AI chatbot for business: what it does and what it costs.

An AI chatbot for business answers questions from your own documents, on your website or on WhatsApp, and hands anything else to a person. Two ready-made tools we checked start at about ₹2,700 and ₹3,100 a month. Two Indian firms publish ₹1.5 lakh to ₹6 lakh to build a custom one, plus a monthly running cost. We quote custom builds after a call.

What does an AI chatbot do for a business?

• Answers repeat questions about prices, timings, policies and delivery, using your own documents.
• Collects the details of a new enquiry, then passes them to sales.
• Replies when your team is off for the day.
• Helps staff find a rule or a process in internal documents.
• Hands the conversation to a person, with the history attached, when it should stop.

The method behind answering from your own documents is called RAG. We explain it in what is RAG. A chatbot is also different from an agent: a chatbot talks, while an agent does a job in your other software.

One workflow on our work page shows the idea on WhatsApp. It reads each message, text or voice note, pulls the details into a lead, asks only for what is missing and pauses when a person takes over.

Should you use a rule-based bot or an AI chatbot?

A rule-based bot follows a script of buttons and fixed replies. It is cheap and predictable, and often enough for booking an appointment or capturing a lead. An AI chatbot understands questions in the customer's own words, which helps when people ask the same thing many different ways. It costs more to build and run, and it needs testing, because it can get an answer wrong.

What does an AI chatbot cost?

What does it cost to run each month?

For a custom AI chatbot, Codingclave publishes ₹8,000 to ₹40,000 a month. Navtechy splits it up: ₹2,000 to ₹15,000 for model use on a low-traffic support bot, ₹1,500 to ₹20,000 for hosting and search, and ₹10,000 to ₹50,000 if you want someone watching quality and updating the content.

You can estimate the model part yourself. Providers charge per million tokens, which are small pieces of text. Here is an illustration with our own round numbers, not a measurement: 1,000 conversations a month, five replies each, with 3,000 tokens going in and 300 coming out per reply.

The gap between models is wide, and the cheapest is not always good enough. Test each one on your own real questions. Notice too that model use can be the smaller part of the monthly bill. Hosting and someone keeping the content right may cost more.

On WhatsApp, Meta gives each business number 1,000 free service messages a month, and has charged from the 1,001st since 1 October 2026. Rates change by country and category, so read our guide to WhatsApp pricing and check Meta's rate card.

Should you buy a chatbot or build one?

Buy one if your questions are simple, you use one channel and your volume is low. You can be live quickly, and you can stop paying if it does not help.

Consider building when the bot has to read live data, such as orders or stock, or when you need control over its answers and your data. Volume also matters: at $0.99 an outcome, 1,000 outcomes a month is $990, about ₹94,000. That is well above the monthly running costs published for a custom build, though you would also have to pay for the build itself.

Can you build a chatbot yourself with AI tools?

Yes, and a working one can take a day. Several tools let you upload your documents, connect an AI model and paste a chat window onto your website. That is a cheap way to learn what customers ask, and whether a chatbot helps at all.

Putting it in front of customers safely is a different job. Answer these before it goes live:

• Where is the API key kept? If it sits in your website's code, anyone can copy it and run up your bill.
• Is there a spending limit with the model provider, so a flood of messages cannot run up a large bill overnight?
• Which documents can it read? If it can see internal files, a clever question may get it to repeat them to a customer.
• Can a customer talk it into ignoring its rules, offering a discount or promising a refund?
• Are conversations saved, so you can see what it told people?
• Who is told when it breaks, and how quickly can they fix it?

AI tools can put the chatbot live for you too. If nobody understands how it was set up, you find these problems after a customer does. Paying someone who understands it to check the setup usually costs less than the mistake.

What can go wrong?

The main risk is a confident wrong answer. In 2024, a tribunal in British Columbia, Canada, ordered Air Canada to pay a customer CA$812.02 after its website chatbot gave wrong information about bereavement fares. The tribunal said Air Canada was in effect suggesting that the chatbot was responsible for its own actions, and replied: "This is a remarkable submission." It said Air Canada was responsible for all the information on its website, whether it came from a static page or a chatbot.

That is a small claim in another country, not Indian law. The lesson still travels: customers will treat your chatbot as you.

• Let it answer only from documents you have approved, and show where each answer came from.
• Tell it to say when it does not know, and pass the customer to a person.
• Keep a person in charge of refunds, discounts and anything else that binds you.
• Test it on real customer questions before launch, and again after every change.
• Read the conversations every week, and fix the content behind any bad answer.

Customer data needs care too. India's Digital Personal Data Protection Rules, notified in November 2025, phase in over 18 months, and the rules on notices and most other duties take effect around May 2027. They expect a notice in clear and plain language, and they require you to tell affected people about a data breach. The government's summary adds that people may withdraw consent at any time. If your chatbot collects names and numbers, ask your lawyer what this means for you.

How do you start?

Start with one job, not a platform. We make the same case in AI agents: start with one job.

• Write down the 30 questions customers ask most.
• Check that a current document answers each one, and fix the ones that do not.
• Choose one channel: your website or WhatsApp.
• Decide what must go to a person, and how quickly they will reply.
• Try it on real questions before customers see it.
• After a month, count the questions it answered correctly and the ones it passed on.

If a ready-made tool covers that list, use it. If your bot needs your own systems, see custom AI and agents. Every build we ship has a set of test questions and monitoring. We estimate model use before you start, and we can talk it through.

About the author

Abhinandan Jain, co-founder, clients and AI

Abhinandan looks after clients, sales and the AI side of our work, including AI systems for clients in Dubai. He has built AI systems since 2025 and is usually the first person you speak to.

Meet the team LinkedIn