Loading...
Intelligence›Anomaly Detection›0 anomalies detected
Live
_predict (inverse)
amount × code × CC
Patterns
low p = anomaly
Method
every PO
Coverage
_predict
POST /api/v1/_predict
{
"from": "purchases",
"where": {
"supplier": "Wärtsilä Components",
"category": "production"
},
"predict": "cost_center"
}
// p(actual) = 0.04 → flagged
{
"from": "purchases",
"where": {
"supplier": "Wärtsilä Components",
"category": "production"
},
"predict": "cost_center"
}
// p(actual) = 0.04 → flagged
How it works
Anomaly detection inverts aito.._predict: instead of asking “what's the most likely value?” we ask “how likely is the value that's actually there?”. A PO from Wärtsilä Components coded to a non-Production cost-centre, or a production purchase posted to a wildly off-pattern account, returns a low probability — that's the anomaly score. No rules, no thresholds to maintain.
Learn more