Search for AI chatbot pricing and you'll find ranges so wide they're almost useless. That isn't because vendors are hiding something — it's because two projects both called "an AI chatbot" can differ tenfold in scope. The useful question isn't "what does a chatbot cost?" but "what will ours need?"
This article breaks down the factors that actually move the price, the costs that continue after launch, and how to get a quote you can rely on. For the build process itself, see our custom AI chatbot development guide; for AI projects beyond chatbots — automation, integrations, agents — see how much it costs to develop an AI solution.
The seven factors that set build cost
1. How much content, and how messy
A few dozen clean web pages are quick to index. Thousands of scanned PDFs, spreadsheets with merged cells, or documents in several versions take real work: OCR, cleaning, deduplication and a chunking strategy that respects the structure.
2. Number of integrations
A standalone website widget is the simplest case. Each connection — CRM lead capture, helpdesk tickets, order lookups, Slack or Teams, a login system for personalised answers — adds design, build and testing effort.
3. The interface
An embeddable chat widget costs less than a full internal portal with search, document viewer and dashboards. Most projects also need an admin panel to upload documents, re-index and review conversations.
4. Hosting and data rules
Using a cloud LLM API is the fastest route. Deploying inside your own cloud account adds setup work; a fully on-premise build with open-source models adds infrastructure and model-serving effort. See hosting and privacy options.
5. Security and access control
Role-based access (so users only retrieve what they're allowed to see), audit logging and encryption are essential for confidential material — and each needs to be built and tested.
6. Accuracy requirements
A marketing FAQ bot can tolerate the occasional imperfect answer. A bot answering compliance, legal or medical-adjacent questions needs a proper test set, stricter grounding and more tuning time. Higher stakes mean more testing.
7. Languages and channels
Supporting several languages, or running the same assistant on the website, in Slack and through an API, multiplies testing.
| Scope | Typical shape | Relative effort |
|---|---|---|
| Focused chatbot layer | RAG on existing content, added to an existing app or site | Lower — around 3–4 weeks in our projects |
| Standard custom chatbot | Ingestion pipeline, chat widget, admin panel, one or two integrations | Medium — around 6–8 weeks |
| Multi-module AI platform | Several document libraries, workflows, case management, access control | Higher — around 10–12 weeks |
What you pay every month after launch
- Language model API usage. Billed per token (roughly per word processed). As a rough guide, 100–500 queries a day typically costs around $20–$80 a month; 1,000–5,000 queries a day around $100–$400 a month. You pay the provider directly with your own API key.
- Hosting for the application, plus the vector database or search index.
- Support and improvement — optional, but budget for someone to review unanswered questions and refresh content.
At higher volumes, caching common answers and routing simple questions to cheaper models keeps API costs down. Ask any developer to estimate running cost at your expected volume before you sign.
Costs people forget
- Content clean-up before the project — someone has to decide which documents are current.
- Subject-expert time to review answers during testing.
- Ongoing content maintenance as policies, prices and products change.
- Switching costs if you don't own the code or configuration.
Fixed price or retainer?
We offer two models. A fixed-scope project fee covers design, build and deployment, and you own everything at handover — best if you have a team to run it afterwards. A retainer suits evolving products, where we keep shipping features, tuning and handling infrastructure each month. Neither charges per user or per conversation; the only usage-based cost is the AI provider's API.
A short, fixed-price discovery engagement (1–2 weeks) reviews your data, defines the architecture and produces a realistic scope and cost for the full build — often with a working prototype on a subset of your documents. If you go ahead, the discovery fee is credited against the project.
How to compare chatbot quotes
- Give every vendor the same written scope and the same ten must-answer questions.
- Ask what's included: ingestion, admin panel, integrations, testing, launch support.
- Ask for the estimated monthly running cost at your volume.
- Confirm who owns the code, prompts and index at the end.
- Ask how the bot behaves when it can't find an answer.
- Ask to see an AI system the vendor runs in production — ours are in our case studies.
For an e-commerce store, the feature list on our e-commerce AI chatbot page is a good starting scope. For anything else, send us your use case and we'll come back with a scope and estimate.