← Free AI Tools

AI Margin Calculator & Benchmark

See how your AI spend impacts your margins — cost per customer, AI as a percentage of revenue, break-even — and how your numbers compare to typical SaaS benchmarks below.

Your numbers

Your combined OpenAI / Anthropic / Google bill

Total MRR from paying customers

Paying customers (exclude free tier)

AI is one featureEntire product is AI

Your AI unit economics

AI gross margin

62.5%

Healthy

AI cost per customer

$100.00

per month

Revenue per customer

$333.33

per month

AI as % of revenue

30.0%

Within normal range

Break-even revenue

$15,000.00

to cover AI spend

AI spend vs attributable revenue

AI spend$12,000.00
AI-attributable revenue$32,000.00

This calculator uses blended averages. Actual margins vary significantly by customer — some may be profitable, others not.

How your numbers compare — typical SaaS benchmarks

Rough benchmarks compiled from public reports and customer conversations. The right benchmark depends on whether AI is the entire product or one feature among many.

Stage / modelAI as % of revenueHealthy AI gross margin
Pre-PMF / heavy free tierYou are here
30–60%20–50%
Early B2B SaaS (AI as a feature)
5–15%60–80%
AI-native product (LLM is the product)You are here
15–35%50–70%
Mature, optimized AI product
8–20%70–85%

Your current AI spend is 30.0% of revenue with a 62.5% AI gross margin. The highlighted row shows where most companies in your range tend to live.

A blended margin hides the customers losing you money.

Most products with a healthy-looking blended margin still have 15–30% of customers running at negative margin individually. The average is fine. The distribution isn't. PerUnit breaks the margin out customer by customer so you can see which side of the line each one actually sits on.

Get early access to PerUnit

Frequently asked questions

What's a healthy AI gross margin?
For most companies, 60–80% on the revenue AI features generate. Below 50% is worth investigating; below 30% usually means either the wrong customers are getting the heaviest features or pricing hasn't caught up with usage. Look at the benchmark table above for where teams at your stage typically land.
What does "AI cost per customer" actually tell me?
It's a blended average — total monthly AI spend divided by paying customers. Useful as a sanity check ("are we in the $5/customer or the $50/customer ballpark?") but the real value comes from breaking it down per customer, because the spread is almost always wider than the average suggests.
How do I reduce AI costs without hurting the product?
Four levers, roughly in order of impact: (1) identify the 10–20% of customers driving most of the cost — they're often on the wrong tier; (2) route simpler tasks to cheaper models (mini/Flash/Haiku) instead of routing everything to a flagship; (3) gate high-cost features behind paid plans or per-tier limits; (4) prompt caching for any context you send repeatedly.
Is it normal for free users to cost more than paying customers?
Yes, and it's the most common pattern we see. Free users often have the same access as paid users with zero revenue contribution; in several products we've looked at, the free tier consumes 40–60% of total AI spend. The fix is usually a feature gate or a per-customer monthly cap on the AI features specifically — not killing the free tier.
What percentage of revenue should AI costs be?
Under 15% of AI-attributable revenue is healthy for most teams. 15–30% is normal for AI-native products where the model is the product. Above 30% on a blended basis usually means you have a small group of customers running well above 60% individually — the average hides who they are.

Get a monthly email when AI model pricing changes

One email a month. New model launches, price cuts, deprecations across OpenAI, Anthropic, and Google. No spam, unsubscribe any time.