Should your business get an AI chatbot?

By Sahil Chauhan · Founder & CEO, Verelios Labs · Published 26 June 2026

Every business owner has been pitched an AI chatbot in the last year. Some need one; many don't. The difference comes down to whether you're answering the same questions over and over, and losing customers when you're slow to reply. Here's an honest look at what a modern chatbot does, what it costs, and how to decide. A chatbot is just one slice of the bigger picture — for everything else worth automating, see our broader guide to AI automation for Indian businesses.

What a good AI chatbot actually does in 2026

Forget the old "press 1 for sales" menus. A modern chatbot is trained on your own content — your pricing, policies, product catalogue, FAQs — and answers in plain language. A good one will:

• Answer routine questions instantly, in English or Hindi, day or night.
• Pull from your real information instead of making things up, because it's grounded in your documents (a technique called retrieval, or RAG).
• Capture the visitor's details and log the conversation as a lead.
• Know its limits — and hand off to a human cleanly when the question is genuinely complex or high-value.

The point isn't to replace your team. It's to stop them answering "what are your timings?" for the fortieth time today, so they can handle the conversations that actually need a person.

Website widget or WhatsApp?

In India, WhatsApp is often the better channel — it's where customers already are, and a WhatsApp chatbot can qualify and reply to leads the moment they message. A website widget makes sense when most enquiries start on your site and you want instant answers before a visitor bounces. Many businesses run both, sharing the same underlying knowledge so answers stay consistent.

What it costs to build and run

A focused chatbot trained on your content typically costs ₹49,999 to ₹1,50,000 to build, depending on how much material it needs to learn, how many channels it lives on, and how deeply it integrates with your CRM or systems. A simple FAQ bot sits at the lower end; one that books appointments, checks order status, or talks to your database sits higher.

Then there's the running cost. A hosted chatbot pays per conversation for the model it uses — usually a few hundred to a few thousand rupees a month for a small business, more as volume grows. WhatsApp's Business API also has its own per-conversation pricing. A good build estimates this for your expected volume before you commit, so the monthly bill never surprises you.

Where chatbots go wrong

• It makes things up. A bot not grounded in your real content will invent answers confidently. Insist on retrieval from your own documents.
• It traps the customer. If there's no easy way to reach a human, a stuck bot becomes worse than no bot. A clean handoff is non-negotiable.
• Nobody updates it. Prices change, policies change. A chatbot needs its knowledge refreshed, or it starts giving wrong answers.
• It's built for a problem you don't have. If you get five enquiries a week, a chatbot is solving nothing — a faster human reply does the job.

A simple test: do you actually need one?

You probably do if: you answer the same handful of questions repeatedly, enquiries come in outside working hours, or you lose leads because no one replied in time. You probably don't (yet) if: your volume is low, every enquiry is genuinely different, or your real bottleneck is fulfilment rather than first response. Be honest here — a chatbot fixes a response-speed problem, not a demand problem.

What to measure once it is live

Most chatbot dashboards lead with the wrong number. “Conversations handled” feels impressive and tells you almost nothing — a bot that answered 500 messages badly scores well on it. Four numbers actually matter:

• Containment rate — the share of conversations resolved without a human. Somewhere between 50% and 70% is healthy for routine enquiries. Much higher and it is probably refusing to hand off.
• Handoff quality — when it does escalate, does the human receive the full context, or does the customer have to repeat themselves?
• First-response time — the metric the bot exists to fix. Compare the month before launch with the month after.
• Enquiry-to-conversation rate — the commercial one. Did more enquiries turn into real conversations because someone replied in time?

Read the failed conversations weekly for the first month. That transcript log is the most useful document a chatbot produces — it tells you exactly which questions your business has never properly answered anywhere.

A worked example: a Kanpur jeweller

Concrete numbers help more than principles, so here is the shape of a typical small-retail case. A jewellery shop gets roughly 40 enquiries a week across Instagram DMs, WhatsApp and a website form. Around 30 of those are the same four questions: are you open today, do you make custom pieces, what is the making charge, and do you buy old gold. The remaining ten are real conversations about specific pieces.

The 30 routine messages are not hard, but they arrive at 9pm and on Sundays, and a customer who waits four hours for “yes, we do custom work” has usually messaged two competitors in the meantime. A grounded bot answers those instantly from the shop's own price list and policy sheet, and routes the other ten to the owner with the conversation history attached. Nothing about the ten changes — those were always going to need a person. What changes is that the person now sees them within minutes instead of after wading through thirty “what time do you close” messages.

Build cost for that scope lands near the bottom of the range above, because the knowledge base is small and there is no inventory integration. The monthly running cost at 160 conversations is modest. The number worth tracking is not messages handled — it is how many of the ten real enquiries convert now that they get answered the same evening.

Hindi, Hinglish, and how customers actually type

This matters more in India than most vendors admit. Your customers will not type clean English. They will type Hinglish in Roman script, switch mid-sentence, use “rate kya hai” and “kitne din lagenge”, and abbreviate heavily. Modern language models handle this far better than the keyword-matching bots of a few years ago, but it has to be tested rather than assumed.

The practical test before you sign off on any chatbot: collect twenty real messages your customers have actually sent — copy them verbatim out of WhatsApp, typos and all — and check the bot answers all twenty correctly. If a vendor demos only with tidy, well-formed questions they wrote themselves, you have learned nothing about how it will behave on Monday.

What you are responsible for after launch

A chatbot is not a one-time purchase, and treating it as one is the most common reason they quietly stop working. Someone has to own three things: refreshing the knowledge when prices or policies change, reading the conversations where the bot failed or handed off, and deciding what to add to its knowledge as a result. That is perhaps an hour a month once things settle, but it is not zero.

Ask any vendor how you update the bot's knowledge yourself, without them. If the honest answer is “raise a ticket with us”, you have bought a dependency rather than a tool. You should be able to edit a document and have the bot pick it up.

How a chatbot project runs

1. Gather knowledge. We collect your FAQs, pricing, policies and catalogue — whatever the bot should know.
2. Build & ground. The bot is set up to answer only from that material, with a clear handoff path to a human.
3. Test on real questions. We run it against the questions your customers actually ask — including the messy Hinglish ones — and tune the answers.
4. Launch on your channel(s). Website, WhatsApp, or both.
5. Tune for 30 days. Real conversations reveal gaps; we close them.
6. Hand over the controls. You get the document that feeds the bot's knowledge, so you can correct a price without waiting on us.

Wondering if a chatbot is right for your business? Tell us the questions your team answers most, and we'll tell you honestly whether a bot helps — and what it would cost. See our AI automation work or send us a two-line message.