Per-SKU Profit Across Marketplaces: Our Currency and Blend Errors

✨ Written with AI assistance from our own settlement data and our own code history, and reviewed by Avanta Global EOOD before publication. Learn more

Short answer: per-SKU profit across marketplaces goes wrong in three places before any fee is misread: the exchange rate, the blend and the owner of the row. We made all three mistakes in our own software. An exchange rate was hard-coded with no date and had drifted 8%. Our unit economics were measured per account, not per SKU. And our order table held 2,621 Amazon Australia orders, of which only 453 were ours.

This is one seller's account, measured line by line, plus the bugs we found in the code that reads it.

What we can actually measure

Per-SKU profit per marketplace needs settled sales in that marketplace. Here is what our own account holds, from settlement reports up to 28 September 2026:

Marketplace	SKU	Settlement currency	What the settlements contain	Can we measure profit?
Amazon.com.au	bubble mailer bundle	AUD	396 orders, 543 units, 27 May – 28 Sep	Yes
Amazon.com.au	kraft mailer bundle (our second product, not yet sold)	AUD	0 lines	No, Amazon's estimate only
Amazon North America	—	CAD	one settlement, 18 Feb – 24 Jun, C$ −14.87, no order lines	No sales
Amazon.com (US)	—	—	no orders; the store was closed for inactivity on 3 Sep 2026	No
Amazon EU	—	—	no orders	No

So for us, honest per-SKU profit across marketplaces is one SKU in one marketplace. We found the mistakes below while preparing for the second SKU, which is when they would have started to cost money.

Mistake 1: an exchange rate with no date

Until 19 September 2026, our margin calculator converted currencies from seven rates written into the code in May 2026, under a comment promising they were "stale-tolerant in the 5-10% range".

On 19 September the table's AUD rate was 1.52 per US dollar; the live rate was 1.40. A US-dollar cost converted into Australian dollars came out about 8% too high, outside the tolerance its own comment promised.

The currencies not in the table were worse. The lookup was .get(currency, 1.00): any currency it did not know was converted at parity. Nine currencies sat in that state (INR, BRL, SEK, PLN, EGP, TRY, SAR, AED, SGD). With USD→INR at 95.99 that morning, 10 US dollars became "10 rupees", off about 96 times, with no error raised: it looks like a conversion.

The real defect was not the rate. A number in code has no date, so nothing could say it had gone stale. Rates now carry an as_of date and a source and refresh daily; a missing rate refuses to convert, and an old rate says it is old.

Scope: this was our sourcing and margin-estimate code. Our own listing's profit never needed a conversion, because its settlement and product cost are both in AUD.

Mistake 2: one blend for two SKUs

Our unit-economics code filtered by account and marketplace, not SKU. With one SKU that gives the per-SKU answer, so for months nothing looked wrong. We found it on 13 September, setting up the second SKU. One column below is measured; the other is declared, because that product has not sold yet:

Per unit, AUD	Bubble mailer bundle (measured, 16 Jul – 28 Sep, 439 units)	Kraft mailer bundle (not yet sold)
Item price after promotions	32.68	declared list price
Referral fee	4.25	declared, gross of a temporary fee promotion
FBA fulfilment fee	10.85	Amazon estimate + 10% GST
Product cost	7.40	supplier invoice
Contribution before ads, refunds and storage	10.17	more than twice as high (declared)

Any single number made from these columns describes neither product, and looks measured while half of it is an estimate. Estimates also need translating before they can sit next to a settlement. For the bubble mailer bundle, Amazon's FBA fee estimate on 30 September was A$9.86, and the settlement has charged A$10.85 on every unit since 27 May. The gap is exactly Australia's 10% GST: the estimate excludes it, the charge includes it (9.86 × 1.10 = 10.85). Put the raw estimate next to measured fees and you overstate a new SKU's margin by about a dollar a unit. That is why the kraft column adds GST to its estimate. We cannot check that figure until the kraft bundle sells.

We use the gross referral fee for the kraft bundle, because a break-even built on a temporary fee promotion becomes a loss the day it ends. And on 30 September the kraft bundle's product cost was still recorded under a retired SKU name, so a lookup by the live SKU finds nothing.

Economics are now computed per SKU. An unsold SKU is labelled declared, and our ad automation may only reduce spend on it until it has measured contribution.

Mistake 3: rows that are not yours

Our order table holds orders for every account connected to AgentXray, including public demo accounts. Without the account filter, Amazon.com.au shows 2,621 orders; ours are 453. The table's 3,185 US orders all belong to demo accounts: counted unfiltered, they look like a US business we do not have.

Subtler: our only Canadian settlement is stored under the US marketplace ID, because our North America account's sync requested it with that ID. Its currency header says CAD. Group by marketplace ID and Canadian dollars get added to US dollars; group by settlement currency and they land correctly.

Two traps inside a single marketplace

Principal appears twice. In the settlement, Principal is the description on both the item-price line and the (negative) promotion line. On our SKU, filtering on the description alone counts 700 units at an average of A$25.23. Filtering on amount type and description gives the true 543 units at A$33.10. The details are in how to read an Amazon settlement report.

Storage has no SKU. None of our 7 storage-fee lines (A$750.26, 6 June – 6 September) carries a SKU. With two products it must be allocated, by cubic volume rather than units, because a bulky SKU uses more space.

How to compute per-SKU profit per marketplace

1. Filter by account first. If your tool or spreadsheet holds more than one seller's data, check the account before you check anything else.
2. Group by settlement currency, not by marketplace label. The currency on the settlement is the currency Amazon paid you in.
3. Do the whole calculation in that currency. Price, fees and product cost, per SKU. Record product cost in the same currency, or convert it once, with a dated rate you write down beside it.
4. Convert only the final profit if you need one number across marketplaces, and state the rate and its date. A rate with no date cannot be checked.
5. Identify a line by amount type and description together. Never by description alone.
6. Split by date where fees changed. Our referral fee went from 5–6% to 13% in July (what that did); an average across the change describes neither period.
7. Allocate shared costs (storage, subscription) explicitly, and say how.
8. Label unsold SKUs as estimates, and never blend them with measured ones.

To try this on one report first, the settlement profit reader runs in your browser. The break-even ACoS calculator turns a per-SKU contribution into an ad ceiling, and ACoS vs profit explains why the ceiling moves with price.

FAQ

Should I compare marketplaces in one currency? Only at the end. Compute each SKU's profit in the settlement currency, then convert the result with a rate you date. Converting each fee line separately multiplies the chance of a mixed or stale rate.

Is Amazon's fee estimate good enough for a new SKU? It is a starting point, not a measurement. On Amazon Australia the estimate excludes GST while the settlement charges it, so on our live SKU the estimate read A$0.99 below the fee actually charged. Add the tax before you compare, and replace the estimate with settled fees as soon as the first orders post.

Can I blend SKUs if they are in the same category? Only if price, size and cost are close enough that the per-SKU numbers barely differ, and you can only know that by computing them per SKU first. On the declared numbers, the contribution of our two mailer bundles differs by more than a factor of two.

About this article

Figures come from Avanta Global EOOD's own Amazon settlement line items (Amazon.com.au, 27 May – 28 September 2026; one Amazon North America settlement, February – June 2026), Amazon's fee estimates for our two Australian listings (read 26 and 30 September 2026), our order table, and our own code history for the exchange-rate change (19 September 2026). Measured on 30 September 2026. Currency is Australian dollars unless stated. The kraft mailer bundle figures are declared, not measured: it has no settled sales yet. We are not accountants; check your own settlement before acting on a number.