The situation
A Nairobi ecommerce retailer sells mostly through WhatsApp. Customers find a product on the website, tap WhatsApp, and ask the same questions: is it in stock, how much, how fast can you deliver? Answering all of that by hand meant slow replies at busy times and none overnight.
What I built
An AI sales agent that answers the store’s WhatsApp in natural Kenyan English, from the live catalogue, and hands over to a human when it should.
- Replies in seconds in the business’s own tone, 24/7
- Product catalogue synced both ways with Odoo over JSON-RPC: prices, stock, SKUs and images always current
- Every customer categorised by sales stage, so the team sees who is browsing and who is ready to pay
- Captures orders and alerts admins instantly
- Escalates anything it can’t answer to a human, with the full conversation
- n8n runs scheduled follow-ups and marks abandoned chats for a nudge
- Ships with a self-diagnosis health page so problems get caught, not discovered by customers
The result
First response went from hours to seconds, and routine questions (price, stock, delivery) are handled without anyone picking up a phone. The team spends its time on the conversations that need a person.
Measuring what it sells
Because every WhatsApp conversation is exportable, this is one of the few funnels where the chat stage is fully measurable. I’ve set it up as a tracked funnel: sessions, WhatsApp clicks, agent conversations, qualified, order intent, paid. Each conversation carries its source, so we can see which pages and campaigns produce paying customers, not just chats.