Most booking requests came in after hours, when no one was there to answer. A WhatsApp chatbot picked up the slack — and started closing bookings a phone line never could.
The marketplace connects customers with home service providers — plumbers, electricians, appliance repair — and almost all of its inbound demand arrived one of two ways: a phone call during business hours, or a WhatsApp message at any hour, since that's where the marketplace's customers already were. The phone line was staffed and worked fine during the day. WhatsApp was the problem. Messages sent at 9pm sat unanswered until morning, and in a category where customers are usually comparing two or three providers at once, the first one to reply is often the one that gets the job.
The instinct to "just add a chatbot" undersells how specific this problem actually was. A generic FAQ bot wouldn't have helped — the message that needed answering wasn't "what are your hours," it was "can someone come fix my geyser tomorrow morning," which requires checking real provider availability, capturing the right details (service type, location, urgency), and actually creating a booking, not just deflecting the conversation until a human woke up.
So the bot was scoped around that one job: understand what service the customer needs and when, check live availability against the marketplace's existing scheduling system, and either confirm a slot on the spot or offer the nearest alternatives. We built it on the WhatsApp Business API with a hybrid design — structured flows for the booking mechanics, where getting it wrong means a wrong appointment, and a language model layer for understanding free-text requests that don't arrive in neat form-field order, which is how people actually type into WhatsApp.
Escalation mattered here too, just differently than in a support queue. Anything outside a clean booking — a complaint, a price negotiation, a request the model wasn't confident it parsed correctly — handed off to the human team the next morning with the full conversation attached, so no context got lost overnight. The bot's job was never to replace the team; it was to make sure a customer messaging at 11pm didn't just give up and message a competitor instead.
Since launch, the bot handles the large majority of after-hours inquiries end-to-end, converting a share of them into confirmed bookings before the office even opens. The support team now starts each morning with a queue of already-scheduled jobs and a short list of conversations that genuinely needed a person — instead of a backlog of messages that had been sitting since the night before.
This engagement drew on our Chatbots and AI Agents services.
It's grounded in the real availability and service data behind the booking system, so it only offers slots that actually exist — no placeholder answers that fall apart when someone tries to confirm.
It hands off to the team the next business morning with the conversation context attached, rather than pretending it can resolve something outside its scope.
Yes — it reads and writes directly to the same scheduling system the team already used, so bookings made by the bot show up exactly like ones taken by a person.
A few weeks from scoping the booking flow to a monitored production launch on WhatsApp Business API — most of the time went into getting the availability logic right, not the conversation design.
Yes — the booking logic was built to branch by service type and location from the start, so it scales to a multi-category or multi-location business without a redesign.
A practical comparison of rule-based and LLM-powered chatbots — cost, predictability, and how to decide which one your use case actually needs.
Read the guideA practical checklist for taking a chatbot from a working demo to a production deployment that holds up against real customers.
Read the guide