Anomaly detection is the automatic identification of data points that deviate significantly from the expected pattern. In hospitality it is used to spot unexpected deviations in revenue, cost of goods, payment transactions or labor costs early — before they show up as a monthly loss in the BWA (the German monthly management accounts).
Typical areas of application
| Area | Example anomaly |
|---|---|
| Revenue | Daily revenue 25% below forecast with no known cause |
| Cost of goods | Food cost ratio suddenly +4% within one week |
| Void rates | Void share three times the weekly average |
| Payment methods | Cash share dropped sharply — possible input error |
| Labor costs | Overtime 40% higher than the previous week with no revenue increase |
In everyday use, the Morning Briefing in Kiosa One View does this job: early-warning notes from the POS and staff data are in the summary every morning, before you have to look for them yourself.
Methods
Simple anomaly detection works with statistical thresholds: if a value is more than two standard deviations from the average, an alert is triggered. More advanced systems use context-dependent models — a revenue drop on a Monday is normal, the same drop on a Saturday is an anomaly.
Combination with forecasting
Anomaly detection is closely linked to forecasting: the forecast defines the expected level, anomaly detection measures the deviation from it. A system that knows a Friday evening should be at €14,000 can send an alert straight away at €9,000 — long before the operator notices manually.
Business value
Anomalies detected early allow quick action: with unusually high cost of goods, the kitchen can be asked immediately. With unusual void rates, individual payment transactions can be checked. Every hour earlier a problem is detected means less loss.
Related terms
- Forecasting — basis for the expected values in anomaly detection
Track this KPI automatically
Kiosa calculates all restaurant KPIs automatically every day — for every location, without spreadsheets.
Book a Kiosa demo →