From consultant to closer.
The goal: upgrade an existing consultant bot into a full sales manager that recognizes products from photos and places orders.
We added computer vision to the existing bot rather than rebuilding it: a customer sends a photo, the agent recognizes the model and finds it in the catalog, reads text from screenshots and labels with OCR, then brings the dialogue to an order and writes it into 1C. It works in Viber and Instagram, the channels where the store's customers already write.
Before, a manager sorted every "I want this one" photo by hand, the bot only answered questions, and orders were carried into the accounting system manually. Now the model is recognized automatically, the agent closes the dialogue with an order, and that order lands in 1C straight from the chat, without anyone retyping it.
What was in the way
- ✗The existing bot only consulted — it did not sell
- ✗Customer photos (“I want this one”) were sorted by hand
- ✗Orders were carried into the accounting system manually
What the system does
Search by photo
A customer sends a photo — the agent recognizes the model and finds it in the catalog.
Order placement
The agent brings the dialogue to an order and writes it into 1C.
OCR in the dialogue
Reads text from screenshots and labels customers send.
On top of the existing bot
Not from scratch: the existing consultant bot was upgraded into a seller.
How we roll it out.
Requests and access analysis
We study historical requests, verify CRM and API access, and score the automation potential.
A slice of real traffic
The agent runs on part of the live requests with weekly syncs — real dialogues, not a demo.
Metrics and scale-up
We evaluate volume, response speed and escalation share — then expand the agent to full traffic.