What a complete chatbot build includes
A quote should cover four things. One that omits any of them is pricing a demonstration rather than a system.
- Grounding. The assistant answers from your own documentation, with a citation on every answer so a reader can check it and so can you.
- Handoff. A designed route to a person, with the conversation and its context carried across rather than restarted.
- Channels. Web, in product, or inside the helpdesk your agents already use. Each additional channel is real additional work.
- Analytics. Reporting on what customers actually asked and where the assistant failed them, which is what lets the thing improve after launch.
What pushes chatbot cost up
The state of your content
An assistant grounded in well maintained documentation is inexpensive to build. One grounded in content that is contradictory, out of date or scattered needs that resolved first, and that work is frequently larger than the assistant itself.
Answering versus acting
Answering questions is cheap. Taking actions is not. An assistant that checks an order, raises a ticket or processes a return needs each of those integrations designed with permissions and failure behaviour, and each one adds meaningfully to the build.
Regulatory exposure
In healthcare, legal and financial contexts what the assistant may say is constrained, every exchange has to be logged, and the escalation path has to be defensible. That governance is a real share of the cost and it is not optional.
Languages and channels
Each language adds evaluation work rather than only translation, because answer quality has to be measured separately in each one.
Why is my chatbot quote so much cheaper elsewhere?
Usually because it prices a widget rather than a system. The common omissions are grounding, so the assistant answers confidently from the wrong source; a handoff, so customers reach a dead end; evaluation, so nobody can say whether a change improved anything; and logging, so nothing can be reviewed afterwards. A chatbot switched off in month three costs more than the one that was quoted properly.
How to reduce the cost of a chatbot build
- Start with the questions you already answer most. Your support tickets tell you what to ground the assistant in, and narrow grounding is both cheaper and more accurate.
- Launch on one channel. Prove answer quality in one place before paying to maintain it in four.
- Answer before you act. Ship an assistant that answers well, then add the integrations that let it do things.
- Fix the content first. Time spent making documentation correct reduces the build and improves the result, and you own that work regardless.
How long does an AI chatbot take to build?
The timeline is driven mainly by how ready your content is to be grounded on, not by the assistant. You see it answering your own questions from your own content well before it meets a customer, so the launch date is set by observed quality rather than by a plan drawn at the start.
Related
- What AI development costs: how pricing works across every engagement model.
- AI chatbot development: what the build actually covers.