AI Cost Estimation
Start with the unit you can price or defend: request, workflow, active user, ticket, developer, or account. Then estimate usage, component cost, fixed overhead, and the price floor in the same unit.
Question
How do I estimate AI software, agent, or workflow cost before production usage is fully known?
Quick answer
Formula: monthly_ai_cost = active_units * requests_per_unit_month * cost_per_request + fixed_monthly_ai_costs
Request cost formula: cost_per_request = generation + retrieval + reranking + vector_db + cache + infra
Price floor formula: required_price = cost_per_unit_month / (1 - target_margin_pct / 100)
- Assumption: choose one unit before estimating so requests, costs, and price use the same basis.
- Assumption: fixed costs such as embedding refresh or hosting are amortized across active units.
- Assumption: model p50 and p90 usage ranges before trusting a flat monthly price.
Example: 500 active users making 80 requests/month at $0.028 per request create $1,120 in variable spend. Add $180 fixed monthly overhead and total AI cost is $1,300, or $2.60 per active user. At an 80% target gross margin, the required price is $13.00.
Fastest Working Method
- Pick the unit that maps to the buying decision: user-month, ticket, workflow, developer, or account.
- Estimate requests per unit, then split cost per request into generation, retrieval, cache, and infra terms.
- Spread fixed monthly costs across the active unit count instead of hiding them in a vague overhead bucket.
- Run the same math for typical and heavy usage before setting caps, credits, or unlimited packaging.
- Convert unit cost into a required price at the target gross margin.
Worked Monthly Cost Example
A docs assistant has 500 active users. Each user averages 80 AI requests/month, and the current workflow costs $0.028 per request after generation, retrieval, reranking, cache, vector, and infra terms are summed.
Variable spend is 500 * 80 * 0.028 = $1,120. Add $180 for fixed embedding refresh, vector storage, and hosting. Total monthly AI cost becomes $1,300, or $2.60 per active user.
If the same feature needs an 80% gross margin, the cost floor is 2.60 / (1 - 0.80) = $13.00. That is the internal price floor before market willingness to pay, bundle value, discounts, and support cost.
Scenario Table
| Scenario | Unit | Starting Assumption | Estimated Cost | Next Step |
|---|---|---|---|---|
| AI SaaS feature | Active customer | 80 requests/customer-month at $0.028 per request plus $180 fixed refresh and infra | $2.60/customer-month before margin | Open |
| Support assistant | Resolved ticket | 2.8 AI requests/ticket at $0.035 per request before escalation labor | $0.10/ticket | Open |
| Coding workflow | Active developer | 110 requests/developer-month at a blended $0.044 per request | $4.84/developer-month | Open |
| Heavy user plan | Paid account | Typical users cost $8/month; 10% of users cost $80/month | $15.20 blended cost; $76 price floor at 80% margin | Open |
Open the Right Calculator
- Use AI Workflow Cost when the unknown is cost per request or cost per active user.
- Use Break-even Price when the unknown is the monthly price needed to hit margin.
- Use Compare Model Costs when the unknown is whether a cheaper or stronger model really improves margin.
- Use AI Agent Pricing when the hard question is credits, caps, heavy users, and usage billing.
Where Estimates Go Wrong
- Mixing request, user, and account units in one estimate.
- Using average users only, then discovering p90 users consume the margin.
- Counting token cost but skipping retrieval, reranking, cache, vector, or hosting overhead.
- Ignoring retry loops and fallback calls when quality is not stable.
- Treating the cost floor as the launch price instead of the minimum viable margin line.
Related reads
Start with baseline cost: AI Workflow Cost
Turn cost into price: Break-even Price
Compare model choices: Compare Model Costs
Plan caps and credits: How To Price AI Agent Usage With Credits, Caps, and Margin
Run the Calculator
Open the related calculator with your own assumptions before you compare infra, packaging, or rollout choices.
Open Related Calculator