AI Automation
AI Chatbot for Business in India: What It Can Do, What It Costs to Run, and How to Start Safely
Most businesses that ask about an AI chatbot for business use have the same problem: the same questions arrive again and again, often outside working hours, and every missed reply is a customer who may go elsewhere. A well-built AI chatbot can answer those questions instantly, in English, Hindi or Hinglish. A badly built one can confidently tell a customer the wrong price.
The short answer: an AI chatbot is worth it when customers ask many different questions that your own information can answer, such as prices, timings, services, policies or order status. Build it to answer only from your content, to say when it does not know, and to hand over to a person easily. Expect two kinds of cost: a one-time build, and running costs for AI usage, hosting and, on WhatsApp, Meta's per-message charges. Start with one channel and one job, test on real customer messages, and measure before expanding. If you only get a handful of repeated questions, quick replies in the free WhatsApp Business app may be all you need.
What an AI chatbot can actually do for a business
- Answer common questions about services, prices, timings, location, eligibility and policies, at any hour.
- Understand messy messages. Current AI models handle typos, Hinglish and mixed languages far better than older menu-based bots.
- Capture leads. Collect a name, number and requirement, then pass them to your team with the conversation attached.
- Take actions, if connected to your systems. Check a booking slot, create an appointment or look up an order status.
- Hand over to a person when the question is unusual, sensitive or the customer asks for one.
Chatbot or AI agent?
A chatbot answers questions. An AI agent can also do things, such as booking a slot or checking an order, by calling your booking or order system. Agents save more time but need clearer limits: what they may do, what needs approval, and what they must never do. Start with answering; add actions once the answers are reliable.
Website chat or WhatsApp?
| Website chat | ||
|---|---|---|
| Where customers are | Visitors already on your site | Where most Indian customers already message businesses |
| Setup | Added to your website | Needs the WhatsApp Business Platform (API), not the free app |
| Running cost | AI usage and hosting | AI usage, hosting and Meta's message charges |
| Follow-up | Ends when the visitor leaves, unless you collect a number | The conversation stays in the customer's WhatsApp |
Many businesses start with whichever channel gets more enquiries. Both can use the same knowledge and logic. For the WhatsApp side, our WhatsApp automation guide explains the difference between the free app and the Platform, and what can be automated without AI at all.
How to stop the chatbot from making things up
AI models can produce fluent, confident answers that are wrong. For a business, that is the main risk. These practices reduce it a lot:
- Answer only from your own content. The chatbot is given your website pages, price list, FAQs and policies, and instructed to answer from those, not from general knowledge.
- Allow "I don't know". When the answer is not in your content, it should say so and offer a person, not guess.
- Keep prices and availability live. Read changing facts from your system rather than from a copy that goes stale.
- Test with real messages. Before launch, run it against actual questions your customers have sent, including the awkward ones.
- Log and review. Keep a record of conversations and check them regularly, especially in the first weeks.
- Limit what agents can do. Anything costly, irreversible or sensitive should need a person's approval.
Chatbots can also be manipulated by cleverly worded messages, a risk known as prompt injection. The OWASP Top 10 for LLM applications lists this and other risks; a careful build limits what the chatbot can see and do, so a manipulated reply cannot leak data or take harmful actions.
What it costs to build and run
There are two separate parts:
- Build cost (one-time): setting up the knowledge, the conversation design, the channels, any connections to your systems, testing and launch. It depends mostly on how many channels and actions you need, so ask for a written quote that lists them.
- Running cost (ongoing): AI providers charge by usage, so cost grows with the number and length of conversations. Hosting is usually modest. On WhatsApp, Meta charges for business messages according to its published WhatsApp Business Platform pricing, which depends on the message category.
A good quote estimates running costs from your actual volumes, such as how many enquiries you get a day, so there are no surprises. Be wary of any offer that hides usage costs or locks your data into a platform you cannot leave.
Customer data and privacy
A chatbot handles personal data: names, phone numbers and sometimes health or financial details. India's Digital Personal Data Protection Act sets rules for how businesses collect and use personal data, and the government notified its rules in 2025 (see PIB's announcement and MeitY's data protection page). In practice:
- Collect only what you need, and tell customers why.
- Know which services see which data, including the AI provider.
- Do not send sensitive details to an AI model unless the task truly needs them.
- Keep access to conversation logs limited to people who need it.
For clinics, legal practices and finance, keep a person in the loop for anything sensitive and take advice on your sector's specific rules.
How to start: a simple plan
- Collect your top questions. Scroll through a month of WhatsApp chats and calls and list the 20 to 30 questions that come up most.
- Write the answers down. Prices, timings, policies and edge cases. This content is what the chatbot will use, and it improves your website too.
- Pick one channel and one job. For example, answering service and price questions on WhatsApp, with handover for bookings.
- Test before launch on real messages, with your team trying to break it.
- Launch, watch and measure. Track how many questions it answers correctly, how many it hands over and whether response time improves.
- Then expand, for example by adding bookings, order status or a second channel.
The same approach works for back-office automation such as data entry, reminders and reports; see our guide to automating repetitive business tasks with AI.
When you should not build an AI chatbot
- You get only a few repeated questions a day. Quick replies and an away message in the free WhatsApp Business app may be enough.
- Your information changes constantly and is not written down anywhere yet. Fix that first.
- Every enquiry needs a personal judgement call. A chatbot can still collect details, but should not answer.
How we build AI chatbots at Stackquil
We start with a consultation about the questions and tasks that take your team's time, then quote one chatbot or workflow in writing, with running costs estimated from your volumes. We ground answers in your own content, test on your real messages, build handover to your team and log every conversation. Read more on our AI chatbot development and AI automation pages, or in Hinglish: AI chatbot banwana hai?
