Skip to content

Stockertim Dictionary · Technology · Human resources

Precision

Also called positive predictive value, precision (classification)

In short

In a classification model, the share of cases the model flagged that turned out to be true, such as flagged employees who really did leave.

  • Advanced
  • Ratio or metric
  • Checked

In detail

Precision answers the question: when the model says yes, how often is it right? It is calculated as true positives divided by all positive predictions, that is, true positives plus false positives. A model that flags 80 employees as likely to leave, of whom 40 actually leave, has a precision of 0.5.

Precision is paired with recall, which asks how many of the real cases the model found. The two usually trade off: setting a lower threshold flags more people, finding more real cases but with more false alarms, so recall rises and precision falls. Which matters more depends on the cost of each kind of error. In HR, a false positive might mean labelling a loyal employee a flight risk or wrongly flagging someone for misconduct, which can cause real harm; a false negative might mean missing a good candidate.

Precision and recall can differ between groups even when overall figures look good, which is a form of unfairness in itself. The UK regulator's guidance on AI and data protection treats the statistical accuracy of a system's outputs as part of meeting the fairness principle, so these measures are chosen and monitored deliberately.

Formula

Precision = true positives / (true positives + false positives)

true positives
cases flagged by the model that were real
false positives
cases flagged by the model that were not real

Worked example

A made-up attrition model at Northgate flags 80 employees; 40 of them leave within the year.

  1. False positives: 80 - 40 = 40
  2. Precision: 40 / (40 + 40) = 0.5

Half of the flags are correct.

In practice

Model reports in HR show precision and recall at the chosen threshold, broken down by group, and explain why that threshold was chosen given the costs of each type of error.

Watch out

  • High precision with very few flags can hide a model that misses most real cases.
  • Check precision by group: equal overall figures can mask unequal error rates.

Learn it properly

Official sources

Facts last checked on 3 November 2026.

Education, not advice: the Stockertim Dictionary explains what terms mean. It never recommends buying or selling anything.

English