What AI really costs a small business
AI pricing looks cheap per request and confusing in total. There are four cost layers, and most budgets only count the first one.
1. Usage costs (tokens)
AI model providers usually charge per million tokens — chunks of text roughly three-quarters of an English word long. Input tokens (what you send) and output tokens (what comes back) are priced separately, and output is typically several times more expensive.
Worked example: a support assistant handles 2,000 requests a day, each sending 1,200 input tokens and receiving 400 output tokens. At an illustrative $3 per million input tokens and $15 per million output tokens, each request costs about $0.0096 — roughly $19 a day or $576 a month. Prices vary widely between providers and models and change often, so plug in your provider’s current rates.
2. Subscriptions
Seat-based AI assistants and AI features inside existing software are usually billed per user per month. They are predictable but add up across a team, and overlapping tools are common. Audit them twice a year.
3. Setup and integration
Connecting AI to your email, help desk or database takes time: prompts, testing, permissions and edge cases. For a small team this is often a one-off project measured in days or weeks of work, and it is the cost most often left out of ROI estimates.
4. Human review — the hidden cost
AI output needs checking, especially anything customer-facing or financial. If a task took 10 minutes manually and now takes 2 minutes to generate plus 4 minutes to review, the real saving is 4 minutes, not 8. Measure review time during your pilot.
Putting it together
| Cost layer | How it’s billed | How to control it |
|---|---|---|
| Tokens | Per million tokens | Shorter prompts, smaller models for simple tasks, caching |
| Subscriptions | Per seat per month | Give seats to heavy users only; remove overlap |
| Setup | One-off time or contractor fee | Pilot one task before building more |
| Review | Staff time | Templates, checklists, spot-checks instead of full review where safe |
Is it worth it?
Use a simple test: net annual time savings (after review) multiplied by the loaded hourly cost of the people involved, minus all four cost layers. If the result is at least three times the tool cost, roll it out. The AI time-savings ROI calculator does this in one step, and the AI vs new hire calculator helps when you are weighing automation against headcount.