A stock count that everyone distrusts is worse than none, because people start guessing. Counts go wrong for predictable reasons: goods leave without being recorded, a delivery is put away in the wrong place, or a count is typed in twice.
Count one shelf at a time, record the count against the system's figure, and investigate only the differences. Small, regular counts beat one huge annual count, and they show you where the losses happen.
Try this now: pick your three fastest-moving items and count them every Monday for a month. The pattern will show you where the trouble is.