Challenge
Orders don't arrive as orders. They arrive as "the usual, 2 boxes" at 11pm. As a voice note in mixed languages. As a product nickname exactly one person in the company understands. Someone experienced has to decode every one of them—and that person becomes the bottleneck for the whole outlet. The group needed a dependable order record without stopping to build an ordering API first.
Approach
Five outlets went first, each with its own supervisors, customers, catalog, product aliases, units of measure, languages, and exception rules.
Every customer's own order history does the matching, so "the usual" resolves to the right product and the right quantity for that customer—not a generic best guess.
Anything low-confidence, unusual, or expensive stops for a human before it becomes an order. The AI never quietly guesses with a customer's money.
Boni runs the order backend and hands over a daily export the group uses in its own reporting. Nothing new for their team to babysit.
Conversation, order state, review decisions, escalations, and follow-up all live in one place in Bow Chat.
Outcome metrics will be added only after the agreement is signed and the first implementation phase is paid.
What it does today