The answer ranges from $0 to $250K+ per year, depending on what you're actually buying. The bigger question — and the one most companies get wrong — is whether you need a "chatbot" at all, or whether a properly-configured general AI tool like Claude does the job better at 5% of the cost.
Tier 1 — Off-the-shelf general AI ($0–$30/seat/mo): Claude, ChatGPT, or equivalent with custom Projects or GPTs configured for your use case. Best for internal workflows, customer-facing only when staffed by a human reviewer.
Tier 2 — Productized customer-facing platforms ($100–$1,000/mo): Intercom Fin, Drift, Tidio, etc. — chatbot platforms with AI built in. Designed for website/in-app deployment with reasonable out-of-the-box setup.
Tier 3 — Custom-built on AI APIs ($10K–$60K to build, $500–$5K/mo to run): A custom chatbot built on Claude or OpenAI APIs with your specific workflows, knowledge base, and integrations. Significant engineering required.
Tier 4 — Enterprise custom builds ($50K–$250K+ per year): Fully custom builds with deep CRM integrations, complex routing logic, compliance requirements. Typically for enterprise customer service or specialized vertical applications.
If you're under 50 employees: skip the dedicated chatbot. Configure Claude Projects (Tier 1) for the internal use case. For customer-facing, use a Tier 2 productized platform — don't build custom.
If you're 50–250 employees with complex CS volume: Tier 2 productized platform (Intercom Fin or equivalent) is the right answer. Roughly $500–$2,000/mo all-in. Custom builds at this stage usually don't pencil.
If you're 250+ employees with specific vertical or compliance requirements: Tier 3 custom build can pencil. Budget $30K–$60K for the initial build plus ongoing API costs.
Knowledge base maintenance. The chatbot is only as good as the docs feeding it. Budget 2–4 hours per week of ongoing content work to keep it useful.
Escalation paths. 100% of useful chatbots need a path to a human. Building that handoff well is a non-trivial design decision — and often where chatbots fail.
Hallucination management. Customer-facing AI that invents policy answers creates legal and brand risk. The mitigation work (guardrails, verification flows, restricted scopes) is real engineering.
Measurement. What does success look like? Deflection rate? CSAT? Time saved? Most companies don't define this up front and end up with a chatbot they can't evaluate.
Not quite. A chatbot is a specific UX (chat widget in a website or app). AI in customer service is a broader category — it includes chatbots, but also email response drafting, ticket triage, knowledge base authoring. For most companies, the broader AI deployment provides more value than the chatbot specifically.
For 95% of companies under 250 employees, buy. The productized platforms (Intercom Fin, etc.) have solved 80% of the build problems and ship in days vs. months.
Not directly out of the box. Claude is best as the engine behind a customer-facing platform — but you need the platform layer for things like conversation persistence, escalation handoff, brand UX, and rate limiting.
A chatbot answers questions in chat. An AI agent can take actions (look up an order, refund a payment, update a record). Agents are meaningfully harder to build well and meaningfully more valuable when they work — but the cost gap is significant: $50K vs. $5K for typical deployments.