The quote you get for an AI chatbot covers the build. It doesn't cover keeping it alive. The running cost is the number that actually shows up on the card every month—and depending on volume, that's $200 to $10,000.
TL;DR
- Monthly running cost splits into four buckets: model API ($20–$6,000), hosting and infra ($200–$3,000), vector database ($50–$2,000), and human upkeep ($1,500–$5,000 if you want it actively maintained).
- A low-volume FAQ bot can run under $300/month. A high-volume qualifying chatbot at 10K+ conversations runs $2,000–$10,000/month, most of it tokens.
- The line item vendors skip: maintenance. A $50K chatbot typically needs $7.5K–$10K/year in updates, retraining, and integration fixes just to keep working.
- The biggest lever is architecture: caching and model routing can cut API cost 60–70%, so ask how your vendor handles it before you sign rather than after.
- Budget the run rate before the build. A chatbot you can't afford to operate is worse than no chatbot.
The four things you pay for every month
Running cost isn't one number. It's four, and they scale differently.
1. Model API (the LLM itself)
Every message in and out is billed by the token. At low volume this is trivial; at high volume it's your biggest line item, and often the only one worth optimizing.
| Monthly conversations | Model | Rough API cost |
|---|---|---|
| ~5,000 | GPT-4o-mini / small model | $20–$75 |
| ~5,000 | Frontier model (GPT-4o-class) | $200–$800 |
| ~10,000 | Frontier model | $1,500–$6,000 |
The 30x gap between the cheap and frontier rows is the most important cost decision you'll make. Route simple, early messages to a small model and save the frontier model for the hard turns—that's where the savings are, and it's purely an architectural decision.
2. Hosting and infrastructure
The app itself, the API gateway, auth, logging, monitoring. Realistically $200–$3,000/month depending on scale. Skimping here is how you end up with a chatbot that breaks silently and nobody notices for a week.
3. Vector database (if it uses RAG)
Any chatbot that answers from your own content needs a vector store: $50–$2,000/month depending on corpus size. Small product-docs index is cheap; large, frequently updated knowledge base is not.
4. Human upkeep: the line that gets skipped
This is the one missing from most quotes. A chatbot isn't a set-and-forget asset. Content drifts, models get deprecated, integrations break when a CRM updates its API. Actively maintaining a production chatbot runs $1,500–$5,000/month. A $50K chatbot typically needs $7,500–$10,000/year just to keep working as built—before you've improved anything.
Monthly totals at two volumes
| Low-volume FAQ bot | High-volume qualifying bot | |
|---|---|---|
| Conversations/mo | ~2,000 | ~10,000 |
| Model API | $30–$120 | $1,500–$6,000 |
| Hosting + infra | $200–$500 | $800–$3,000 |
| Vector DB | $50–$200 | $300–$1,500 |
| Active maintenance | minimal / DIY | $1,500–$5,000 |
| Monthly total | $300–$800 | $4,000–$10,000+ |
The low-volume bot is cheap and you can mostly leave it alone. Once you're at 10K conversations a month, it's a real operating line—and at scale, tokens and people dominate, not hosting. Optimizing your AWS bill is a rounding error. Optimizing model routing is thousands of dollars.
What most people get wrong
They budget the build and forget the run rate. The conversation is always "what does it cost to build a chatbot," almost never "what does it cost to keep it running for three years." So the project gets approved on a $30K number, and the $3K/month operating cost shows up as a surprise in month two.
Do the three-year math before you sign. $30K to build plus $3K/month is a $138K commitment over three years. That's often still the right call—but only if you made it with the actual number. The off-the-shelf comparison flips on exactly this math, which is why the build vs. buy decision turns on running cost, not sticker price.
The maintenance thing trips people up too. A chatbot that worked at launch and got ignored is, eighteen months later, citing docs that no longer exist and calling a CRM endpoint that changed. It's quietly losing you leads, and nobody flagged it.
Keeping the cost down
The biggest lever is how you route model calls. Simple early turns go to a cheap model; the frontier model handles the hard ones. Cache answers to repeated questions instead of regenerating them every time. Done right, that's 60–70% off your API bill with no quality drop. Ask any vendor specifically how they do this—if the answer is vague, they don't.
Second, match the model to what you actually need. A well-built FAQ bot on a small model with good retrieval will often beat a frontier model with sloppy context. You don't need the most expensive model; you need the right one.
Third: nail down who maintains it and at what frequency before you launch. Monthly upkeep on a schedule is cheaper than fixing things after they've been broken for two weeks and a customer noticed.
Before you approve the build
Run the three-year number. A chatbot you can't afford to operate isn't an asset. The cheapest one to run long-term is the one that was built for efficiency from day one—not the one with the lowest build quote.
Next step: See how running cost factors into the full AI chatbot cost breakdown, or compare it against off-the-shelf monthly fees in custom vs. platform chatbot pricing.
How much does it cost to run an AI chatbot per month?+
Between $200 and $10,000+ per month, driven by volume. A low-volume FAQ bot (~2,000 conversations) runs $300–$800/month across API, hosting, and vector database. A high-volume lead-qualifying bot (~10,000 conversations) runs $4,000–$10,000+/month, with model API tokens and active maintenance making up most of the bill.
What are the ongoing costs of an AI chatbot besides the build?+
Four recurring buckets: model API (token cost, $20–$6,000/month by volume), hosting and infrastructure ($200–$3,000/month), a vector database if it uses RAG ($50–$2,000/month), and human maintenance ($1,500–$5,000/month for active upkeep). Maintenance is the one most often left out of vendor quotes.
How much does chatbot maintenance cost per year?+
A custom chatbot typically needs 15–20% of its build cost annually in maintenance, so a $50,000 chatbot runs $7,500–$10,000/year for security patches, model updates, retraining, and fixing integrations when connected systems change their APIs. Skipping maintenance doesn't save money; it degrades answer quality and breaks integrations over time.
What's the biggest cost driver in running a chatbot at scale?+
At high volume, model API tokens and human maintenance dominate, not hosting. That's why architecture matters: routing simple messages to a cheaper model and caching repeated answers can cut API cost 60–70%. Optimizing infrastructure saves comparatively little once you're past a few thousand conversations a month.
Is it cheaper to run a custom chatbot or an off-the-shelf one?+
Off-the-shelf platforms charge a predictable monthly fee ($500–$2,000+) with no infrastructure to manage, while a custom chatbot runs $200–$800/month at low volume but rises with usage and needs maintenance you arrange yourself. Custom usually wins on per-conversation cost at scale; platforms win on simplicity and predictability at low volume. The decision turns on your conversation volume and how specific your needs are.
Written by
Pankaj Kumar
Founder · Metageeks Technologies
Metageeks builds production-ready AI products for $1M–$15M companies — shipped in fixed-price sprints, not open-ended retainers. We write about what actually works in the field.
Connect on LinkedInThe AI Build Brief
Ship AI that actually works.
Practical playbooks on building, pricing, and shipping production AI — one email, every other week. No fluff.
