Cost of goods sold goes wrong in five predictable places: when a purchase becomes an expense, what gets counted inside unit cost, how costs are averaged across batches, what happens to units that never sell, and whether marketplace fees have been folded in where they do not belong. A seller who fixes those five has a gross margin number worth making decisions on. A seller who has not carries a margin figure off by several points, without knowing which direction.
Each error has a specific signature in a real set of books.
1. Expensing inventory when the supplier gets paid
A purchase order for 2,000 units at $6.40 hits the books as a $12,800 expense in the month the wire went out. The units then sit on the water for five weeks and sell across the following four months.
That produces a heavy loss in March and inflated profit from April through July. None of those months is real. Inventory is an asset until it sells, and its cost belongs to the period the unit left the warehouse.
The IRS frames this as a method of accounting question rather than a matter of preference. Publication 538 states that a small business taxpayer may account for inventory by treating it as non-incidental materials and supplies, or by conforming to its treatment of inventory in an applicable financial statement, and that whatever method is chosen has to clearly reflect income. Expensing a container on the payment date rarely satisfies that, and it makes internal margin reporting unusable as a side effect.
2. Unit cost that stops at the invoice
The second error is quieter. The supplier invoice says $6.40 per unit, so $6.40 becomes the cost. Freight, duty, customs brokerage, drayage, prep and labeling, and inbound shipping to the fulfillment center all land in operating expenses instead.
Run the arithmetic on a typical container. Ocean freight and drayage of $4,100 across 2,000 units adds $2.05. Duty at 7.5 percent adds $0.48. Prep and labeling at $0.35 and inbound freight at $0.42 bring the landed figure to $9.70. That is a 52 percent increase over the invoice price, and every margin calculation built on $6.40 is wrong by that much.
The distortion is worse than a simple understatement, because those costs do not scale with the invoice. A cheap, bulky product carries proportionally more freight than an expensive compact one. A seller using invoice cost will systematically overrate their heavy products and underrate their light ones.
3. One blended average across every batch
Landed cost moves between purchase orders. Freight rates change, the supplier raises prices, a tariff line gets reclassified. Books that carry a single standing cost per SKU, updated whenever someone remembers, report margin against a number that matched reality at some point in the past.
Weighted average and FIFO both handle this, and they produce different answers during a period when costs are climbing. That difference is a real decision with real tax consequences, not a software default to accept without examining. A manual cost field, edited whenever someone remembers, fails a different way: it makes month over month margin comparison meaningless. Margin appears to improve in any month nobody updated the cost.
4. Units that never sell, treated as though they did
Every product business loses units. They arrive damaged, get returned unsellable, disappear inside a fulfillment network, get disposed of, or sit long enough that they are written down. None of that shows up in a bank feed.
A seller who reduces inventory only by units sold ends up with a balance sheet inventory figure higher than the physical count, and a COGS figure lower than actual cost. The gap accumulates. When someone finally reconciles to a physical count, the correction arrives as one ugly adjustment in a single month, and that month’s margin gets blamed on something else entirely.
Reimbursements complicate it further. A marketplace that reimburses for a lost unit generally pays a value derived from selling price, not from what the seller paid. Booking that reimbursement as revenue while the unit’s cost stays parked in inventory overstates both sides at once.
5. Marketplace fees mixed into cost of goods
The fifth error runs the other way. Some sellers push referral fees, fulfillment fees, and storage into COGS so that gross margin looks like real product economics. Others go further and record revenue net of fees, which buries the fees where nobody can see them at all.
Amazon’s published seller pricing shows why that matters. Referral fees vary by category, charged as a percentage of total price or a minimum amount, whichever is greater, with categories running as high as 45 percent for device accessories against a $0.30 floor. Media items carry an additional closing fee of $1.80 per item. Those are channel costs, and they belong on their own lines where a seller can watch them move.
Once fees disappear into cost of goods or net revenue, the two questions a seller needs to answer separately become one blurred figure: is this product cheap enough to make, and is this channel cheap enough to sell it on. This is exactly how a product can lose money for months while the sales report still looks healthy. Unit economics that hold up at the warehouse fail after a fee change nobody isolated.
Working it through on one SKU
Take a product selling at $34.99, in a 15 percent referral category, fulfilled by the marketplace.
- Selling price: $34.99
- Landed cost: $9.70
- Referral fee at 15 percent: $5.25
- Fulfillment: $5.69
- Returns and shrink provision at 4 percent of revenue: $1.40
Gross margin on landed cost alone is 72 percent, which is the number a seller quotes when asked. Contribution after channel costs is $12.95, or 37 percent. Advertising at $4.10 per unit sold takes it to $8.85, or 25 percent. All three figures are correct and they describe different things. The mistake is not choosing one; the mistake is reporting the first while making decisions that require the third.
What to fix first
Start with landed cost on the top ten SKUs by units. It is a contained afternoon of pulling freight invoices and duty entries, and it usually moves reported margin more than any other single correction.
Then make the timing right, so cost follows units out rather than cash out. Then give fees their own accounts. Order matters here, because the first two fixes change every downstream number, and redoing fee analysis afterward is wasted work.
For the recordkeeping baseline underneath all of this, the Small Business Administration publishes a plain summary of what financial records a business is expected to keep and why in its guide to managing business finances. None of it is ecommerce specific. The specialization sits on top of an ordinary requirement to know what your inventory cost and when it sold.