TL;DR & Quick Summary
A custom AI chatbot costs about $8,000 to $150,000+ to build, depending almost entirely on what it has to know, who it has to recognise, and what it's allowed to do. Running costs come on top and scale with conversations. You can estimate them from token counts before you build.
| Chatbot type | Typical build cost | Typical timeline |
|---|---|---|
| Website FAQ bot answering from your content | $8,000–$25,000 | 3–5 weeks |
| Account-aware bot for logged-in customers | $25,000–$60,000 | 6–10 weeks |
| Action-taking bot (bookings, refunds, updates) with human handoff | $40,000–$100,000 | 10–16 weeks |
| Multi-channel enterprise assistant | $80,000–$150,000+ | 4–6+ months |
Running cost: tokens per conversation × price per token × conversations per month
Biggest cost drivers: account-specific answers, actions in other systems, channels and languages
The model choice is rarely the main cost. Integration, permissions and testing are.
Key Takeaway: Decide what the chatbot must do, not just what it should say. Each step from answering to recognising to acting moves you up a price band.
Get Started: Want a scoped estimate? Schedule a strategy call with Cogniq AI or see our AI customer support services.
Where Each Price Band Comes From
Like all software, a chatbot's build cost follows people × weeks × rate. A typical team is two engineers, with design and product help. At a blended $60–$90 per hour, each week of a two-person team costs roughly $4,800–$7,200. The timelines above set the bands. For how rate alone can triple a quote, see our AI MVP development cost guide.
Website FAQ Bot: $8,000–$25,000
Answers questions from your website, help centre and policy documents, citing its sources, and hands off to a person or a contact form when it can't help.
The work: ingesting and indexing content, keeping it current when pages change, answer quality testing on real customer questions, and the chat widget.
The trap: outdated content. If a retired policy is still indexed, the bot will eventually quote it.
Account-Aware Bot: $25,000–$60,000
Recognises logged-in customers and answers questions about their account: order status, bookings, invoices, plan details.
The work: authentication, permission checks so customers only ever see their own data, read integrations with your order, billing or booking systems, and testing every permission boundary.
The trap: one permission bug that shows a customer another customer's data costs more than the whole project. That's where most of the testing time goes.
Action-Taking Bot: $40,000–$100,000
Doesn't just answer. It changes things: reschedules a booking, issues a refund within limits, updates an address, creates a support ticket.
The work: write integrations, confirmation steps, limits on what can be done without a person, human approval for sensitive actions, audit logs, and a clean handoff to staff with the full conversation.
The trap: letting the bot act without limits. Start with low-risk actions and expand as the record shows they're handled reliably.
Multi-Channel Enterprise Assistant: $80,000–$150,000+
Runs across web, mobile app, WhatsApp, email or voice, in several languages, with integrations into several systems and enterprise security requirements.
The work: everything above, multiplied by channels and languages, plus single sign-on, data residency, security review and formal service levels. See WhatsApp AI customer support and AI voice agent cost for channel-specific costs.
Running Costs: Work Them Out Per Conversation
Language models charge per token for the text sent (input) and returned (output). In a chatbot, each turn usually sends the conversation so far plus retrieved knowledge, so input tokens grow as the conversation goes on.
Cost per conversation ≈ turns × (input tokens per turn × input price + output tokens per turn × output price)
For illustration only, assume $3 per million input tokens and $15 per million output tokens (check your provider's current rates, for example on Anthropic's pricing page):
- Each turn sends about 3,000 tokens and returns 300: $0.009 + $0.0045 = $0.0135
- A six-turn conversation: about $0.08
- 5,000 conversations a month: about $400 a month, plus hosting and search infrastructure
Ways to keep it down: retrieve only the most relevant passages rather than whole documents, summarise long conversations instead of resending them, use a smaller model for simple steps such as classifying the question, and cache common answers.
What Moves the Price Most
| Driver | Why it costs more |
|---|---|
| Account-specific answers | Authentication, permission checks, read integrations |
| Actions in other systems | Write integrations, approvals, audit logs, error handling |
| Number and condition of knowledge sources | Scanned PDFs and scattered documents take longer to make reliable |
| Channels | Each channel has its own formats, limits and testing |
| Languages | Content, testing and handoff in every language |
| Security and compliance | Single sign-on, data residency, reviews, logging |
| Answer-quality requirements | Larger test sets and more review before launch |
Custom Build or Chatbot Platform?
Subscription chatbot platforms are faster and cheaper to start with when you need a standard FAQ bot or simple lead capture. A custom build earns its cost when the chatbot must:
- Answer from your own systems, not just public content
- Recognise customers and respect their permissions
- Take actions in your CRM, booking or order system
- Meet security or data requirements a platform can't
- Run at a volume where per-conversation platform fees become significant
The three-year break-even method in custom AI agent vs SaaS works the same way for chatbots.
A Worked Budget Example
Here's how a first-year budget comes together for a typical mid-sized request: an e-commerce business wants a chatbot that answers product and policy questions, checks order status for logged-in customers, and hands over to staff when it can't help. That's an account-aware bot.
Build: two engineers for about 8 weeks:
- Weeks 1–2: gather the top 100 real customer questions from chat and email logs, design authentication and permission checks, connect read-only to the order system
- Weeks 3–5: knowledge indexing, order-status lookups, answer generation with sources
- Weeks 6–7: chat widget, handoff to the help desk with full context, failure handling
- Week 8: testing against the 100 questions plus permission edge cases, then release to a share of visitors
At $60–$90 per hour, 640 hours is about $38,000–$58,000.
Running: at the illustrative 8 cents per conversation, 5,000 conversations a month is about $400 a month in model usage. Add hosting and search, plus a few hours a month to review failed conversations and update content.
First-year total: roughly $45,000–$70,000, depending mainly on the rate and on how much ongoing review you do in-house.
How to Tell If It's Paying Off
Track these from the first week:
- Containment rate: conversations fully resolved without a person
- Escalation quality: whether handoffs arrive with enough context that staff don't need to ask again
- Customer satisfaction on bot conversations versus human ones
- Cost per resolved conversation, bot versus staff
- Repeat contacts: customers coming back with the same question means the answer didn't work
If containment is low, look at which questions fail and fix content or integrations before adding features. A chatbot that answers ten common questions perfectly is worth more than one that answers a hundred poorly.
Costs People Forget
- Keeping knowledge current. Someone must own updating content as products and policies change.
- Reviewing bad conversations. A weekly look at failed or escalated chats is how the bot improves.
- Handoff design. When the bot hands over, staff need the full context, or customers repeat themselves.
- Analytics. Containment rate, escalation rate, satisfaction and cost per conversation need building and watching.
- Maintenance. Model updates, API changes in connected systems and new products all need engineering time.
Fixed Price or Time and Materials?
Chatbot projects are quoted in two main ways:
- Fixed price: you know the cost upfront, which suits a well-defined FAQ bot. The risk is that anything not written into the scope becomes a change request, and quality can suffer if the provider underestimated.
- Time and materials: you pay for the hours used. This suits bots with integrations, where unknowns are more likely, as long as there are weekly demos and a budget cap.
A common middle ground is a short fixed-price discovery phase that produces the question set, integration map and acceptance criteria, then a build quoted from those. Either way, insist that the price includes testing against real questions and a staged launch. Those are the parts most often cut from low quotes.
How to Get an Accurate Quote
Before talking to developers, decide:
- The top questions and tasks the bot must handle, from real support tickets or chat logs
- Whether it must recognise customers, and which systems hold their data
- Which actions it may take, and which need a person
- Channels and languages for the first release
- How you'll judge quality: 50–100 real questions with the answers you'd expect
With those answers, quotes become comparable. If you're shortlisting providers, our guide to AI chatbot development companies covers what to ask them.
Conclusion
AI chatbot development costs about $8,000 for a simple website FAQ bot and $150,000 or more for a multi-channel enterprise assistant. The jump between bands comes from recognising customers and taking actions, not from the AI model. Estimate running costs per conversation before you build, budget for keeping knowledge current, and scope the first release around the questions and tasks your customers actually bring.
Schedule a strategy call with Cogniq AI for a scoped chatbot estimate, or explore our AI customer support services.
