AI Chatbot for Business: Uses, Limits and Setup

An AI chatbot can answer your routine questions, collect enquiry details and take booking requests at any hour, on your website or on WhatsApp. It can't think for your business, and it will sometimes state wrong information with complete confidence. The small businesses that get value from chatbots keep them narrow: answer from the business's own content, capture the lead, and hand anything unusual to a person. This guide covers what a chatbot can and can't do, where to put it, how to keep it accurate and within the rules, and how to start without a big project.

What we mean by an "AI chatbot"

There are two broad kinds of chatbot, and many businesses only need the first.

Button-based (rules) bots follow a fixed script: "Tap 1 for timings, 2 to book, 3 to talk to us." They are predictable and easy to check, but they can't understand a question typed in the customer's own words.
AI chatbots use a large language model to understand free-text questions, including mixed English, Bengali and Hindi, and write a reply. They feel more natural, and they can also make things up.

Most good setups combine the two: buttons for the common paths, AI for open questions, and a person for everything else. If all you need are automatic replies to forms and bookings, read our WhatsApp automation guide first. You may not need AI at all.

What a chatbot can do well

Use case	Example	What it needs
FAQs	"Are you open on Sunday?" "Do you deliver to Behala?"	An up-to-date FAQ, timings, service areas and policies
Enquiry qualification	Asking budget, location and timeline before a salesperson calls	Three to five agreed questions, and a CRM or sheet to save answers
Booking	Offering free slots for a consultation, class or site visit	A connection to your calendar or booking system
Order status	"Where is my order 1042?"	Read access to your order or courier data

The pattern is the same in every row: the bot is only as good as the information and systems you connect it to. A chatbot with no access to your calendar can't book anything. It can only promise that someone will call.

What it can't (or shouldn't) do

Make decisions with consequences: refunds, credit, discounts, exceptions to policy.
Quote anything it can't check live: stock, availability, custom prices. If the data isn't connected, the bot should say a person will confirm.
Handle complaints. An upset customer wants a person, and a bot's polite non-answer usually makes it worse.
Give professional advice: medical, legal, financial or tax. Keep it to factual information about your business.
Negotiate. In our view, any conversation about money beyond your published prices belongs with your team.

Where the chatbot lives: website or WhatsApp

On your website

A chat widget catches visitors while they're already reading about you. It works well for FAQs and for turning a browsing visitor into an enquiry. The weakness: once the visitor closes the tab, the conversation is usually over unless you've collected a phone number or email.

On WhatsApp

If your customers already message you on WhatsApp, a bot there meets them where they are, and the conversation stays on their phone for later. It comes with rules:

Opt-in. WhatsApp's Business Messaging Policy says you may contact people only if they have given you their number and opted in to receive messages from you.
The 24-hour window. When a customer messages you, a 24-hour customer service window opens. Meta's pricing documentation says non-template messages, the normal replies a bot sends, can only be sent inside that window and are free. Outside it, you can only send pre-approved templates.
An escalation path is required. The policy allows automation during the 24-hour window but says you must also have "prompt, clear, and direct escalation paths", such as transfer to a human agent, a phone number, email or a support form.
The bot must serve your business. Meta's Terms for WhatsApp Business Platform (updated 23 September 2026) prohibit AI providers from using the platform where AI is the primary, rather than incidental or ancillary, functionality. A bot that answers questions about your own shop or clinic is a different thing from a general-purpose AI assistant, but check the current terms with your provider.

Human handover: decide the triggers first

A chatbot without a way out frustrates people and, on WhatsApp, breaks the policy. Before you build anything, write down when the bot must hand over. A practical starting list:

the customer types "agent", "human", "call me" or similar, in any language you serve;
the bot can't find the answer in your content;
the question involves money, a complaint or a cancellation;
the customer repeats the same question, which usually means the bot misunderstood;
a qualified lead is ready for a call or visit.

Then decide who receives the handover, how they're alerted and what happens outside office hours. "Our team will reply after 10 am" is a perfectly good answer at midnight.

Accuracy: the hallucination problem

The US National Institute of Standards and Technology (NIST), in its Generative AI Profile (NIST AI 600-1), describes "confabulation" as "the production of confidently stated but erroneous or false content (known colloquially as 'hallucinations' or 'fabrications')". It explains that this is a natural result of how generative models work: they predict likely text, which is often correct but not always. NIST also notes that people tend to believe false content because of how confident the response sounds.

For a business, that means a bot inventing an opening time, a price or a policy, and a customer acting on it. You can't remove the risk completely, but you can shrink it:

Ground the bot on your own content. Give it an approved knowledge base: FAQ, price list, service pages, policies. Instruct it to answer only from that material.
Make "I don't know" an acceptable answer. The instruction should be: if the answer isn't in the content, say so and offer a person.
Keep the content current. In our view, outdated source material, such as last year's prices or old holiday hours, is one of the easiest ways for a grounded bot to go wrong. Assign someone to update it.
Test before launch with real questions from your inbox and call log, including awkward ones.
Review conversations every week at first. Note every wrong or unhelpful answer and fix the content or the instructions.
Tell customers it's automated. It sets the right expectations and makes the handover option obvious.

Data protection and consent basics

This is general information, not legal advice. Check your obligations with a legal adviser.

A chatbot collects personal data: names, phone numbers, sometimes addresses and details of what the customer needs. In India, the Digital Personal Data Protection Act, 2023 (No. 22 of 2023) governs this. A few points from the Act that matter for chatbots:

Consent must be "free, specific, informed, unconditional and unambiguous with a clear affirmative action", and limited to the data necessary for the stated purpose (section 6(1)).
A notice describing the data and the purpose should accompany or precede the request (section 5), and people must be able to read it in English or any language in the Eighth Schedule to the Constitution, which includes Bengali.
Withdrawing consent must be as easy as giving it (section 6(4)).
Reasonable security safeguards are required to prevent a personal data breach, including for processing done on your behalf by a vendor (section 8(5)).

The government notified the DPDP Rules, 2025 on 14 November 2025, and the Press Information Bureau says they introduce an eighteen-month period for phased compliance. Even before every obligation applies, these are sensible habits:

ask only for what you need to answer or book;
show a short notice and a link to your privacy policy before the bot collects details;
never ask for full card numbers, bank account numbers or ID numbers in chat (WhatsApp's policy prohibits this too);
check where your chatbot vendor stores conversations, who can read them, and whether they're used to train AI models. Meta's WhatsApp terms say platform data must not be used to train AI systems, other than fine-tuning a model for your exclusive use.

Clinics: factual information only

A clinic chatbot can usefully answer timings, location and directions, which doctor is available on which day, declared consultation fees and how to book. It must not suggest diagnoses, assess symptoms or recommend treatment. Route anything clinical to staff, and anything urgent to a phone number.

The NMC's October 2026 advertising guidelines list messaging platforms such as WhatsApp as a medium of advertising, prohibit AI-generated promotional campaigns for commercial interest, require disclosure where AI-generated content's artificial nature matters to the audience, and say any use of patient information as input for an AI system must comply with privacy, confidentiality and data-protection requirements (clause 7.2). So keep the bot informational, label it as automated, and don't let it collect medical history it doesn't need. See our NMC checklist for clinics and our guide to WhatsApp automation for clinics.

Real estate: qualify, then hand over fast

Property enquiries suit a chatbot well because the first questions are always the same: budget, preferred area (say, New Town or EM Bypass), configuration and when the buyer plans to move. The bot can ask those, share approved project facts from your brochure, offer site-visit slots and pass the lead to the right agent. It should not quote prices or unit availability unless it reads them from a live inventory. We cover the full flow in WhatsApp and CRM for real estate leads.

How to start small

Pick one job. Usually FAQs or enquiry qualification, not both at once.
Collect the questions. Go through the last month of WhatsApp chats, emails and call notes and list what people actually ask.
Write the answers in plain language. This becomes the bot's knowledge base.
Set handover rules and name the person who receives them.
Add consent and a notice before any details are collected.
Launch on one channel, website or WhatsApp, and review every conversation for the first couple of weeks.
Expand only when it's working: add booking, then order status, then the second channel.

How to measure it

Decide what success looks like before launch, and measure against your own starting point rather than someone else's benchmark.

Questions answered without handover, and how many of those answers were correct when reviewed.
Handovers, and how quickly a person picked them up.
Unanswered questions: each one is a gap in your content.
Leads captured and bookings made through the bot, recorded in your CRM with the source.
Complaints, blocks and opt-outs, especially on WhatsApp.
Staff time spent on repetitive questions, before and after.

If the review shows the bot is often wrong, cut its scope back. A small bot that is always right is worth more than a clever one customers can't trust.

Thinking about a chatbot?

We build chatbots grounded on your own content, with clear handover to your team and connections to your CRM or booking system. See our AI automation services or tell us which questions you'd like to automate.

Based on WhatsApp's Business Messaging Policy, Meta's WhatsApp pricing documentation and Terms for WhatsApp Business Platform, NIST AI 600-1, the Digital Personal Data Protection Act, 2023, the PIB note on the DPDP Rules, 2025 and the NMC public notice of 6 October 2026, as read on 18 October 2026.

Sources

Written by Koushik Paikar

Founder of Pocket Web Dev, a digital marketing and web development agency in Kolkata. About us

AI Chatbots for Small Businesses: What They Can Do, What They Can't, and How to Start