Assistant, 09:00
ai solutions

Owner's assistant

Restaurant chain

clientRestaurant chain
year2026
workMorning digest, Anomalies, Profit from POS data, Antifraud

task

The owner used to learn about a failing location a week later and saw profit together with the accountants, a month later. Loyalty bonuses were being redeemed, and nobody checked them against receipts. We decided a chain cannot be run this way.

what we did

  1. Every 6 hours we pull location data from the POS and compare yesterday with the median of four matching weekdays.
  2. At 09:00 the owner gets one message per brand: deviations with the likely cause and profit since the 1st of the month.
  3. Antifraud matches bonus redemptions against receipts for each cashier.
  4. Every night the assistant goes through a quality check on 28 real questions from owners.
  5. We run the server and the models, reconcile profit with the P&L, and show every new format before it goes out.

result

83%of redeemed bonuses came from one cashier, and the client confirmed the scheme
11 → 1false Friday alerts after switching to the weekday median
18,3%the chain's October margin is visible before the books close
106pairs of duplicated bonuses in a month, caused by a POS integration failure

how it works

POSYesterday's POS data for each location.
comparisonMedian of the same weekday, deviation threshold.
diagnosisExternal cause, the location, or the whole city.
morningOne message to the owner in Telegram.
discuss a project
discuss a project on telegram
€250 reviewbook