Numbado
Back to cases
Computer Vision

Camera-based sales control (Computer Vision)

We turn camera footage into a sales-control tool. An AI model recognizes the moment a customer hands over cash and reconciles every such transaction with orders in the CRM — if money changed hands but there’s no order in the system, the owner sees it.

Client: A cannabis cafeshop (Alicante, Spain) with a bar counter and a flow of cash transactions. Sales go through the bartender, who is supposed to enter every order into the CRM.

Introduction

Cameras in a venue usually work only as a “just in case” archive — reviewing thousands of transactions per shift by hand is unrealistic. We turn that video into working analytics: the model finds the relevant moments itself and matches them against the till, with no manual review.

Pain → Request

The owner suspected the bartender was running some sales past the till — selling his own goods and not entering the order into the system. Proving or disproving it by eye from camera footage was impossible. The ask: get an objective check of the hypothesis, not guesses.
PAID: YES
CAM_04 · CHECKOUTREC 00:14:32
Camera frame: the system detects the moment of payment

The project

We trained an AI model to recognize cash being handed from customer to bartender and to tell it apart from giving change or handing over goods. Every cash transaction was reconciled with CRM orders for the same time. Mismatches — payment made but no order in the system — were collected and shown to the owner.
Video feed
Object detection
Sync with CRM
Flag ✓
How it works: from video to a CRM record

Impact

The owner’s hypothesis was confirmed: some orders really did bypass the till — the bartender was selling his own goods instead of the bar’s. For the first time the owner got an objective picture of what happens behind the counter instead of suspicions. It gave a fact-based basis for a conversation with staff, not gut feeling.

Conclusion

We close a blind spot that can’t be controlled manually: we match real actions at the counter with what reaches the till. The approach scales to any venue with cash sales and cameras — and works continuously, not as a one-off.
100%
of transactions covered
24/7
around the clock
Metrics after the first month in production

Tech stack

Computer Vision

Have a similar task?