Back to Blog
Inventory & Landed Cost

Inventory Accuracy: Why It Drifts and How the Books Catch It

Inventory accuracy by product, units and value, with a worked count, why it drifts at 3PLs and Amazon, and the month-end check that ties the count to the books.

Oz Merchant17 min read

A stack of cardboard boxes stands in front of a mirror whose reflection shows a shorter stack with boxes missing.

The 3PL's stock file and the balance sheet are about $7,200 apart on $1.5 million of inventory: 99.5% agreement, and most CEOs would sign off on it. The numbers belong to an example dog-treat brand at the end of September, with the stock file priced at landed cost. Under that small gap sit two larger errors pointing in opposite directions: about $15,900 of stock the accounts carry that the company does not have, and about $8,800 of stock the 3PL's records show that is not on the shelf.

Inventory accuracy is how closely the stock on record matches the stock that is there, measured in units against a physical count or in dollars against the inventory balance. The lesson of this example is simple to repeat: the warehouse record, the physical count and the accounting balance can each be wrong for a different reason, and fixing the wrong one creates a second problem.

What is inventory accuracy?

The phrase is used two ways. Inventory record accuracy is the operations measure: how many products, or product locations, have a system quantity that matches the count. Inventory accuracy by value is the finance measure: what the counted stock is worth at cost against the inventory balance on the balance sheet. You need both, because each one misses what the other catches. Where accuracy sits among the other stock numbers a CEO reviews each month is covered in inventory KPIs.

The inventory accuracy formula

There are three versions, and each answers a different question.

By product (inventory record accuracy). Accuracy = products whose system quantity matches the count ÷ products counted.

By units. Accuracy = 1 − (sum of the absolute differences ÷ system units for the products counted).

By value. Accuracy = counted value at cost ÷ book value at cost. If the count comes in above the books, the result tops 100%, which is an error too; read the gap, not just the ratio.

The units version adds up absolute differences, so an over on one product does not cancel an under on another. Netting them hides the most common warehouse error. If a picker sends 40 large dog beds against orders for medium ones, the medium bed is 40 units over and the large bed 40 under; net units do not move, and two records are wrong.

The product version needs a tolerance per product, agreed before the count: a $31 dog bed should match exactly, while a 40-cent chew might miss by a few units. Keep it the same from count to count, or the trend means nothing.

How to calculate inventory accuracy: a worked count

The dog-treat brand from the opening, with illustrative figures throughout, has about $12 million of sales a year and also sells toys and beds. It stores stock at one 3PL and at Amazon. In the last week of September 2026 the 3PL counted the ten products that hold most of the cost in its building.

ProductLanded costSystem unitsCounted unitsDifferenceDifference at cost
Salmon treats, 12 oz bag$4.2018,40017,200−1,200−$5,040
Rope toy$3.109,6008,400−1,200−$3,720
Eight other products, including the medium orthopedic bed and a gift bundleVarious26,00026,0000$0
Total54,00051,600−2,400−$8,760

By product, 8 of 10 matched: 80%. By units, the absolute differences add to 2,400, and 1 − (2,400 ÷ 54,000) is 95.6%. At cost, the counted products are $8,760 short of what the 3PL's system says.

That last figure compares the count with the 3PL's record, not with the books. The comparison that reaches the P&L is against the accounts, and for these two products it gives a surprise: the accounts carry the right number of treats and toys. The reconciliation below shows why.

What counts as good inventory accuracy

Published accuracy targets vary with who measured them and how, so your own numbers are the better yardstick, read three ways.

The first is the trend: the match rate should hold or rise from count to count, and a fall means a process changed, often at the 3PL. The second is the dollars, set beside a month's gross profit. In the example below, $6,640 of the correction reaches the P&L, under 2% of a monthly gross profit of about $520,000, and the rest is a claim on a supplier. The third is repeat offenders: the same products missing month after month point to a process, such as receiving or kitting, that can be fixed.

Why inventory accuracy drifts when someone else holds the stock

A brand that ships from a 3PL and from Amazon rarely touches its own stock. Every unit moves through someone else's process, and each process has its own way of getting a record wrong. The table shows the common ones and which record goes wrong first.

CauseWhat happensWhich record goes wrong
Receiving errors at the 3PLCartons are keyed to the wrong product, or the quantity is typed from the packing list instead of countedThe 3PL's records
Supplier short shipmentsThe factory ships fewer units than the bill says, and the bill is booked as receivedThe books
Lost or damaged at AmazonUnits go missing or are damaged in a fulfillment centerThe books, until the loss is recorded
ReturnsReturned units sit unprocessed, go back on the shelf unrecorded, or come back unsellableEither, depending on who processes the return
Bundles and kitsComponents are built into a bundle and never taken out of stockThe 3PL's records
Transfers between locationsStock moving from the 3PL to Amazon is counted in both places, or in neitherWhichever report is read
Mis-picksThe wrong product ships; one record is over and another underThe 3PL's records, by product
Received, not put awayThe 3PL has the cartons but has not entered them, so they are missing from the stock file for daysThe 3PL's records
Samples and giveawaysUnits go to influencers, trade shows or replacement orders as $0 orders or by handThe books

Several of these need the accounting stated plainly.

Amazon. For stock at Amazon, its reports are the only count a brand gets. Units Amazon has received but not yet made available to buy are inbound, not sellable, and they are still your inventory; a check that reads only the sellable figure makes them look missing. Units Amazon lists as reserved, moving between its own warehouses or held for an order, are yours too, and a check that skips them makes them look lost. Brands that also keep stock in Amazon's storage program (AWD) have a third location to read. Units lost or damaged come off inventory at cost when the loss happens, or in the month it is found when the date cannot be known. Any reimbursement is booked separately, when the claim is approved or payment is probable. How claims work is in the guide to Amazon reimbursements, and how FBA stock sits on the balance sheet is in Amazon FBA accounting.

Transfers. Stock on a truck between the 3PL and Amazon is still your inventory and belongs on its own in-transit line. Counting it twice overstates the record; counting it nowhere invites a write-off of stock that is on its way.

Returns. A returned unit that will sell again goes back into inventory at cost, and cost of goods comes down by the same amount. A unit that comes back damaged is written down to what it will sell for after selling costs, when that is below cost, and one that will not sell at all is written off. Returns waiting for inspection in a returns area are still owned and belong in the count. The write-down rules are set out in inventory accounting for ecommerce.

Shrink. Units still unexplained after a recount are shrink and come off inventory in the month the count finds them. Normal shrink is usually part of cost of goods. An abnormal loss, such as a flood at the 3PL or a pallet stolen in transit, is expensed when it happens and may be shown on its own line so it does not distort gross margin; the CPA decides where.

The three numbers that should agree

The check at the center of this guide compares three figures at every month end.

  1. Systems. Units in the 3PL's stock file, Amazon's inventory reports and any office stock, at landed cost.
  2. Count. Units a count finds: this month's cycle count for the products counted, and the location reports for the rest.
  3. Books. The inventory balance on the balance sheet, which is units times cost as recorded in the accounts.

At cost, all three are in dollars and can be compared directly. When they do not agree, the pair that disagrees points at a different part of the process.

What disagreesWhat it usually meansWhere to look
Systems above the count; books agree with the countThe warehouse records are wrong and the books are notKitting, receiving keyed to the wrong product, mis-picks
Books above systems and count, which agree with each otherThe books hold units that never arrived, or that left without a cost entryBills booked at the ordered quantity, $0 orders, Amazon losses not yet booked
Count above bothUnits are on the shelf that nobody recordedReturns restocked but not processed, receipts not entered
Totals agree, products do notOvers and unders are cancellingMis-picks and look-alike products
Units agree, dollars do notA cost problem, not a count problemLanded cost not applied, freight booked to expense, a write-down booked in one place only

Direction matters as much as size. Books above the count usually mean costs the P&L has not yet taken, so profit and gross margin have been too high; part of the gap can instead be money a supplier owes or a cost that belongs in marketing. Count above the books usually means the reverse.

What a year-end-only true-up does to this check

Many brands keep cost of goods as an estimate through the year, often a flat percentage of sales, and true up inventory only after the year-end count. For them the third number is an estimate for eleven months of the year, so the comparison cannot be made. Every drift above piles up unseen and lands as one December adjustment that says nothing about which month caused it. By then the receiving report for a March short shipment is nine months old, and a supplier is far less likely to credit a claim raised that late. Recording cost of goods at landed cost as units ship, the approach explained in perpetual vs periodic inventory, is what makes a monthly check possible.

Walking one month: the dog-treat brand in September

Back to the opening figures. At September 30, before any adjustment:

NumberValue at cost
Books: inventory balance$1,498,200
Systems: 3PL file, Amazon reports and office stock, at landed cost$1,491,020
Count: the systems figure with the ten counted products at their counted units$1,482,260

Accuracy by value, the count against the books, is $1,482,260 ÷ $1,498,200, or 98.9%. The books are $15,940 above the count. The systems figure sits between the two, $7,180 below the books and $8,760 above the count, which is why the first comparison looked like 99.5%.

Step 1: systems against the count

The whole count gap is in two products, treats and rope toys, 1,200 units each. That pairing is the clue. In August the 3PL built 1,200 gift bundles, each a bag of salmon treats and a rope toy in a box, and set them up in its system as a new product. It added the bundles and never took the components out, so every treat and toy inside a box was on record twice: once as part of a bundle and once as loose stock.

The books had it right. They carry a bundle at the cost of its parts, $7.30, plus the box and any kitting fee the 3PL charges, which this example leaves out. Building one does not change how many treats and toys the company owns. So this $8,760 needs no journal entry; it needs the 3PL to correct its records. This is where fixing the wrong record does damage. Writing $8,760 off to make the accounts match the 3PL file would have left two records wrong instead of one, and next month's count would have found treats and toys the balance sheet no longer carried.

Step 2: books against the count

With the 3PL's record corrected for the bundles, the systems figure equals the count, and the books are still $15,940 above both. Books high while the stock records agree means units never arrived or left without a cost entry. Two reports show why.

The first was the receiving report for an August purchase order of 2,400 medium orthopedic beds at $31.00 delivered, the price on the supplier's bill. The 3PL received 2,100; the factory had shipped six cartons of 50 short. The bill for 2,400 had been booked as received and paid in full. That is 300 beds and $9,300 paid for and never received. It is not a loss. It is money the supplier owes, as a credit against the next order or a refund. If the supplier disputes the shortfall, any part it is unlikely to repay is a loss.

The second was a Shopify export of $0 orders since the last count. The company had sent 160 medium beds to influencers and a trade show, and 400 bags of duck jerky, landed at $4.20, as replacements for customer orders that arrived damaged. The 3PL shipped them and took them out of its records. The books took cost out of inventory only for paid orders, so none of these units left the balance sheet: 160 × $31.00 is $4,960, and 400 × $4.20 is $1,680, a total of $6,640.

$9,300 plus $6,640 is $15,940, so this month every dollar traced. Many months it will not. When the recount and the reports leave part of a gap unexplained, the remainder is shrink, written off in the month of the count. A small remainder that changes direction from month to month is normal. One larger than the tolerance set for the products involved, one that keeps the same sign, or one that sits in the same products each month is a process still broken, and it is worth looking further before the write-off becomes routine.

Step 3: the adjustment

CauseRecord to correctDirectionAmount
Bundles built without taking components outThe 3PL's fileDown: 1,200 treat bags and 1,200 rope toys$8,760
Short shipment, 300 bedsThe accountsInventory down; amount due from the supplier up$9,300
$0 orders, 160 beds and 400 bags of duck jerkyThe accountsInventory down; marketing and cost of goods up$6,640

No cause needs both records changed, and the count is not adjusted to match a record unless a recount shows it was wrong. Afterwards the 3PL's file and the accounts both show $1,482,260, the count.

Each correction belongs to the month it happened: August for the short shipment, and the ship month for each $0 order. If those months' reports have already gone to a lender or investor, the CPA decides whether to reopen them or book the correction in September, labeled with the months it belongs to.

EntryDebitCreditAmount
Short shipment, 300 bedsDue from supplierInventory$9,300
Influencer and trade show beds, 160Marketing expenseInventory$4,960
Replacement duck jerky, 400 bagsCost of goods soldInventory$1,680
Total$15,940

The P&L takes $6,640 across the months the orders shipped: $4,960 as marketing, because samples and giveaways are a cost of promotion, and $1,680 in cost of goods sold, because replacements are part of the cost of fulfilling sales. A free gift that ships with a paid order is part of what the customer bought, so its cost goes to cost of goods, not marketing. The other $9,300 stays on the balance sheet as a claim against the supplier.

The company now checks each receiving report against the supplier bill before paying, tags $0 orders by purpose so their cost leaves inventory when they ship, and has the 3PL confirm components come out whenever a bundle is built.

How to improve inventory accuracy: a routine that holds

Counting finds drift; fixing the process stops it. This routine covers both for a brand without its own warehouse.

Cycle counting by value

Cycle counting means counting a few products at a time on a rotation, instead of everything at once. Rank products by the cost they hold, not by units. One workable split is to count the products that make up most of the inventory value every month, the middle group every quarter and the long tail twice a year. Add any product that missed at the last count, and any product just bundled, relisted or moved between locations.

A full count still belongs at or near year end, because some drift only a full count finds; monthly cycle counts make it a confirmation. How it fits wider stock management is in ecommerce inventory management.

What to ask the 3PL for

  • A stock file by product and location as of the last day of each month, not the day it is pulled.
  • A receiving report for every purchase order, with units counted on arrival, in time to check it before the supplier balance is paid.
  • A log of every adjustment the 3PL makes, with a reason for each.
  • Bundle and kit builds as they happen, with the components taken out.
  • Returns received and what happened to each one: restocked, held for inspection or disposed of.
  • The results of each cycle count, and the clause that sets what the 3PL owes for units it loses. Many contracts cap it, so read the cap before assuming a loss will be paid.

What to claim from Amazon

Each month, compare the units shipped to Amazon with the units Amazon received, and check its reports for units lost or damaged in its fulfillment centers and for refunds issued on items that never came back. Each one is a possible claim. Book the inventory loss when it happens and the reimbursement when the claim is approved or payment is probable, so the two never net against each other and hide how much stock went missing.

What an auditor or a buyer will ask

An auditor or a buyer's diligence team asks for the same few things:

  • A roll-forward for each month: opening balance, plus receipts, less cost of goods, less adjustments, equals the closing balance.
  • The year-end count, its count sheets and the reconciliation from the count to the books.
  • The adjustment log, with the reason for each entry and who approved it.
  • How landed cost is built, and evidence that freight and duty reach inventory instead of expense.
  • Aged stock and any write-downs for stock that will sell for less than cost after selling costs.
  • Confirmations of units held at the 3PL and Amazon at year end, and proof of which shipments in transit the brand already owned.

A buyer reads the adjustments as evidence about gross margin. A run of small monthly corrections, each with a named cause, says the reported margin is the real one. One large unexplained true-up every December says it may not be, and that doubt tends to show up in the offer.

Inventory accuracy is a margin question

The warehouse record, the count and the balance sheet each go wrong for their own reasons. Reading which pair disagrees, and in which direction, tells you which record to fix; fixing the other one makes a second error. In this example a 99.5% match hid about $9,300 owed by a supplier and $6,640 of cost the P&L had not yet taken, found while the paperwork was weeks old.

FAQ

How do you calculate inventory accuracy?

There are three common formulas. By product, divide the products whose system quantity matches the count by the products counted. By units, divide the sum of absolute differences by system units and subtract the result from one, so overs and unders do not cancel. By value, divide counted value at cost by book value at cost. A count of 54,000 system units with 2,400 units of differences is 95.6% accurate by units.

What is a good inventory accuracy rate?

Judge it against your own history instead of a borrowed target. The product match rate should hold or rise from count to count, the same products should not miss month after month, and the dollar correction to the accounts should be small against a month's gross profit. A high total can hide large errors that cancel each other, so read the product match rate beside the dollar figure.

What is the difference between inventory accuracy and inventory record accuracy?

Inventory record accuracy usually means the operations measure: the share of products or locations whose system quantity matches a count. Inventory accuracy is often used more broadly and includes accuracy by value against the books. A 3PL can score well on record accuracy while the books are wrong, for example when a supplier short shipment was booked at the billed quantity.

See the whole platform

Reading it is one thing. Running it is another.

Book a demo for an overview of the whole platform.

Margin Matters

The ecommerce news, read for your margin

Twice a month: the changes that touch your P&L, the money behind each one, and one move to make. Three minutes, then back to work.

Twice a month. Unsubscribe anytime.