What It Really Costs to Run an AI Feature (Per User)

The scariest line in an AI product budget is the one nobody can estimate: what the AI will cost to run once real users show up. Some founders discover it is a few cents per user. Others discover it is more than they charge.

You can know which one you are before you launch. Here is the simple formula, a worked example, and five ways to keep model costs under control.

The formula

Monthly AI cost per user = requests per user per month × average tokens per request × price per token

Tokens are the units AI models charge by. Roughly, a token is about three quarters of an English word. Most providers charge separately for input tokens (what you send) and output tokens (what the model writes back), and output usually costs more.

A worked example

These numbers are illustrative. Check your provider's current pricing page before you plan.

Say your feature summarizes customer support tickets:

• Each user summarizes 200 tickets a month
• Each request sends about 1,500 input tokens and gets back 300 output tokens
• Your model charges $3 per million input tokens and $15 per million output tokens

Input: 200 × 1,500 = 300,000 tokens, which is about $0.90 Output: 200 × 300 = 60,000 tokens, which is about $0.90

Total: about $1.80 per user per month. If you charge $49 a month, AI cost is not your problem.

Now say the feature reads 50-page PDFs instead. Each request might be 50,000 tokens or more, and the same math can quickly reach tens of dollars per user. That is when cost design matters.

Five ways to keep AI costs down

1. Cache results so the same input is never processed twice.
2. Use the smallest model that works for each step, and save the strongest model for the hard steps.
3. Send less. Trim documents to the relevant pages or sections before calling the model.
4. Set usage limits per plan, and show users how much they have left.
5. Log every call with tokens and cost from day one, so surprises show up in a dashboard, not on an invoice.

Frequently asked questions

How much does it cost to add AI to an app?

Building the feature is a one-time cost. Running it costs a few cents to a few dollars per user per month for many features, and more for document-heavy or high-volume use.

Are AI API prices going down?

Prices per token have generally fallen over time as new models are released, but check current pricing rather than assuming.

Should I charge more for AI features?

If AI usage varies a lot between customers, usage-based tiers or limits protect your margins.