The real numbers
How much does an AI chatbot really cost?
Far less than most quotes suggest — and far more than "the API is cheap" implies. A chatbot's running cost has three layers: model tokens (every reply reads the system prompt, the chat history and — for knowledge-base bots — your retrieved documents), infrastructure (vector database, hosting, the workflow tool that orchestrates it), and the one-time build. The calculator above prices all three; the defaults describe a typical support bot: 2,000 conversations/month on a mini-tier model with a document knowledge base, landing near $110/month — about 5–6 cents per conversation.
Chatbot cost per month: what actually drives it
- Model choice — the single biggest lever. A mini-tier model answers most support questions at 5–10× less than a flagship. Route hard cases up instead of paying flagship prices on every "where's my order?".
- Input tokens, not output — chat history and retrieved context make chatbots input-heavy. Summarize old turns and cap retrieved chunks; that's where the bill grows.
- Volume scales linearly — 10× conversations ≈ 10× token bill, but your platform and vector DB costs barely move. Busy bots get cheaper per conversation.
Chatbot cost to build (one-time)
A simple FAQ bot on an existing platform is a $500–2,000 build. A knowledge-base (RAG) bot that ingests your docs, retrieves and cites them runs $2,000–10,000 depending on integrations and polish. In-house, count your hours honestly — including the prompt-tuning week everyone forgets. Whether that build pays off is exactly what our automation ROI calculator answers.
The knowledge-base layer (RAG)
Bots that answer from your documents add two costs: extra input tokens (the retrieved chunks pasted into every prompt — often more tokens than the question itself) and a vector database. At small scale, serverless vector DBs like Pinecone cost $0–50/month. Model that layer precisely — corpus size, embeddings, storage, queries — in the RAG cost calculator.
Chatbot platforms vs building your own
Per-seat chatbot SaaS often charges $50–500/month for what tokens deliver at $20–100. Building on a workflow tool like n8n plus a model API keeps you in control of model choice and cost per message — that's the stack this calculator models, and the one we build for clients.
Worked example
The defaults above: 2,000 conversations × 6 messages on GPT-5.4 mini with RAG. Token bill ≈ $38/month (32.4M input + 3M output tokens), vector DB $25, platform $50 → ~$113/month, or 5.6¢ per conversation. First year with a $1,500 build: ~$2,850. Compare that to one part-time support hire — then check the ROI calculator to see the payback in hours saved.