WhatsApp AI Sales Assistant
A WhatsApp Business agent for a Pakistani-dress retailer that answers product questions, takes orders and bookings in two languages, and escalates cleanly when it should.
Problem
Customers messaged on WhatsApp in a mix of English and Bangla, at all hours, with everything from stock questions to orders to appointment requests landing in the same inbox. Bad phone numbers, incomplete addresses and timezone slips on bookings were quietly turning into lost orders.
Before
One inbox for stock questions, orders, complaints and booking requests, with no consistent handling for incomplete details or an angry customer.
After
The agent answers from the real, current product list, saves orders and bookings itself, works around a bad phone number or vague address instead of getting stuck, and hands off to a human only when a conversation actually needs one.
Interactive workflow
Click through the actual node graph
Demo video
See it running end to end
One address, two very different messages
WhatsApp sends both its one-time setup check and every real customer message to the same address. The workflow replies "got it" immediately to both — the setup check gets its confirmation, a real message gets a quick acknowledgement — before doing anything else, then makes sure what arrived is actually a customer message (and not just a "delivered" or "read" notice) before pulling out the sender's number, their message, their name and which business number they wrote to. Answering fast matters here too: if WhatsApp doesn't hear back quickly, it resends the same message, which risks a customer getting answered twice.
Working around problems instead of getting stuck
A bad phone number or a vague address isn't treated as a dead end. The assistant asks once for the missing detail, and if it's still unclear, it saves whatever the customer gave along with a short note for staff to double-check by phone, rather than asking the same question over and over. Bookings work the same way in spirit but with a firmer rule: the assistant is never allowed to tell a customer their appointment is confirmed until the calendar slot has actually been created, and every booking is saved with the correct Bangladesh time attached so it can never land six hours off.
Every handoff leaves a paper trail
Angry customers, refund questions, and repeated frustration all get logged for a staff member before the customer is ever told someone will follow up — never after, and never skipped. That way, every handoff stays visible to staff even if the customer doesn't message again. The assistant also remembers what each customer already said earlier in the same conversation, so it never asks for a name or address it's already been given.
Impact
Orders and bookings captured directly from chat, in the customer's own language, with a paper trail for every escalation instead of a message someone might miss.