AI Cost Benchmarking: What's Normal for Your Stage?

March 17, 2026

An investor asked us: "Is 18% of revenue on AI normal?" We didn't know. We dug into what other early-stage B2B SaaS teams were seeing. Rough ranges: 5–15% for early stage, 10–25% for growth. We were at 18%. Not crazy — but we had no idea if our distribution was healthy or if we were bleeding on certain segments. AI cost benchmark for SaaS: there's no universal number, but there are useful ranges.

AI spend benchmark: rough ranges we've seen

Early-stage B2B SaaS with AI features: often 5–15% of revenue on AI. Growth stage: 10–25%. If you're above 30% and not in a heavy AI-first product, it's worth digging in. AI spend normal for your stage depends on your product mix. A writing tool will have higher AI cost as % of revenue than a CRM with one AI feature. The benchmark is directional, not prescriptive.

The number that matters more than AI cost as % of revenue

The company-wide percentage is a starting point. The real question: are you profitable per customer? We had a 20% AI spend. Fine for customers with 70% gross margin. Terrible for customers where our margin was 25%. The average hid both. AI cost benchmark SaaS teams should care about: cost per customer and margin by tier. The blended number can be misleading.

How we benchmarked ourselves

We got cost per customer, cost per feature, and AI as % of revenue by tier. Compared tiers to each other. Our free tier: 100% AI cost, 0% revenue. Expected. Our Pro tier: some customers had AI costs approaching revenue. That was the problem. We raised prices on the heavy users and gated the expensive features. Margin improved. An AI cost benchmark that only looks at the company average misses the segments that need fixing.

Our free AI margin calculator helps you model this. For ongoing attribution, PerUnit connects AI costs to Stripe revenue.