Almost every business we talk to in India has the same question in some form: "can we do this on WhatsApp?" It is a fair instinct. WhatsApp is where your customers already are — they don't need to install anything, learn anything, or remember a password.
But the gap between "WhatsApp chatbot" as a phrase and WhatsApp chatbot as a working system is wider than most vendors let on. Meta's platform has rules that quietly decide what you can build, and a chatbot that ignores how people actually use WhatsApp will annoy customers faster than no chatbot at all.
Here is what we have learned building these, including the cases where we have told clients not to bother.
First: do you need the API, or just the app?
There are two completely different WhatsApp products, and choosing wrong wastes months.
The WhatsApp Business app is free, installs on a phone, and gives you a catalogue, quick replies, away messages and labels. If one or two people handle your messages and you get a manageable number a day, this is genuinely all you need. No development, no monthly cost. We tell businesses this regularly and it is usually the right answer.
The WhatsApp Business Platform — the API — is what you need when you want software in the loop: automated replies driven by your own data, messages triggered by events in your systems, several agents sharing one number, or conversations that write back into your CRM. It has no interface of its own. It is a pipe, and somebody has to build what sits on either end.
A useful test: if you can describe what you want as "reply faster with the right information", the app plus a trained person may beat any bot. If you can describe it as "our system should talk to the customer without a person", you need the Platform.
The rules that shape what you can build
This is the part that surprises people, and it is worth understanding before you scope anything, because these constraints are not negotiable.
You cannot message whoever you like
WhatsApp is deliberately hostile to unsolicited marketing. You need the customer's opt-in, collected somewhere you can evidence, before you send them anything. Buying a list and blasting it is the fastest way to get your number quality rating downgraded and, eventually, your ability to send restricted.
The 24-hour window
When a customer messages you, a service window opens for 24 hours. Inside it, you can reply freely with whatever content the conversation needs. Once it closes, you cannot simply send a normal message — you have to use a template that Meta approved in advance.
This single rule shapes more architecture than anything else. It means your chatbot has two distinct modes: a conversational one inside the window, and a rigid, pre-approved one outside it. Designing as though you can message anyone at any time produces a system that fails in production.
Templates need approval
Any business-initiated message — an order update, an appointment reminder, a payment link — runs through a template submitted to Meta for review. Approval is usually quick, but it is not instant and it is not guaranteed. Templates that read as promotional get rejected. Build a review cycle into your timeline rather than assuming you can change copy on the day.
The number gets consumed
A number connected to the Business Platform cannot also be used in the regular WhatsApp or WhatsApp Business app. Businesses regularly discover this after connecting the number their customers have used for years. Decide early, and if in doubt start with a new number.
What it actually costs
Three separate costs, and vendors tend to quote only the one that flatters them.
- Meta's messaging charges. You pay Meta per message or conversation, with different rates for different categories — marketing costs more than a utility update, and service replies inside the window are treated differently again. Rates vary by country and Meta has revised the model more than once, so check the current pricing for India rather than trusting a figure in any article, including this one.
- Your provider. Most businesses connect through a Business Solution Provider rather than directly. They charge a monthly platform fee, a per-message markup, or both. This is the number that differs most between vendors and the one worth negotiating.
- The software itself. Whatever actually decides what to say — the flows, the integrations into your CRM or ERP, the handover to a human. This is a one-off build plus ongoing maintenance, and it is where the value is created or lost.
The trap is optimising the first two and skimping on the third. A cheap provider running a bad bot costs you customers, and that does not show up on any invoice.
What to automate first
The best candidates are questions that are asked constantly and have one correct answer. In roughly the order we would tackle them:
- Order and delivery status. High volume, zero judgement required, and customers genuinely prefer not to wait for a human.
- Appointment booking, confirmation and reminders. Reminders alone often justify the project by cutting no-shows.
- Lead qualification. A few structured questions before a salesperson gets involved, so they inherit a warm, pre-qualified enquiry instead of "hi, price?".
- Payment and renewal reminders. Delivered where people actually read them, unlike email.
- The genuine top five FAQs. Not fifty. Five.
Lead qualification is where a chatbot most often pays for itself, and it only works if the answers land somewhere your team acts on. Feeding qualified enquiries straight into a system like Orbis Lead CRM — assigned to a rep, with a follow-up reminder attached — is what turns a conversation into a deal. A bot that collects information nobody reads is theatre.
Where WhatsApp chatbots go wrong
Almost every failure we have been asked to fix comes down to one of these.
- No fast route to a human. If a customer types "talk to someone" and the bot keeps offering menu options, you have made your service worse. Handover should be one message away, always.
- Rebuilding your website inside a chat. Deep nested menus are miserable on a phone. If a task needs more than a few steps, send a link.
- Pretending to be a person. Customers work it out, and they resent it. Say it is an assistant. Nobody minds a bot that is upfront and useful.
- Treating it as a broadcast channel. Because WhatsApp is personal, over-messaging costs you more than it would on email. People block, and blocks damage your quality rating and your reach.
- An LLM with no guardrails. Language models make WhatsApp bots dramatically better at understanding messy, mixed-language, typo-ridden real messages. They will also state a price or promise a refund policy that does not exist. Ground them in your actual data, constrain what they are allowed to assert, and keep anything involving money or commitments on rails. This is the part of AI and ML work that takes the real engineering.
When we tell people not to build one
If your message volume is low enough that a person handles it comfortably, a chatbot adds cost and a layer of frustration for no gain. If your product genuinely requires consultation — where every enquiry is different and the answer depends on judgement — automation will hurt you. And if your underlying data is a mess, a chatbot only exposes that faster: it cannot tell a customer where their order is if your own systems do not reliably know.
Fix the data and the process first. The bot is the last layer, not the first.
A realistic path
Start narrow. Pick the single highest-volume question you get, automate only that, and keep a human one tap away. Measure how many conversations resolve without escalation and how many customers ask for a person anyway — that second number tells you more than any satisfaction score. Expand only into the things the data says are working.
The businesses that get real value from WhatsApp are not the ones that automated the most. They are the ones that automated the few things customers were happy to have automated, and left everything else alone.
If you are weighing this up and want a straight answer on whether it is worth building for your situation — including the possibility that it is not — talk to us. You will speak to an engineer, not a salesperson.