How many standard deviations a value sits from the average of its own history; used to flag a measure that moved beyond its normal range.
A z-score measures how unusual a value is against its own history, in standard deviations. A dashboard flags a measure when its z-score passes a stated limit, often 2 either way. The flag is computed by a tool from the data; the AI model only explains a movement the tool has already found and figured.
Z-score = (this period's value − mean of the comparison periods) ÷ standard deviation of the comparison periods. In Excel: =(B14-AVERAGE(B2:B13))/STDEV.S(B2:B13). Exclude the current period from the mean and standard deviation.
Over the last twelve months revenue averaged $200,000 with a standard deviation of $10,000. This month came in at $172,000: (172,000 − 200,000) ÷ 10,000 = −2.8, so it is flagged.
Applied to seasonal measures, so every December is flagged. A history with one huge month inflates the standard deviation and hides later movements. A near-zero standard deviation turns trivial changes into large scores. The full guide is AI dashboard.