AI Chatbot Cost Breakdown
Cost to Build an AI Chatbot
The build cost is the easy part. The running cost is what catches people.
A chatbot that answers from your own documents is a retrieval problem with a language model attached, and it has a monthly bill.
Quick Answer
Updated August 22, 2026
How much does it cost to build an AI chatbot?
An AI chatbot costs $10,000–$45,000 (₹8–37 lakh) to build through an Indian agency in 2026, plus $200–$3,000 a month in model and infrastructure costs depending on volume. A scripted assistant over a small, fixed knowledge base sits near $10,000–$18,000; a retrieval-augmented assistant over your real documentation, with evaluation, guardrails, and human escalation, runs $25,000–$45,000. The evaluation harness is the part most quotes omit and the part that determines whether the thing is usable.
Build range
$10,000 – $45,000
Monthly running cost
$200 – $3,000
Typical timeline
6 – 16 weeks
Best for
- Support teams answering the same questions repeatedly
- Companies with substantial written documentation to draw on
- Products where deflecting even 30% of tickets is material
Not best for
- Businesses with no written knowledge base to retrieve from
- Questions where a wrong answer carries legal or safety risk
- Teams unwilling to review and correct answers over time
Price Bands
What an AI chatbot costs, by scope
Bands rather than a single figure, because a single figure without the assumptions behind it is a sales number rather than an estimate.
| Tier | Scope | INR | USD | Timeline | Team |
|---|---|---|---|---|---|
| Scripted assistantNarrow, high-volume questions with predictable answers | Fixed flows, small knowledge base, handoff to a human | ₹8,00,000 – ₹15,00,000 | $10,000 – $18,000 | 6 – 8 weeks | 1 – 2 engineers |
| RAG assistantSupport and internal knowledge assistants over real documentation | Retrieval over your documents, citations, evaluation set, escalation | ₹15,00,000 – ₹28,00,000 | $18,000 – $34,000 | 8 – 12 weeks | 2 engineers |
| Tool-using assistantAssistants that do things — check orders, update records, book slots | Actions against your systems, auth-aware answers, guardrails, analytics | ₹28,00,000 – ₹37,00,000 | $34,000 – $45,000 | 12 – 16 weeks | 2 – 3 engineers |
Figures are indicative ranges for delivery by an Indian agency at senior rates, current as of August 2026. INR and USD are shown at an approximate 83:1 rate and are not a quotation.
Cost Breakdown
Where the budget actually goes
| Phase | Share | What it covers |
|---|---|---|
| Data preparation and retrieval | 25 – 30% | Ingestion, chunking, embedding, and search quality — the part that decides accuracy. |
| Evaluation harness | 15 – 20% | A labelled set and a scoring pipeline, so quality is a number rather than a feeling. |
| Conversation and interface | 20 – 25% | Chat surface, streaming, citations, and the escalation path to a human. |
| Integration and guardrails | 20 – 25% | Tool calls into your systems, permissions, and limits on what it may answer. |
| Deployment and monitoring | 10% | Cost tracking, latency monitoring, and conversation review tooling. |
Cost Drivers
What changes the price, and by how much
If two quotes for an AI chatbot differ wildly, the cause is almost always one of these being assumed in or out.
Knowledge base quality
Clean, structured documentation makes retrieval cheap. Scanned PDFs and tribal knowledge make it expensive.
Tool use and actions
Letting the assistant change state requires permissions, confirmation, and audit.
Multi-language support
Retrieval quality degrades across languages and needs separate evaluation per language.
Voice interface
Speech in and out, plus latency budgets far tighter than text.
Compliance and PII handling
Redaction, retention limits, and audit trails in regulated sectors.
Scope Of A Quote
What is in our price, and what is not
Included
- Discovery, scope definition, and a written delivery plan
- UI design or implementation of your existing design system
- Engineering, code review, and automated tests
- Deployment, environment setup, and CI/CD
- Documentation and a handover your next engineer can actually use
- Post-launch stabilisation for the first two weeks
- A labelled evaluation set and a measured accuracy baseline
Not included
- Third-party service subscriptions billed directly to you
- Cloud hosting and infrastructure costs after launch
- Paid marketing, content production, and app store fees
- Ongoing maintenance beyond the stabilisation window
- Hardware, devices, and physical logistics
- Model API usage charges, which are billed to your account
After Launch
What it costs to keep running
The line most quotes leave out, and the one that decides whether the product is still alive in two years.
| Item | Cost | Notes |
|---|---|---|
| Model API usage | $100 – $2,000 / month | Driven by conversation volume and context size. Retrieval quality is the main lever on this number. |
| Vector store and infrastructure | $50 – $500 / month | Often just pgvector on your existing database, which keeps this low. |
| Ongoing tuning | $800 – $3,000 / month | Reviewing conversations and improving retrieval. Without it, quality drifts down. |
How We Quote
From a brief to a number you can plan against
Step 1
You send a written scope
A feature list, an existing product, or a problem statement. Anything we can read and ask questions about.
Step 2
We ask the questions that move the number
Roles, integrations, compliance, and volume. These change a quote far more than the interface does.
Step 3
You get a banded estimate, not a single figure
A range with the assumptions written next to it, so you can see which assumption to challenge.
Step 4
We recommend the lightest shape that works
Fixed-scope, phased, or a dedicated pod. We say when a smaller engagement would serve you better.
Prefer to work it out yourself first? The project cost estimator gives you a band in a couple of minutes, and the MVP scope planner helps you work out what to cut before you ask anyone for a quote.
Quick Brief
Start the conversation here
Send the scope or the problem. We will come back with a banded estimate and the assumptions behind it.
Your Name
Work Email
What do you need help with?
Proof
Comparable products we have shipped
FAQ
Pricing questions we get asked most
How much does it cost to build an AI chatbot?
$10,000–$45,000 to build, plus $200–$3,000 a month to run. A scripted assistant is $10,000–$18,000; a retrieval assistant over your real documentation with evaluation and escalation is $18,000–$45,000.
What are the ongoing costs of an AI chatbot?
Model API usage of $100–$2,000 a month at typical volumes, infrastructure of $50–$500, and ideally a tuning budget. Token spend scales with conversation volume and how much context you send — better retrieval directly lowers the bill.
How accurate will it be?
That depends almost entirely on your documentation quality, and it is measurable. We build an evaluation set from your real questions first, so you get a number before launch rather than an impression. If the number is not good enough to ship, we will say so.
Can it use our internal data safely?
Yes, with permission-aware retrieval so the assistant only surfaces what the asking user is entitled to see. This is a common failure in quick implementations and worth checking in any quote.
Which AI model should we use?
We build primarily on the Claude and Gemini families, selected per workload on cost, latency, and quality. The integration is designed so switching models is a configuration change rather than a rebuild.
Related Cost Guides
Other budgets worth checking before you commit
Ready to staff it instead?
Next steps
Where buyers usually go from here
The service pages behind this work, plus the tools that turn a budget into a scope.