Cost guide
AI chatbot development cost in India (2026): the full breakdown, with an estimator
By Samuel Marndi · Updated
The short answer
An AI chatbot in India costs ₹30,000–₹75,000 for a scripted FAQ bot, ₹1,00,000–₹4,00,000 for a business bot with WhatsApp and CRM integration, and ₹6,00,000–₹15,00,000 for a RAG system answering from your private documents. On top of that sits a monthly bill of ₹2,000–₹80,000 for model usage, hosting and monitoring — the number most quotes omit and the one that decides whether the thing survives its first year.
The single biggest cost driver is not conversation volume. It is integration count. A bot that answers from your documents is a fraction of the work of one that also writes to your CRM, checks a calendar and raises a ticket — because each of those is an API, an error path, an approval rule and a thing that breaks when the other side changes.
What actually drives the price
Six things move a quote, and they are not the six most people expect. Conversation volume barely features — it changes the monthly bill, not the build. What changes the build is how many places the bot has to reach into, and how much it is allowed to get wrong.
| Driver | Cheap end | Expensive end | Why the gap |
|---|---|---|---|
| Where answers come from | A list of questions you write | Ten thousand pages of documents, with permissions | Retrieval, chunking, permission modelling and an evaluation set are most of the work |
| Integration count | None — it answers and stops | CRM, calendar, ERP, helpdesk | Each one is an API, an error path, an approval rule and something that breaks when the other side changes |
| Channels | Website widget | Website + WhatsApp + Instagram + in-app | WhatsApp needs Business API approval, a verified sender and per-conversation charges from Meta |
| Languages | English | English + two Indian languages | Each language needs its own evaluation set; code-switching mid-sentence is where cheap builds fail |
| Tolerance for being wrong | It can say "I don't know" | It quotes prices or commits stock | Anything that creates an obligation needs guardrails, approval gates and an audit trail |
| Who owns it afterwards | The vendor's platform | Your repository, your accounts | Portability is a build decision. It costs a little up front and a lot to retrofit |
The two numbers: one-time and monthly
Almost every chatbot quote in India is a single figure, and a single figure is always the build. The running cost is real, it is usage-based, and it arrives every month whether or not anyone remembered to budget for it.
Model usage is the volatile part. It scales with conversations and with how much context each conversation carries — a RAG bot that stuffs eight retrieved passages into every prompt costs several times more per conversation than an FAQ bot answering from a paragraph. Hosting and monitoring are steadier and smaller.
| Line | Typical range | What moves it |
|---|---|---|
| Build (one-off) | ₹30,000 – ₹15,00,000 | Integration count, knowledge source, channels, languages |
| LLM API usage | ₹1,500 – ₹60,000 / month | Conversations × context size. The biggest and most variable line |
| Hosting & vector store | ₹500 – ₹8,000 / month | Managed Postgres with pgvector at the low end, a dedicated vector service higher up |
| WhatsApp conversations | Billed by Meta, per conversation | Charged to your Meta account directly, not by the developer |
| Monitoring & logging | ₹0 – ₹4,000 / month | Free tiers cover a small bot; volume needs a real tool |
| Care plan (optional) | ₹8,000 – ₹40,000 / month | Tuning against logged failures, content refreshes, model migrations |
Estimate your own build
The numbers below are the same ones I would use to scope a project. Change the four inputs and the total moves — the arithmetic is shown underneath so you can check it, or redo it on paper if you would rather not trust a form.
Pick the four things that actually drive a chatbot quote. Everything else is detail.
The formula: Build = knowledge source + channels + (integrations × ₹45,000). Monthly = hosting ₹4,000 + model usage at your volume. Year one = build + (monthly × 12).
Indicative Indian market ranges, not a rate card — a real quote comes from a scoping conversation. Model and hosting costs are billed to your own accounts at cost. WhatsApp adds Meta's per-conversation charges on top, billed by Meta.
| Option | Price | Basis |
|---|---|---|
| What it answers from — A fixed set of questions you write | ₹50,000 | one-off |
| What it answers from — Your website and product pages | ₹1,50,000 | one-off |
| What it answers from — Your private documents, with citations (RAG) | ₹5,00,000 | one-off |
| Channels — Website widget only | included | one-off |
| Channels — Website + WhatsApp Business API | ₹60,000 | one-off |
| Channels — Website + WhatsApp + one more | ₹1,10,000 | one-off |
| Systems it writes to — None — it answers and stops | included | one-off |
| Systems it writes to — One (CRM or helpdesk) | ₹45,000 | one-off |
| Systems it writes to — Two (CRM + calendar) | ₹90,000 | one-off |
| Systems it writes to — Three or more | ₹1,35,000 | one-off |
| Conversations per month — Up to 500 | ₹5,500 | per month |
| Conversations per month — 500 – 2,000 | ₹10,000 | per month |
| Conversations per month — 2,000 – 10,000 | ₹32,000 | per month |
| Conversations per month — Over 10,000 | ₹64,000 | per month |
Three real-shaped quotes
Abstract ranges are hard to act on, so here are three builds at the shapes I actually get asked for. Each total is the estimator above with different inputs.
| Scenario | What it does | Build | Monthly | Year one |
|---|---|---|---|---|
| Single-clinic front desk | Website widget answering opening hours, services, preparation instructions and directions. Hands off to a phone number for anything clinical. No integrations | ₹50,000 | ₹5,500 | ₹1,16,000 |
| D2C brand support | Website and WhatsApp, answering from product pages and the returns policy, writing qualified enquiries to the CRM and looking up order status | ₹2,00,000 | ₹10,000 | ₹3,20,000 |
| Manufacturer's internal assistant | RAG over ten years of specifications and service manuals, with citations and role permissions. WhatsApp for field engineers, writing tickets to the helpdesk | ₹6,95,000 | ₹32,000 | ₹10,79,000 |
DIY, freelancer, or agency?
The build price is the part people compare, and it is the smaller part of the decision. What separates these three over a year is not the invoice — it is who absorbs the work when the bot is confidently wrong in week three.
The chart is year-one total cost of ownership for a comparable business bot: build, twelve months of running, and — for DIY — the internal time it actually consumes, valued honestly. That last figure is the one DIY comparisons always omit, and it is usually the largest number in the column.
- DIY is right when the bot answers ten fixed questions, nothing integrates, and someone genuinely has a few hours a week for it. Below that bar it works well and costs very little.
- DIY goes wrong at the first integration and at the first wrong answer in front of a customer. No-code platforms make the easy 80% trivial and the remaining 20% impossible.
- An independent suits one build with real integrations where you want the person writing it to be the person you talk to — and you accept that one person can be ill.
- An agency earns its premium when you need several specialists in parallel, contractual cover, or a procurement process that will not approve an individual.
- Whoever builds it, insist the accounts are yours, the code is in your repository, and there is an evaluation set. Those three make leaving possible, which is the only thing that keeps a retainer honest.
Why RAG costs ten times more than an FAQ bot
An FAQ bot matches a question to an answer you wrote. A RAG system indexes your entire document estate, retrieves the right passages at question time, respects who is allowed to see what, cites its sources, and refuses when it has none.
Most of that cost is not the model. It is data preparation, chunking strategy, permission modelling, and building an evaluation set so accuracy is measured rather than assumed. Preparing ten thousand pages of documentation and support history adds ₹50,000–₹3,00,000 to a project on its own — before a single answer is generated.
Controlling the monthly bill
The running cost is the line that quietly kills chatbot projects in month four. All five of these are decisions made before launch, and all five are cheap then and expensive later.
- Forecast from real volume: conversations per day × average context size, not a guess
- Route routine questions to a cheaper model and hard ones to a stronger one — most traffic does not need the expensive model
- Cache repeated context so the same system prompt and documents are not paid for twice
- Set a hard spending cap at the provider before launch, not after the first surprise
- Log and review monthly — usage patterns shift as staff find new uses for it
Questions worth asking any vendor
The answers separate a quote that will hold from one that will not. None of them are unreasonable, and a vendor who bristles at them has told you something useful.
- What is the monthly running cost at my volume, and what happens if volume triples?
- Show me the accuracy numbers against a real question set — not a demo
- What does it do when it does not know? Show me that path
- Whose accounts hold the model keys and the data, mine or yours?
- If I leave, what do I take with me — and can I have that in writing?
- Who fixes it when it is wrong in production, and how quickly?
FAQ
Related questions
How much does an AI chatbot cost in India?
₹30,000–₹75,000 for a scripted FAQ bot, ₹1,00,000–₹4,00,000 for a business bot with WhatsApp and CRM integration, and ₹6,00,000–₹15,00,000 for RAG on private documents. Add ₹2,000–₹80,000 a month for model usage, hosting and monitoring. The estimator above gets you to a number for your own shape in about thirty seconds.
What is the difference between a chatbot and an AI agent, in cost terms?
Roughly three to five times, and the reason is actions rather than intelligence. A chatbot answers; an agent does something — books, updates, escalates. Every action needs an error path, an approval gate and an audit trail, so an agent starts around ₹2,50,000 where a comparable chatbot is ₹75,000. The chatbot and agent service page breaks both down.
Why do quotes for the same chatbot vary by ten times?
Usually because they are not the same chatbot. Ask each vendor what it answers from, how many systems it writes to, and what it does when it does not know. A ₹40,000 quote and a ₹4,00,000 quote frequently differ on all three — and occasionally the cheap one is right for what you need.
Is there an ongoing cost even if nobody uses it?
Yes, a small one — hosting and the vector store run whether or not anyone asks a question, typically ₹500–₹8,000 a month. Model usage is genuinely usage-based and falls to near zero in a quiet month. Anyone quoting a large fixed monthly fee for a bot with no traffic is charging you for a platform, not for usage.
Can I start small and add to it later?
Yes, and it is usually the right sequence — provided the first build is structured for it. Starting with a website FAQ bot and adding WhatsApp and a CRM integration later costs perhaps 20% more in total than building everything at once, and it removes the risk of spending ₹4,00,000 on something nobody uses. What it must not be is a throwaway: agree at the start that the knowledge base and evaluation set carry forward.
Does WhatsApp cost extra?
Yes, in two ways. Building the WhatsApp channel adds roughly ₹60,000 to the project, and Meta charges per conversation on the Business API, billed to your own Meta account rather than through the developer. For most Indian businesses it is still worth it — customers are already there, and open rates are not comparable to a website widget.
Will it invent answers?
A properly built one will not: answers are grounded in retrieved content, cite their source, and refuse plus escalate when no source exists. That is a build decision rather than a property of the model. Ask any vendor to show you the refusal path working before you buy — it is the single most revealing demo you can ask for.
How long before it pays for itself?
For an FAQ bot deflecting genuine repeat questions, usually within a few months — it is the cheapest tier and the volume is immediate. For RAG on internal documents the payback is real but slower and harder to measure, because the saving is senior staff not being interrupted. If a vendor gives you a confident ROI figure for a RAG build, ask how they measured it.
Is a chatbot worth it for a small business?
If your team answers the same twenty questions daily, usually yes, and an FAQ bot at ₹30,000–₹75,000 pays back quickly. If enquiries are varied and consultative, a bot often frustrates customers, and document extraction or lead qualification is the better first AI project. I will tell you which of those you are before quoting.
What about a voice agent instead?
Different economics entirely — voice is billed per minute of talk time rather than per conversation, so the arithmetic works out differently and volume matters far more. The voice agent page has the cost-per-call comparison against a receptionist, including the volumes at which a person is still cheaper.