Because each is using a different denominator, panel or time basis for the same word. "Pipeline" measured as created-in-period versus open-at-period-end are different quantities. Fix it by publishing one definition per metric with the filter logic attached, then making every report reference it.
Run the same four checks you would use on an external benchmark: does the denominator match, does the population match, is it the same statistic (median vs mean), and is it the same time basis? Most internal disagreements fail on the first or third, which means the gap is measurement rather than performance.
Every claim above comes from work LeanScale published. These are the sources.