What is anomaly detection?

Anomaly detection is finding values or patterns that don't fit what is normal. Wikipedia describes it as the identification of "rare items, events or observations" that differ markedly from most of the data and "do not conform to a well defined notion of normal behavior". It learns what normal looks like, then flags what doesn't fit.

Rules and detectors

Some errors can be stated as rules: the rate on a ticket should match the contract, every ticket should have an invoice. Checking those is a job for matching rules. They find every case and need no history. From Quarri's own work with a forestry operation: a levy rate was being accrued at a fraction of the rate actually invoiced, a gap of 500%+. A gap like that is a broken relationship between two records, the kind of error a rule catches.

Anomaly detection is for what can't be written down in advance. A load far heavier than usual, a price outside its range, a day's production far below normal: these catch typing errors and breakdowns.

Normal depends on context

Wikipedia describes two harder types. A contextual anomaly "may be normal in summer but unusual in winter". A collective anomaly is a group of records "anomalous as a whole, even when its individual observations are not anomalous by themselves", such as a run of slightly short loads from one contractor.

Timber data has strong seasons, such as spring thaw and mill shutdowns. A detector that doesn't know them fires every year, and people learn to ignore it.

What to do with an anomaly

Route it to the person who can explain it, with the records beside it. Label it as an error or a real event, so the detector learns and the same pattern isn't flagged again.

What it isn't

Anomaly detection isn't proof of a problem, only a reason to look. Nor does it replace reconciliation, which checks every record against its counterpart.

Quarri for finance and strategy teams is built for the people who close the month, explain the margin and answer the board.

Quarri is an AI-native data platform for the timber supply chain. It connects buying, production, sales and inventory for forest management, sawmill, wood products and pulp, paper and packaging operators.