“How much does a chatbot cost?” is the question behind almost every AI enquiry we receive, and the honest answer is a range wide enough to be useless without the breakdown behind it. The AI chatbot development cost in 2026 runs from under $15,000 for a simple FAQ bot to seven figures for enterprise multi-agent systems — and the difference is rarely the model. It is integrations, knowledge preparation, guardrails and evaluation. This guide breaks down what businesses actually pay, tier by tier, what drives the number up or down, and the running costs most quotes leave out. (If you are still deciding whether to build at all, start with our companion piece: AI chatbot for your website — build or buy.)
AI chatbot development cost in 2026, by type
- Simple FAQ bot — under $15,000. Answers from a fixed knowledge base, one channel (usually your website), no account actions. The right tool for “where is my order / what are your hours” traffic.
- RAG support chatbot — $30,000 to $150,000. Retrieval-augmented generation over your documents, help centre and policies, with cited answers and human hand-off. This is where most serious businesses land.
- Mid-market production build — $40,000 to $120,000. The typical range we see for a production system with CRM or helpdesk integration, guardrails, analytics and an evaluation set.
- Focused MVP — $30,000 to $60,000, shipping in 8–14 weeks. One channel, one knowledge domain, the integrations that matter, room to grow.
- Enterprise multi-agent systems — $80,000 to $1M+. Multiple specialised agents, write-back actions into core systems, compliance review, several channels and languages.
Notice what is not on that list: the price of the AI model itself. Model licences are a running cost (below), not the build cost. What you are buying in the build is software engineering around the model — and that is where the tiers separate.
What actually drives the price
- Integrations. Reading a PDF is cheap. Checking a live order status, raising a ticket in your helpdesk, updating a CRM record — each integration is design, build, testing and error-handling work. Two or three integrations routinely double an MVP budget.
- Knowledge preparation. Your documents are messier than you think: duplicates, contradictions, outdated policies. Cleaning and structuring the knowledge base is unglamorous, essential, and priced in days.
- Channels. Website widget, WhatsApp, in-app, voice — every surface adds UI, testing and edge cases.
- Guardrails and evaluation. A bot that can say anything is a liability. Production builds include refusal rules, PII handling, prompt-injection defences and a test set of hundreds of questions the bot must keep passing after every change. This is the line item cheap quotes skip — and the one that decides whether legal lets you launch.
- Languages and compliance. Multilingual support, data-residency requirements, sector regulation (health, finance) and audit logging all add real engineering time.
How long it takes
Timelines track the tiers. A focused MVP ships in 8–14 weeks. A serious production build with integrations, guardrails and evaluation typically takes 3–6 months. Enterprise systems with compliance and multi-channel deployment run 6–14 months. In our experience the schedule is rarely set by development speed; it is set by how fast the business can make decisions about knowledge, tone and escalation rules. Assign one owner on your side and the timeline holds.
The running costs nobody quotes
Build cost is the headline; running cost is the marriage. Budget for three things. LLM usage: at SMB volumes, $50–800 a month is typical — but model choice dominates everything. One 2026 analysis priced a 100,000-conversation month at roughly $960 on a budget-tier model versus roughly $24,000 on a frontier model. Routing most queries to a small model and escalating only hard ones is the single biggest cost lever, and it is an architecture decision you make at build time. Maintenance: knowledge updates, prompt tuning, reviewing failed conversations — a few hours a month for an MVP, a named role for enterprise. Infrastructure: hosting, vector storage and monitoring are usually the smallest line, but they are not zero.
RAG or fine-tuning? (It changes the budget)
For most business chatbots, retrieval-augmented generation (RAG) is the right and cheaper architecture: the bot answers from your documents, updates by editing a document instead of retraining, and needs far less ongoing engineering. Fine-tuning earns its cost only at very high query volumes — think 100,000+ conversations a day — or when you need relentlessly consistent tone and behaviour that prompting cannot hold. If a vendor quotes fine-tuning for a standard support bot, ask why.
From our experience: where budgets actually blow up
We build these systems — the assistant on this website is one of ours, with a lead-capture gate, model routing between a free-tier model and a paid fallback, and conversation limits to keep it sane. Two lessons repeat across projects. First, scope creep arrives disguised as integrations: “can it also check stock / book the engineer / chase the invoice” turns a chatbot into an automation platform. That can be the right call — it is most of what our AI integration services work is — but price it as a phase two, not a favour. Second, evaluation is not optional. The projects that launch on time are the ones where “good answers” were defined as a test set in week two, not argued about in week twelve. Cut features before you cut either of those, and the ranges in this guide will hold.
Frequently asked questions
What is the average AI chatbot development cost in 2026?
Most mid-market production builds land between $40,000 and $120,000. A focused MVP costs $30,000–60,000 and ships in 8–14 weeks. Simple FAQ bots start under $15,000; RAG chatbots run $30,000–150,000; enterprise multi-agent systems start at $80,000 and can exceed $1 million.
Is RAG cheaper than fine-tuning for a business chatbot?
Yes, for most business use cases. RAG costs less to build, updates by editing a document instead of retraining, and needs far less ongoing engineering. Fine-tuning only wins at very high query volumes (100,000+ a day) or when you need relentlessly consistent tone and behaviour.
How much do ongoing LLM API costs add to a chatbot?
At SMB volumes, $50–800 a month is typical, but model choice dominates: one 2026 analysis found a 100,000-conversation month costs about $960 on a budget-tier model versus about $24,000 on a frontier model.
How long does it take to build an AI chatbot?
A focused MVP ships in 8–14 weeks. A serious production build with integrations, guardrails and evaluation typically takes 3–6 months. Enterprise systems with compliance and multi-channel deployment run 6–14 months.
Should my business build a custom chatbot or buy a packaged one?
Buy when your knowledge base is small, your tickets are mostly FAQ and you do not need write-back actions. Build custom for proprietary integrations, multi-surface deployment, compliance packaged tools cannot meet, or when volume makes SaaS pricing uneconomical.
Want a number for your specific case instead of a range? Tell us what the bot needs to do and which systems it must touch, and we will scope it properly — start the conversation.