Blog
Adding AI to a SaaS product is the easy part. Pricing it without crushing margins is the hard part. A feature that costs $4/user/month in tokens cannot live in a $10/month base plan. Here is the pricing math that keeps AI features profitable.
Your AI feature has three costs you can control and one you can't:
A safe rule: your blended AI delivery cost is roughly 1.5x your raw token cost. Plan around that, not the token cost alone.
For SaaS AI features to fund growth (sales, marketing, R&D), you want gross margin of 70%+. That means the price needs to be at least 3.3x the delivery cost.
| Token cost / user / mo | Min addon price | Recommended price | Margin at recommended |
|---|---|---|---|
| $0.10 | $0.50 | $2.00 | 92% |
| $0.50 | $2.50 | $5.00 | 85% |
| $1.00 | $5.00 | $10.00 | 85% |
| $3.00 | $15.00 | $20.00 | 78% |
| $8.00 | $40.00 | $49.00 | 76% |
| $20.00 | $100.00 | $129.00 | 77% |
If you can't justify the recommended price to your customers, the feature is not viable on the model you picked. Drop to a cheaper model and recalculate.
Find your per-user cost across every model — pick the one that fits your price.
Open AI Cost Calculator →1. Bundled in higher tier. Include AI features in your $49 or $99 plan, not the $9 starter. The 5-10x price jump funds the AI cost without making the base tier expensive.
2. Usage credits with overage. Each tier includes N AI credits per month. Going over costs $X per credit. Common ratio: include enough credits to cover average use, charge 2x cost for overage.
3. Per-seat with fair use. Charge per user with a soft cap (1,000 AI requests/user/month) and a hard cap (5,000). 95% of users stay under the soft cap. The 5% who blow through subsidize themselves into a higher plan.
4. Standalone AI addon. Sell the AI feature as a separate $X/month addon to the base subscription. Customers self-select into the addon, and you can adjust pricing without touching base SKUs.
1. "Unlimited AI" in any plan under $100/mo. One whale will eat all your margin. There is no defensible "unlimited" at the cheap end.
2. Free AI in the free trial. 14-day trials with unlimited AI burn cash. Cap the trial at 50-200 AI requests. If users want more, they convert.
3. Adding AI to base plan without raising the price. Your existing customers will love it. Your gross margin will collapse. Either grandfather and add AI to a new tier, or raise the base price.
4. Charging less than 2x cost. Below 2x, support burden and edge-case spikes will eat your margin. There is no safety net.
Imagine you sell a $49/month help-desk SaaS. You want to add AI ticket summarization. Each ticket summary uses ~2,500 input tokens (the conversation) and ~150 output tokens. Average customer processes 200 tickets/month.
Per customer: 200 × (2,500 input + 150 output) = 500,000 input + 30,000 output tokens.
| Model | Per customer / mo (token cost) | Cost as % of $49 plan |
|---|---|---|
| Gemini 2.0 Flash | $0.06 | 0.1% |
| GPT-4o mini | $0.10 | 0.2% |
| Claude Haiku 3.5 | $0.52 | 1.1% |
| Gemini 2.5 Pro | $0.93 | 1.9% |
| GPT-4o | $1.55 | 3.2% |
| Claude Sonnet 4 | $1.95 | 4.0% |
| Claude Opus 4 | $9.75 | 19.9% |
On a cheap model, the AI summary costs almost nothing — you can include it in the base plan. On Claude Opus, it eats 20% of your plan price, which is an addon, not a feature. The model you pick determines whether the feature is "free" or "$10/mo addon."
Calculate your per-user AI cost across every major model.
Open AI Cost Calculator →