The real cost of an AI chatbot for a small business (spreadsheet included)
What an AI chatbot actually costs to run in 2026. Line by line, no vendor math, no hidden items.
When someone quotes "AI chatbot for $49/month," the useful follow-up is always: where's the rest of the bill? Most of the time there's a material line missing. Let's walk through what a realistic cost ledger looks like.
Fixed costs (whoever owns the site pays these)
| Item | Monthly | One-time |
|---|---|---|
| Domain + email | $2 | — |
| VPS hosting (small) | $6–20 | — |
| SSL certificate | $0 (Let's Encrypt) | — |
| Small language model (self-hosted) | $0 | — |
| Setup labor | — | $200–1,000+ |
Variable costs — the part that breaks "cheap chatbot" quotes
If the chatbot uses a cloud LLM API instead of a local model, pricing scales with traffic:
| Workload | Rough monthly |
|---|---|
| 300 chats/month × ~5k tokens | $12–25 |
| 1,500 chats/month | $60–125 |
| 5,000 chats/month (busy dentist) | $200–400 |
A $49/month "flat rate" that routes through an OpenAI API is cheap until volume arrives. At which point it stops being flat.
The hidden cost: your time
The line item nobody prices: the human hours to:
- Review conversations for wrong answers
- Update the FAQ when hours or services change
- Follow up on leads the bot captures
Budget 1–2 hours a week for the first month, less after. An admin panel that lets non-technical owners make FAQ edits reduces that to minutes.
Local vs. cloud — when each wins
Cloud LLM API wins when:
- Volume is low (under 500/month) and you don't want to manage a box
- You need frontier-model quality on rare long-form questions
- Data sensitivity isn't a concern
Local model wins when:
- Volume is sustained (500+/month) and total cost matters
- Privacy is a hard constraint (HIPAA, attorney-client, payment data)
- You want predictable cost regardless of usage
Most small-business FAQ chatbots fit the "local wins" column. Your mileage will vary.
— Alex Kargin. More engineering writing at kargin-utkin.com.