A chatbot is cheap to build and easy to demonstrate, which is why so many businesses have one that nobody uses. The ones that pay for themselves share a pattern: they live where customers already are, which in India means WhatsApp, and they answer the handful of questions that make up most of the incoming calls. This article is about finding that pattern for your business and doing the arithmetic before you spend.
Start by counting the questions
Take a week and have whoever answers the phone or the WhatsApp number keep a tally of what people ask. Every business we have done this with finds the same shape: a small number of questions account for most of the volume.
For a distributor: where is my order, what is the price of this item, is this in stock, send me the invoice. For a clinic: is the doctor available today, is my report ready, what are the timings, how do I book. For a college: what is the fee, is admission open, what documents are needed, when are results. For a retailer: do you have this, what time do you close, do you deliver.
If the top five questions are more than half of the volume, a chatbot has a job. If every conversation is different, it does not.
Where the ROI is real
WhatsApp first, everything else second
Indian customers do not download your app or find the chat widget on your website. They message the number on your board or your invoice. A WhatsApp Business API bot answers there, in Hindi, Marathi or English, on any phone, without an install. A website widget is a reasonable second channel; an app-only bot is usually a waste.
Order and delivery status
"Where is my order" is the single most common question for anyone who ships goods. A bot that reads the order number or the customer's phone number, looks it up in your system and replies with the status and expected date removes those calls entirely. This needs an integration with your order or dispatch system, and that integration is the real cost and the real value.
Fixed-answer FAQs
Timings, location with a map link, price lists, documents required, return policy, fee structure. These change rarely and are asked constantly. A bot handles them perfectly and a language model makes it tolerant of how people actually type: spelling, Hinglish, voice notes transcribed.
Appointment, booking and enquiry capture
Booking a slot, registering interest in an admission, requesting a callback. The bot collects the details, writes them into your CRM or a sheet, and confirms. Nobody sits at a desk copying WhatsApp messages into Excel any more.
Reminders and outbound messages
Strictly this is automation rather than a chatbot, but it uses the same channel and is often the best return of all: payment reminders with a UPI link, appointment reminders, fee-due notices, report-ready messages. See our notes on workflow automation.
Where it fails
- Complaints and angry customers. A person who has a problem wants a person. Route to a human immediately and visibly. Making them fight the bot first is how you lose them.
- Negotiation and B2B pricing. Trade customers expect to talk about rates and credit. A bot quoting list price annoys them.
- Anything needing judgement. Medical advice, legal questions, exceptions to policy. The bot should say "let me connect you" and mean it.
- Businesses with no system behind the bot. If order status lives in a notebook, the bot has nothing to read. Fix the operational system first.
- Bots that pretend to be human. Customers notice, and they stop trusting every message from your number.
- Free-form generative bots with no boundaries. A language model allowed to say anything will eventually promise a discount you do not offer or give an answer that is wrong. Constrain it to your content and to the systems it can look up.
A simple way to estimate ROI
You need five numbers, all of which you can measure in a week.
| Input | Example (small distributor) |
|---|---|
| Incoming queries per month (calls plus WhatsApp) | 3,000 |
| Share the bot can fully answer (from your tally) | 60 percent |
| Minutes a person spends per query | 3 |
| Loaded cost of that person per hour | ₹150 |
| Bot cost per month (platform, WhatsApp fees, support) | ₹8,000 |
Monthly saving is queries multiplied by share automated multiplied by minutes, divided by 60, multiplied by hourly cost. In the example: 3,000 × 0.6 × 3 ÷ 60 × 150 = ₹13,500 per month in staff time, against ₹8,000 running cost. Net ₹5,500 a month, which repays a ₹1.5 lakh build in a little over two years on staff time alone.
That looks thin, and often it is, on staff time. The cases where the bot clearly wins add one of two other numbers.
- Queries that currently go unanswered. If 30 percent of calls are missed after hours or during rush, and some of those are orders or bookings, the recovered revenue dwarfs the staff saving. Estimate it conservatively: missed queries multiplied by the share that would have converted multiplied by average order value.
- Staff redeployed, not saved. The person who was answering "where is my order" is now doing collections or outbound sales. Value that at what the new activity produces.
Do this on paper before building anything. If the number is marginal with honest inputs, do the reminder automation alone, which is cheaper, and revisit the bot when volume grows. If you are also weighing the broader case for AI in a small business, the same discipline applies.
What has to be behind the bot
A bot is the visible ten percent. The rest is plumbing, and the plumbing decides whether it works.
- A WhatsApp Business API account through an approved provider, with your business verified and message templates approved. This takes days to weeks and is best started early.
- A system it can read from. Order status, appointment slots, fee ledgers, stock. If these live in Tally, a custom ERP or a booking system, the bot needs an API or a read-only connection to each.
- A knowledge base you maintain. Timings, prices, policies, documents. Someone owns keeping it current, or the bot starts giving last year's fees.
- A handover path. A human inbox, staffed, where conversations land when the bot gives up, with the full transcript attached so the customer does not repeat themselves.
- Logs and consent records. Every message stored, opt-ins recorded, personal data handled as the DPDP Act expects.
What it costs to build
As a rough guide for Indian projects: a WhatsApp FAQ and enquiry bot with no integrations, roughly ₹40,000 to ₹1.5 lakh. A bot integrated with your order, billing or appointment system so it can look things up and write things back, roughly ₹1.5 to 5 lakh. Running costs are the WhatsApp Business API conversation charges, any language-model API usage, hosting, and a support arrangement, together typically a few thousand rupees to a few tens of thousands a month depending on volume. All of it depends on scope and on how clean the systems behind it are.
How to run one well
- Launch with the top five questions only. Add more as the logs show what people ask.
- Put "talk to a person" one message away, always, and staff it during business hours.
- Read the transcripts every week for the first two months. That is where you learn what the bot gets wrong and what customers actually want.
- Measure four things: queries handled without a human, handover rate, response time, and what happens to call volume at the counter.
- Never let it send marketing without opt-in. WhatsApp will block the number, and rightly.
- Tell customers it is automated. Trust is worth more than the illusion.
Where SailRight fits
We build customer service bots and automations on WhatsApp and the web, integrated with the billing, order and booking systems behind them, including our own RightSutra suite. If you want us to do the ROI arithmetic with your real numbers before you decide, send us your weekly query tally and we will reply with an honest estimate.
- Chatbots
- WhatsApp Business
- Customer service
- ROI
- Automation
- AI