A gradual shift in the distribution of a column's values over time, away from what it used to be.
Reviewed by Francisco Ferreira ·
Data drift is when the values in a column slowly stop looking like they used to — the average creeps up, a category that was rare becomes common, a currency or unit silently changes. Unlike a hard break, drift is gradual, which is exactly why it goes unnoticed for so long.
It quietly corrupts anything built on top: dashboards, machine-learning models, business metrics. By the time a chart looks 'off,' the drift has usually been underway for weeks, and the baseline you'd compare against has already moved with it.
Tabkeel compares each column's current distribution to its learned baseline and surfaces drift while it's still small enough to fix cheaply.