You measure the variance and then chase it to a cause. The usual suspects are portioning drift, waste and spoilage, yield loss in prep, over-production, theft, supplier price creep, and specials that were never costed. A weekly stock count narrows it down — if the gap widens the week you ran an un-costed special or a portion crept up, the count points you at it while you can still act.
From Restaurant Food Cost Control: Theoretical vs Actual, and How to Close the Gap — the full reading, with the method behind this answer.