Founders ask this constantly, and the honest answer is 'it depends on what you measure'. There are reasonable benchmarks, but for AI products the more important question isn't the headline number — it's whether that number hides a tail of unprofitable customers.
The benchmarks
Mature software targets 70–80% gross margin. AI-heavy products typically run lower because of the token cost behind every request — industry surveys often put them well under that classic range, sometimes around 50%. As rough anchors: above 70% is healthy and fundable, 50–70% is workable with efficient pricing and growth, and below 50% means model choice or pricing needs to change.
Why AI runs lower
The difference is variable cost. Classic SaaS approaches zero marginal cost per customer; AI SaaS pays for tokens on every interaction. Output tokens cost several times more than input, and a single power user on a flat plan can consume 10–50× the tokens of a light one. That variance is why an AI product's margin is far more sensitive to usage — and pricing — than a traditional one.
Per-customer beats blended
A blended 65% can look perfectly healthy while a fifth of your customers are individually losing you money. Blended margin averages the winners and losers together; per-customer margin shows you the losers. For an AI SaaS the per-customer view is the one that predicts whether growth helps or hurts — because adding more customers like your unprofitable ones makes the average worse, not better.
How to improve it
Three moves, in order: see it (track cost and margin per customer, not just blended), fix the model economics (trim input, control output, route to cheaper models), and align pricing (usage limits or higher tiers for the heavy accounts that break your math). The first move is the one most teams skip — and the one that makes the other two possible.
Check the margin on a single customer in ten seconds with the free calculator, or connect Stripe and your LLM cost with MarginWard to see every customer's margin at once.