two numbers: current-trajectory spend and optimized spend. The gap between them is the ROI of an optimization tool.

The template

Present your budget with four line items:

• -Flagship models (Opus, GPT 5.5, Fable 5): complex reasoning, code generation
• -Mid-tier models (Sonnet, Terra, Gemini Pro): code review, summarization, QA
• -Fast models (Haiku, Luna, Flash): classification, extraction, routing
• -Optimization savings (negative line item): projected reduction from caching, model routing, prompt optimization

Sample budget for a 20-person engineering team

• -Flagship (20% of calls): 6,000 calls/day x $0.12/call x 22 days = $15,840/month
• -Mid-tier (50% of calls): 15,000 calls/day x $0.04/call x 22 days = $13,200/month
• -Fast (30% of calls): 9,000 calls/day x $0.01/call x 22 days = $1,980/month
• -Subtotal: $31,020/month
• -Optimization savings (caching + prompt discipline): -$9,300 (30% reduction)
• -Net budget: $21,720/month

What the CFO wants to hear

Finance cares about three things: predictability, accountability, and reduction trajectory.

• -Predictability: per-call unit costs from published rate cards
• -Accountability: per-team and per-developer attribution
• -Reduction trajectory: concrete plan to reduce cost per unit over time

If you can show a declining cost-per-call trend alongside growing usage, the budget conversation becomes easy.

Build your cost baseline at /teardown or start tracking with the free dashboard.