"How much is a chatbot?" is roughly as answerable as "how much is a vehicle?". The word covers a decision tree that answers five questions and an agent that reads your documentation, checks your calendar and books the meeting. They cost very different amounts and they solve very different problems. This is how to tell which one you actually need before anyone quotes you.
The three levels, and what separates them
| Level | What it does | Price band |
| 1. Scripted FAQ | Answers a fixed set of questions from a decision tree. No understanding, no memory. | $2,500 – $4,000 |
| 2. Retrieval chatbot (RAG) | Reads your real documentation and answers in your own words. Handles phrasing it has never seen. | $4,000 – $6,000 |
| 3. Agent | Answers, and then acts: checks availability, books, updates the CRM, hands off to a human when it should. | $6,000 – $9,000 |
Why level 2 is where most companies should start
A scripted FAQ fails the moment a customer phrases something the way a human would, and every failure trains them not to bother next time. A retrieval chatbot reads what you have already written, so it answers questions you never thought to script.
- It uses documentation you already own. If you have a price list, a terms page and an FAQ, you have most of the input.
- It degrades gracefully. When it does not know, it says so and hands over, rather than confidently inventing an answer.
- It is auditable. You can see which source it drew from, which matters the first time it gets something wrong in front of a customer.
A real example, with the real numbers
We built a retrieval chatbot for Camping de Acedo, a campsite in Spain drowning in the same questions about availability, prices and facilities every peak season. It answers around the clock from the campsite's own documentation. The measured outcome for that client: 67% of enquiries resolved with nobody involved, response time down from about four hours to instant, and a 45% improvement in their satisfaction scores. Those are their numbers for their business, not an average we apply to everyone.
The comparison against hiring
The honest version of this comparison has to include the things a person does that a chatbot does not, so here it is both ways.
- A chatbot at level 2 costs roughly $4,000-6,000 once, plus small ongoing model and platform costs. It works nights, weekends and August. It never has a bad day and it never leaves.
- A person handling the same volume costs that much every few weeks, fully loaded, and covers about a quarter of the hours in a week.
- But a person handles the awkward case, the upset customer and the thing nobody wrote down. That is exactly why level 2 and 3 include a handoff: the goal is that a human only sees the conversations that need a human.
- The break-even is usually somewhere between six weeks and four months of the equivalent salary. If your enquiry volume is genuinely low, the honest answer is that you do not need this yet.
How the implementation actually goes
- Week 1: we gather your real documentation and map the questions you already get. Most of the work at this stage is deciding what the bot must never answer on its own.
- Week 2-3: build, connect to the channels you use, and wire the handoff. Guardrails first, features second.
- Week 3-4: testing against real historical questions, not invented ones. This is where you find out which of your documents contradict each other.
- Launch, then two weeks of watching. The first fortnight of real conversations teaches you more than the whole build.
What to insist on before you sign
- The guardrails in writing. Which topics it refuses, and what it says instead.
- Full access to the configuration and the sources, so you are never locked in.
- A named handoff path. A bot with no escape hatch is a customer service problem waiting to happen.
- A fixed price. Chatbot scope is knowable in advance; hourly billing here mostly signals that the agency has not scoped it.