Two IP cameras above the bar + AI on your server. They count bottles, measure levels and detect pours. POS cross-check shows where 20 % of the alcohol disappears — and who pours it for free.
An average bar loses tens of thousands of crowns a month to overpour, unbilled shots and spillage. Weighing bottles once a week fixes nothing.
Gap between what was poured and what was sold. Sources: Diageo Bar Academy, BevSpot.
Bartender weighs, counts, writes down. 5 hours a week × hourly rate. Data is stale after one weekend.
Problem that starts Tuesday is found Sunday. Too late to fix, too late to prove.
Largest single share of losses. Without continuous monitoring you can't prove when or who.
Green boxes = detected bottles. Yellow line = liquid level. All computed locally on an edge device, no frames leave the venue.
Detection conf 0.97 · level 86.7 % · level conf 1.0
Level range 37–92 % · works in dim light
YOLOv11 baseline · no fine-tuning · top conf 0.45
Typically 1–2 cameras for a small back bar, 3–4 for a larger venue (Reolink / Hikvision / Dahua). Edge device next to the bar (Jetson Orin Nano). Everything processed locally — video never leaves the venue.
YOLOv11 recognises every bottle on the shelf and assigns a brand. After fine-tuning on your 30–50 labels we reach 95 %+ accuracy.
CV segmentation of the meniscus in a transparent bottle. ±3 % accuracy. Works in dim bar lighting.
SAM2 + temporal classifier detects the act of pouring. Linked to a bottle and timestamp — pairs with POS records.
Integration with Storyous and Dotykačka. We compare detected consumption with POS sales — gap over 10 % triggers an alert.
Competitors require hardware on every bottle — pour spouts, Bluetooth scales, RFID tags. We only need a pair of cameras.
| Solution | Hardware | Price/mo | Continuous | EU support |
|---|---|---|---|---|
| WISK, Partender | Bluetooth scale / manual | $165–299 | weekly | no |
| BarVision | IoT pour spouts (per bottle) | undisclosed | yes | no |
| Decibar, Pavlasoft | Bluetooth scale / manual | from €60 | weekly | CZ only |
| BarBee | 1–2 IP cameras + edge box | from €100 | 24/7 | CZ + EN |
Section 316 of the Czech Labour Code permits monitoring with a substantial reason — 20 % shrinkage qualifies. Camera points down at the shelf and hands, not faces. A DPIA is done, employees are informed in advance, the area is marked.
Bottle detection: >95 % after brand fine-tuning. Level estimation: ±3 % on transparent bottles. Pour events: phase 2 of the pilot. Reliable on a back bar with decent lighting.
Typically 1–2 IP cameras (Reolink / Hikvision, 4K + WDR). For a back bar with 30–50 bottles one front + one top-down view is enough. Bring your own or we include them in the pilot.
No. All processing runs locally on an edge device at your venue (Jetson Orin Nano). Video never leaves. Only aggregated numbers go to the cloud — bottle count, average level, detected pours.
Yes — Storyous and Dotykačka via official APIs. Custom work for other systems (iKelp, Pavlasoft, in-house). Reconciliation is automatic, gaps over 10 % trigger alerts.
90 days free. Then you see our report of what AI found and decide — continue at from €100/month, or stop (you can keep the hardware at cost). No lock-in.
We want to validate accuracy in real-world conditions on different shelf layouts. We supply the camera and edge device, deploy it, and after 90 days we show you how much we found. You decide whether to continue.