In 2026, a chatbot in India costs roughly 40,000 to 1.5 lakh for a rule-based menu bot, 3 to 8 lakh to build a custom AI chatbot that answers from your own content, or a platform subscription of 2,500 to 30,000 rupees a month plus a fee for every conversation it resolves. That last fee is the one that decides most budgets, and almost nobody models it before signing: the major platforms charge between 8 and 170 rupees per resolved conversation, on top of the subscription.
This guide is mostly about buying, because most people searching for a chatbot price should buy rather than build. It ends with the arithmetic for the point where that stops being true. If what you actually need is software that takes actions in your systems rather than answering questions about them, that is an agent, it costs more, and the ranges are in our AI agent development cost guide. The difference between the two, in plain terms, is in chatbot vs AI agent.
The quick-reference table
| What you are buying | What you pay | Live in | Fits when |
|---|---|---|---|
| Rule-based menu bot | 40,000 to 1.5 lakh once, or a few thousand a month on a builder | 1 to 3 weeks | Five to twenty known questions, a booking link, and a hand-off to a person |
| Platform chatbot, configured | 2,500 to 30,000 rupees a month, plus 8 to 170 rupees per resolved conversation | Days | Standard support questions, a help centre to feed it, and nobody to maintain code |
| Custom AI chatbot over your own content | 3 to 8 lakh once, then 15,000 to 40,000 a month to run | 3 to 5 weeks | Your content is not in a help centre, the answers must be exact, or the volume makes per-resolution pricing painful |
| Chatbot that acts in your systems | 8 to 25 lakh | 4 to 8 weeks | It books, refunds, cancels or updates records. This is an agent, and it is priced as one |
The Indian agency market quotes the first and third rows at 35,000 to 1.25 lakh for a rule-based FAQ bot and 1.5 to 3.5 lakh for a typical business website chatbot with natural language, rising to 5 lakh and beyond for enterprise work. Those numbers are real. Our third row sits a little higher than the market's because it includes the evaluation set and the hand-off, which is the part cheap quotes leave out, and which is the difference between a bot that demos well and one that is still answering correctly in month three.
What the platforms actually charge
Here is the part that surprises people. Platform pricing has two meters: a seat or plan fee, and a fee per conversation the bot resolves without a human. The second one is usually larger.
| Platform | Plan fee | Per resolved conversation | In rupees, roughly |
|---|---|---|---|
| Intercom with Fin | $29 to $132 per seat a month | $0.99, with none included | About 83 rupees |
| Zendesk with AI agents | $55 to $169 per agent a month, plus $50 for the advanced AI add-on | $1.50 committed, $2.00 pay as you go | 126 to 168 rupees |
| Tidio with Lyro | $29 and up, Lyro add-on $39 a month with 50 conversations | $0.50 to $1.00 after that | 42 to 84 rupees |
| Freshchat with Freddy | Free for two agents, then $15 to $79 an agent a month | $100 to $1,000 per 1,000 sessions, no rollover | 8 to 84 rupees |
Converted at 84 rupees to the dollar. Two things follow.
First, the plan fee is not the price. A three-seat Intercom plan at the middle tier is about 21,000 rupees a month before a single conversation is resolved. Two thousand resolved conversations on top of that is another 1.66 lakh.
Second, the definition of "resolved" is the contract. Intercom bills a resolution when the customer confirms the answer helped or leaves without escalating. Zendesk bills an automated resolution on its own definition. Freshchat bills sessions rather than resolutions, so a customer who asks three questions in one visit may be one session or three depending on the timeout. Ask for the definition in writing, then ask what your current ticket volume would have cost last month under it.
The crossover: when buying stops being cheaper
This is the arithmetic the platform pages do not show you, and it is simple enough to do on the back of an envelope.
Take Intercom's 83 rupees a resolution, and a custom AI chatbot built over your own content at 5 lakh, running at 25,000 rupees a month for hosting, model usage and a few days of attention.
| Resolved conversations a month | Platform, year one | Custom build, year one | Custom build, year two |
|---|---|---|---|
| 250 | About 2.5 lakh | 8 lakh | 3 lakh |
| 500 | About 5 lakh | 8 lakh | 3 lakh |
| 1,000 | About 10 lakh | 8 lakh | 3 lakh |
| 2,000 | About 20 lakh | 8 lakh | 3 lakh |
| 5,000 | About 50 lakh | 8 lakh | 3 lakh |
The platform column includes a modest plan fee alongside the per-resolution charge. The build column is 5 lakh plus twelve months of running cost in year one, and running cost alone after that.
The crossover is somewhere between 600 and 1,000 resolved conversations a month. Below it, buy: the platform is cheaper, faster and someone else maintains it. Above it, the per-resolution meter starts funding a build every year, and at a few thousand conversations a month it funds one every quarter.
Two honest caveats. The build column assumes the bot keeps working, which takes the few days a month that are already in the 25,000. And a custom build has no one to call at 2am, which is worth something real that does not appear in either column.
Where the build money goes: a worked estimate
Tier tables hide the arithmetic, so here is a concrete build: a chatbot on your website and in your app that answers from your help articles, policies and product data, recognises when it does not know, and hands over to a person with the conversation attached. It reads; it changes nothing.
| Work package | Person-days |
|---|---|
| Discovery: the twenty questions that make up most of the volume, and what a good answer looks like | 3 |
| Collecting and labelling 150 real conversations, and building the scorer | 5 |
| Content preparation: getting the help articles, policies and product data into a form the bot can search | 6 |
| Retrieval and prompts, scored against the set | 7 |
| The widget: website and app, mobile layouts, accessibility, your brand | 4 |
| Hand-off to a person, with the transcript and the customer's history attached | 3 |
| Guardrails: refusing confidently, prompt-injection defences, personal data handling | 3 |
| Observability: every conversation logged, a weekly review screen, cost tracking | 3 |
| Testing and a staged rollout, after-hours first | 3 |
| Total | 37 |
At a blended 16,000 rupees a person-day, that is about 5.9 lakh, inside the third row of the table above, and about four weeks with two people. Notice that content preparation at 6 days is nearly as large as the retrieval and prompt work at 7. That is the real lesson of chatbot projects: the bot is only as good as the material you point it at, and most companies discover during the build that their help centre has been wrong since 2024.
Running it costs 15,000 to 40,000 rupees a month: model usage, hosting, the search index, and the few days of attention that keep the answers right. At a few thousand conversations a month the model usage itself is small, around two rupees a conversation, which is the same arithmetic as in what an AI agent costs to run.
The seven things that move the price
- Rule-based or AI. A menu of buttons is a week of work and never surprises anyone. A bot that understands free text needs a test set, a confidence threshold and a hand-off, and that is the jump from 1 lakh to 5.
- The state of your content. A current, well-structured help centre is the cheapest input there is. Scattered PDFs, a wiki nobody has edited since 2024, and knowledge that lives in three people's heads are the most expensive, and the clean-up lands in your budget whether or not the quote mentions it.
- Languages. English alone is the base. Hindi and regional languages each need their own examples and their own testing, and the answers must be checked by someone who speaks them.
- Channels. A website widget is one place. The same bot on WhatsApp carries Meta's fees and template rules, which are their own budget line and are set out in our WhatsApp AI agent cost guide. On the phone it becomes a voice agent, priced per minute, which is in our AI voice agent cost guide.
- Whether it may act. The moment the bot books, cancels, refunds or edits a record, it needs limits, an audit log and approval gates, and it is priced as an agent rather than a chatbot.
- Where the answers come from. A bot answering from documents is retrieval work. A bot answering from live data, stock, order status, account balance, needs connectors to the systems that hold it, and each one is a week or more.
- Who reviews it after launch. A chatbot with nobody reading its transcripts degrades quietly. Budget a few hours a week or accept that it will be wrong in ways you find out about from customers.
When a rule-based bot is the right answer
Agencies rarely recommend the cheap option, so here it is plainly. A menu bot is the right purchase when your questions are few and stable, when the answer is usually a link or a form, and when the real job is routing rather than understanding. Store hours, delivery areas, a booking link, a complaint form, and a clear path to a human: that is a one lakh project that will work for years and never hallucinate anything, because it cannot.
The failure mode of the cheap bot is not that it is cheap. It is a menu with no way out, which trains customers to type "agent" before they read anything. If you buy this row, buy the escape hatch with it.
The costs that are not in the quote
| Cost | Typical size | When it surprises you |
|---|---|---|
| Fixing the content the bot answers from | 5 to 15 days | When the first test answers confidently from an article nobody has updated in two years |
| The per-resolution meter | 8 to 170 rupees each | On the second invoice, when volume rose |
| A plan upgrade for the API or the integration you need | Often a tier above where you are | When the handover to your helpdesk turns out to be an enterprise feature |
| Translation and review | Per language, ongoing | When Hindi answers are fluent and subtly wrong |
| Maintenance | 15 to 25 percent of a custom build a year | The first time a model changes and the tone shifts |
| The people it hands off to | Training, and a harder case mix | Month two, when the easy questions are gone |
| Transcript storage and consent | Small, but a policy decision | When someone asks how long you keep customer conversations |
GST at 18 percent applies to the development invoice and to most platform fees, claimable as input credit if you are registered.
How to read a chatbot quote
Six checks, ten minutes.
- Ask which row of the table it is. A quote of 60,000 rupees for an "AI chatbot" is usually a menu bot with a language model writing the greeting.
- Ask for the per-resolution definition and a worked example using your own volume from last month.
- Ask how it knows it is right. A test set of real conversations with scored answers, or nothing.
- Ask what happens when it does not know. The answer should be a hand-off with the transcript attached, not a confident guess.
- Ask who owns the prompts, the content index and the transcripts. All three should be yours.
- Ask what it costs to run, as a separate number. A build price with no monthly figure beside it is half a quote.
The longer version, for choosing the company rather than the number, is in our guide to hiring an AI development company.
How to spend less without a worse chatbot
- Fix the help centre first. It is the cheapest work in the project and it improves every answer the bot gives. It also helps the customers who never open the chat.
- Start with the twenty questions that make up most of your volume. Deflecting those well beats covering everything badly.
- Buy before you build. Two months on a platform tells you your real resolution volume, which is the number the build decision turns on.
- Keep it read-only at first. Answering is cheap and safe. Acting is where the cost and the risk both live.
- Put the escape hatch one click away. It costs nothing and it is the difference between a bot customers tolerate and one they complain about.
- Review transcripts weekly for the first month. An hour a week in month one saves a rebuild in month six.
When not to buy a chatbot at all
- Under a few hundred conversations a month. A good contact page, a short FAQ and a WhatsApp link will serve customers better and cost nothing to run.
- When the questions need judgement. Pricing negotiations, complaints, anything with money or emotion in it. Route those to a person immediately and let the bot handle the rest.
- When nobody will own it. A chatbot is a product, not an installation. If no one is accountable for its answers, it will be wrong and nobody will know.
Questions people ask
How much does it cost to make a chatbot in India?
A rule-based menu bot costs 40,000 to 1.5 lakh. A custom AI chatbot that answers from your own content costs 3 to 8 lakh, and the worked estimate above lands at about 5.9 lakh for 37 person-days. A configured platform chatbot costs no build fee, a subscription of 2,500 to 30,000 rupees a month, and a per-conversation charge on top. Which is cheapest depends almost entirely on your monthly volume.
What do chatbot platforms charge per conversation?
Between 8 and 170 rupees per resolved conversation in 2026. Intercom's Fin is $0.99, about 83 rupees, with none included in the plan. Zendesk charges $1.50 to $2.00 per automated resolution. Tidio's Lyro is $0.50 to $1.00 after the first 50 a month. Freshchat prices sessions at $100 to $1,000 per thousand. The definition of a resolution differs by vendor and is worth reading before signing.
At what point is building cheaper than buying?
Somewhere between 600 and 1,000 resolved conversations a month, on the arithmetic above. At 1,000 a month a platform costs roughly 10 lakh a year, against about 8 lakh for a 5 lakh build plus its running costs in year one and 3 lakh a year after that. Below 500 a month, buying wins comfortably and you should buy.
How much does a chatbot cost to run each month?
Fifteen to forty thousand rupees for a custom build at modest volume: model usage, hosting, the search index and a few days of attention. Model usage alone is around two rupees a conversation. On a platform, the running cost is the subscription plus the per-resolution meter, which at a thousand conversations a month is 83,000 rupees and up.
Is a rule-based chatbot still worth building in 2026?
Yes, for a narrow job. If your questions are few, stable, and usually answered with a link or a form, a menu bot is cheaper, faster, impossible to confuse and will not invent anything. Buy the escape hatch to a human with it, because the common failure is a menu with no way out rather than the technology itself.
How long does a chatbot take to build?
One to three weeks for a rule-based bot, days to configure a platform, and three to five weeks for a custom AI chatbot over your own content. The step that stretches it is almost never the engineering. It is getting the content into a state the bot can answer from.
Can a chatbot book appointments or process refunds?
It can, and at that point it is an agent rather than a chatbot, with action limits, an audit log and approval gates to match. That is a different project, at 8 to 25 lakh, and the ranges and a worked estimate are in our AI agent development cost guide.
What about a chatbot on WhatsApp instead of the website?
Same bot, different channel, and a bill of its own: Meta charges per message, and since 1 October 2026 it charges for the replies your bot sends inside the 24 hour window too. The arithmetic is in our WhatsApp AI agent cost guide.
Getting a real number
Ranges end where your own conversations begin. We price chatbot work after a short discovery and, where it is useful, a two-week prototype on your real questions, so the number you approve is based on how well it answers them rather than how it demos. What that looks like is on our AI agents and automation page.