WhatsApp Chatbot Development

WhatsApp is not a channel you control, and teams that treat it like email discover this expensively. Meta enforces opt-in, restricts free-form messaging to a 24-hour window after the customer's last reply, requires pre-approved templates outside that window, and maintains a quality rating on your number that falls when people block or report you. A falling rating cuts your sending limits, and in serious cases the number is disabled. So the first conversation we have on any WhatsApp project is about which of your intended messages will actually clear those rules — and we would rather tell you a campaign idea will damage your rating before we build it than afterwards.

The 24-hour window drives more design decisions than anything else, and it is the detail most often missed in planning. Inside it you can say anything; outside it you can only send an approved template. That means a follow-up sequence spanning several days is not a matter of scheduling messages — it needs templates written and approved in advance, and each one opens a new billable conversation window. Designing a flow without accounting for this produces something that works perfectly in testing, where every message is within the window, and then fails or costs several times the estimate in production.

We build bots that let people out, and this is not just courtesy — it is protective. A bot that will not connect a customer to a human gets blocked, and on WhatsApp a block is not a lost conversation, it is a hit to the quality rating that governs your entire number. So asking for a person always works, and the bot escalates on its own when it detects frustration, cannot classify a message confidently, or hits a complaint or cancellation. The commercial incentive here is unusually well aligned with the customer's interest, which is worth pointing out because on most channels it is not.

Voice notes and informal language matter more on WhatsApp than anywhere else. Customers send audio, use heavy shorthand, switch between languages mid-sentence, and put three unrelated questions in one message. Menu bots collapse immediately under this, which is why so many WhatsApp bots feel broken in real use even though they passed testing. We transcribe audio, handle multi-intent messages by addressing each part, and test against real message samples from your existing threads rather than the clean phrasing that ends up in a specification document.

The number you register is a decision with consequences worth understanding before you make it. A number on the Business API cannot also be used in the regular WhatsApp or WhatsApp Business app on a handset — so if your team currently answers on a phone, that stops. Most clients either migrate the business number and move day-to-day chat to a new one, or register a fresh number for automation and keep the old one running during transition. Either path is workable; discovering the constraint after registering the number your customers already know is not.

We instrument the things Meta cares about, not only the things you care about. Alongside resolution rate and response time, we track block rate, report rate, template rejection reasons, and quality tier. These are leading indicators: a rising block rate signals a problem well before sending limits are cut, and at that point it is a copy or targeting fix rather than an account recovery exercise. Most WhatsApp deployments monitor none of this and find out something is wrong when messages stop being delivered.