Inventory shrinkage is the difference between the inventory your records say you have and the inventory you actually have, valued at cost. You account for it by writing the missing units off the inventory asset account and recognizing the loss as an expense in the period you discover it. The mechanics take one journal entry. The harder part is measuring it at all, which is why most ecommerce sellers carry shrinkage they have never quantified.
The definition, precisely
Shrinkage is recorded quantity minus physical quantity, multiplied by unit cost. It never touches your sales figures, because units that vanish were never sold in the first place and there is no revenue to reverse.
The word covers several distinct causes that happen to produce the same accounting outcome:
Damage. Units broken in transit, in the warehouse, or during returns processing.
Warehouse and fulfillment error. Items lost, misplaced, mislabeled or shipped to the wrong customer without a corresponding record.
Theft. External or internal.
Administrative error. Receiving 480 units against an invoice for 500 and never catching it. This is the largest category for most ecommerce sellers and the one they least expect.
Supplier shortfall. Goods paid for and never received in full.
Separating the causes matters because the remedies differ completely. Theft and receiving errors do not respond to the same controls.
What the numbers look like across retail
The most recent industry-wide benchmark comes from the National Retail Federation’s 2023 National Retail Security Survey, which reported an average shrink rate of 1.6 percent of sales for fiscal 2022, up from 1.4 percent in fiscal 2021, representing $112.1 billion in total losses.
Two caveats belong with that figure every time it is used. It measures retail broadly rather than ecommerce specifically. And the NRF discontinued the annual shrink survey in 2024, after more than thirty years, citing methodology concerns, so no more recent comparable industry figure exists. A cited 2025 or 2026 NRF shrink rate is not a real number.
Use the 1.6 percent as context rather than as a target. Your own rate is the only one that tells you anything, and an ecommerce seller with third-party fulfillment has a materially different risk profile from a physical store.
The journal entry, with real numbers
Suppose a quarterly count covers a SKU with a landed cost of $14.20 per unit. Your system says you hold 1,850 units, and the physical count returns 1,807.
The variance is 43 units, which at $14.20 landed cost comes to $610.60.
The entry debits an inventory shrinkage expense account for $610.60 and credits inventory for the same amount. Inventory on the balance sheet falls to what you actually hold, and the loss appears in the period you found it.
A note on where to put the expense. Small, routine shrinkage is commonly folded into cost of goods sold, which keeps gross margin honest. Large or unusual losses are better held in a separate expense line so they do not distort your margin trend and so the amount stays visible. Whichever you choose, be consistent, because switching treatment between periods makes your margin history incomparable.
If the count comes back higher than the records, the same entry runs in reverse. Overages are not good news. They mean the recorded quantities are unreliable in both directions, which usually points at a receiving process problem.
Why ecommerce sellers miss it
Three structural reasons.
The inventory is not where you are. With third-party fulfillment, you are trusting someone else’s count. Marketplace fulfillment centers do report losses and issue reimbursements, but reconciling those against your own records is a deliberate exercise, not something that happens by itself.
The books may not track units at all. Sellers who expense purchases when paid rather than capitalizing inventory have no recorded quantity to compare a count against. Shrinkage in that setup is not understated, it is unmeasurable, because the concept requires a book figure to vary from.
Counting feels optional. A physical count of a multi-thousand SKU catalog across several fulfillment centers is genuinely hard, so it gets deferred, and deferred counts turn a series of small correctable variances into one large annual write-off with no diagnosable cause.
Cycle counting instead of an annual shutdown
The practical alternative is counting a subset continuously rather than everything at once.
Rank SKUs by inventory value. Count the top tier monthly, the middle tier quarterly, and the long tail annually. A small share of SKUs usually carries most of the value in any catalog, so this concentrates effort where the money is while still touching everything within a year.
The real benefit is diagnostic. A variance found within thirty days can be traced to a specific shipment, a specific receiving date, or a specific process. The same variance found at year end is untraceable, and an untraceable loss cannot be prevented from recurring.
Keeping the recorded number trustworthy
Shrinkage measurement depends entirely on the book quantity being maintained accurately in the first place. That means recording receipts at actual received quantity rather than ordered quantity, recording returns as they physically arrive rather than when the refund is issued, and reconciling fulfillment center adjustments against your own records rather than accepting them.
Systems that carry real-time inventory alongside the accounting, a group that includes ConnectBooks and several comparable platforms, remove part of this burden by keeping the recorded quantity current as transactions occur. The measurement problem remains either way: software can maintain the book figure, but only a physical count produces the other side of the comparison.
A reasonable starting point
Count your ten highest value SKUs this month. Compare to the recorded quantity, value the variance at landed cost, and express it as a percentage of the inventory value you counted.
That percentage is your first real data point. If it is under half a percent, your controls are working and quarterly cycle counts are sufficient. If it is well above one percent, the value of finding out why exceeds the cost of the counting program that would tell you.
