What is shrinkage and why three points vanish with no culprit
Shrinkage is the gap between the stock the system records and what actually exists in the store, measured as a percentage of sales. It is not the same as known waste. Waste is known loss: expired product, damage, disposal. Shrinkage includes the unknown: theft, process error, fraud.
Confusing the two is expensive, because each is solved differently. Waste you reduce with forecasting and replenishment. Unknown shrinkage you reduce by finding where it lives.
How the operation measures it
Shrinkage shows up at inventory count. You count what exists, compare it to what the system says, and the gap is the loss. Cycle counting measures this by section across the month; book inventory, once or twice a year.
The number usually comes as a percentage of sales. Three points of shrinkage in a thin-margin chain can be the whole year’s profit.
What the number hides
The aggregate is useless for acting. “It rose from 1.8% to 2.4%” does not tell you what to do. Shrinkage becomes a decision only when it is decomposed: how much is external theft, how much internal, how much receiving error, how much self-checkout.
Each source has a different countermeasure, and investing in the wrong one burns money. A camera does not fix a registry error. A process audit does not catch theft at the register. The combined number hides this.
Why this changes an AI project
An AI loss-prevention system does not act on “shrinkage”. It acts on a specific source, with specific data. Asking AI to “reduce loss” without decomposing it is asking it to guess where to bet.
The acceptance criterion has to be per source, not on the total. It is the same logic as OSA measured at the shelf and camera audits under LGPD: the data has to have an address.
Think about the loss number on your monthly slide. Is it a total, or is it decomposed by source? If it is a total, it is telling you something leaks and hiding where.
Tell me the loss term your operation reports and how it is measured today. In one hour I will send back a one-page card: the formula, the measurement points, what the number actually means, and where it misleads. If the card shows the system producing that number has to be rebuilt to separate the sources, the two-week Diagnóstico is the next step.