Back to work

Flores da Ilha

A service and management ecosystem for a flower shop, with an AI answering on WhatsApp and a panel that closes the order.

1.400conversations over two peak days
01The problem

The shop sells through WhatsApp conversations, and every sale means understanding the order, finding the product in the catalogue, measuring delivery from the address, charging, and issuing the work order. Without a system, that is done message by message, by hand.

02The decision

Put an AI on the WhatsApp front line, tied to the shop's catalogue and knowledge base so it answers with real products and real prices. It understands audio and images, measures delivery from the address, generates the payment and issues the work order. On the other side, a panel where the team reads the same conversation, pauses the AI for as long as they want and takes over by hand.

03The outcome

Over the two peak days of Valentine's Day 2025, the operation handled 1,400 conversations with 20% converting into a sale, the AI on the front line, the human team taking over at handoff.

Period
2025
Role
Architecture, AI and product
Tech
ReactTypeScriptPostgreSQLn8npgvectorWhatsApp API
  • Panel and connection

    A login carrying the flower shop's brand, then a grid of cards leading to each area. In Connection the person reads the step by step, scans the QR code with the phone and watches the screen switch to connected. That is the point where the shop's WhatsApp starts being answered by the system.

  • Product catalogue

    A table with category, name, price and description, text search and filters by category and price range. The person registers the product with a photo, price, category and description. It is this catalogue the AI consults during the conversation, and it is from here that the photo sent to the customer comes.

  • Knowledge base

    A screen where the person uploads a PDF, gives it a title and a category, and sees the documents as cards with search. The file is sliced and indexed for search by meaning, and becomes the source the AI reads before answering anything about the shop.

  • Order and work order

    A list of orders with search by customer or number, a status filter and a card for the day's orders. The record carries customer, address, delivery date and shift, products with quantity, subtotal, discount, delivery fee, total, the card message and the name of the person receiving the gift. At the end the work order comes out as a PDF, sent over WhatsApp. The same record can be filled in by hand when the order arrives from outside.

  • Customers and reactivation

    A table of customers with name, phone, type and interest, text search, filters and a side record that opens each person's detail and history. Behind it, two routines run on their own: one sends the right message on the customer's special date, the other sweeps stalled conversations and classifies the reason before picking the contact back up.

  • The operation's numbers

    A panel with sales, profit, orders, average ticket and delivery fees, each one as a total, for the month and for the day, plus charts of sales over time, conversion, cancellation, best selling products and categories, and delivery shifts. In the same area the person enters expenses by category and sees the effect on the result. There is also a record of the AI's usage and cost per flow.