Skip to content

5 sales data errors to check before you trust the numbers

Sales data checks: mismatched SKUs, duplicate rows, unknown order statuses, missing cost of goods, provisional fees. Catch them in Excel or automatically.

By The DataxLand team· · 6 min read

A wrong profit figure is rarely caused by a wrong formula. Most of the time the input data has an error nobody noticed: a fee line exported twice, a new SKU missing from the cost table, an order status the marketplace just added. The formula still runs, the cell still shows a number, and the number looks plausible enough that nobody questions it.

This post lists the five most common sales data errors, each with its symptom, how to check for it yourself in Excel, and how serious it is: which ones only need a note, and which ones should stop the report.

Examples come from the DataxLand sample data: a simulated scented-candle and essential-oil shop selling on Shopee, TikTok Shop and its own website, 1 Aug – 29 Oct 2026. The sample set has a few typical errors planted in it so the checks can catch them.

The five errors at a glance

ErrorSymptomImpactWhat to do
1. SKU not in the catalogAn order line has a product code that is not in the product tableRevenue is complete, but cost of goods and profit by product are missingWarn, fix at the source
2. Duplicate rowsA fee line or order line appears twiceFees or revenue counted twiceWarn, deduplicate by rule
3. Unknown order statusA status that has not been mapped (for example "unpaid")Unclear whether the order was sold or cancelledBlock the report until it is mapped
4. Missing cost of goodsA SKU has no cost on the sale dateProfit inflated because cost is zeroShow the cost coverage rate
5. Provisional fees, data not yet inRecent orders have no settled fees yet; a source has not refreshedProfit for the latest week is not stableState the "data as of" date

Error 1: SKU in an order that is not in the catalog

Symptom. The shop adds a new product on the marketplace, or changes a variant code, but the product and cost tables are not updated. The order line has an "orphan" SKU. The sample data has 3 order lines like this.

Why it is dangerous. Total revenue does not change, so nobody notices. But when you calculate profit by product, these lines either drop out of the report or are counted with a cost of zero.

Check in Excel (sheet orders has the SKU in column C, sheet products has the SKU in column A):

=IF(ISNA(MATCH(C2, products!A:A, 0)), "NOT IN CATALOG", "")

Filter the result column and send the list of SKUs to whoever manages products so they can add them.

Error 2: duplicate rows

Symptom. A fee settlement file is exported twice and pasted on top, or a source resends a record. The sample data has 1 fee line duplicated.

Why it is dangerous. A duplicated fee line lowers profit by exactly that fee; a duplicated order line raises revenue. In a file with thousands of rows, the human eye cannot catch it.

Check in Excel. Identify the row's "key" — the combination of columns that two valid rows never share, for example order ID + fee type:

=COUNTIFS(A:A, A2, B:B, B2)

A result above 1 means a duplicate key. Do not delete right away: check whether it is a true duplicate or two valid fees of the same type.

Error 3: order status not mapped

Symptom. Each marketplace has its own set of statuses, and occasionally adds new ones. A report needs to map every status into a few standard groups: completed, cancelled, returned, unpaid. A status missing from the mapping table means an order that "belongs to no group".

Why it is dangerous. This is an error that should block the report, not just warn: if you do not know whether an order was sold or cancelled, every order, revenue and cancellation-rate metric is untrustworthy. In the sample data, if you remove the "unpaid" status from the mapping table, the status check fails on exactly one row and the build is blocked.

Check in Excel. Keep a status_map sheet (raw status → standard group) and flag rows that cannot be looked up:

=IFERROR(VLOOKUP(E2, status_map!A:B, 2, FALSE), "NOT MAPPED")

Error 4: missing cost of goods on the sale date

Symptom. The cost table has no row for the SKU, or has one whose effective date starts after the sale date. The result is an order line with revenue but no cost of goods.

Why it is dangerous. Profit is inflated by exactly the missing cost. The best fix is not to hide the number but to show it with a health metric: cost coverage of revenue. In the sample data this is 92.35%: nearly 8% of revenue is counted as if cost were zero, so real profit is lower than the figure shown.

Check. Divide total revenue of lines that have a cost by total revenue. Below roughly 95%, fill in costs before using profit by product to make decisions (this threshold is a practical suggestion; set your own according to what you can accept).

Error 5: provisional fees and data not yet in

Symptom. Marketplace fees are only final once the order completes and is settled. Orders from the most recent week usually still have provisional fees; an ad source may not have refreshed up to yesterday.

Why it is dangerous. The latest week's profit looks very different from the week before, and readers draw the wrong conclusion. In the sample data, 12.36% of total marketplace, payment and shipping fees are still provisional on the last data date (29 Oct 2026).

Check. Always write "data as of …" next to every report, and keep provisional fees separate from settled fees in the file. See Provisional fees and settled fees.

The order to check every time you build a report

  1. Up to what date the data has arrived, and which sources have not refreshed.
  2. Are there any unmapped order statuses — if so, stop.
  3. Are there duplicate keys?
  4. Are there SKUs not in the catalog?
  5. What are cost coverage and the provisional fee rate — write both numbers into the report.

How DataxLand does this

Every build, DataxLand runs a set of data checks: unique keys, no empty values, relationships between tables (a SKU in an order must exist in the catalog), valid values. Each check has a level of Blocking or Warning. A warning-level failure still lets the dashboard publish and records the failing rows so you can fix them at the source; a blocking failure keeps the previous published version and does not put wrong numbers out. Before publishing, DataxLand also reconciles a few key totals between the loaded data and the built warehouse. See Verified numbers.

The Data health page lists every check as Passed, Failed or Couldn't run. On a dashboard, a tile is labelled Incomplete when data coverage is below 100%, and the Data as of line shows the latest date with numbers. When you ask Si, the answer includes a Sources section with the period and data date. See also verified numbers on the platform page and return control.

Frequently asked questions

Do data errors stop the dashboard from working?

Only blocking-level errors hold back a new publish; the dashboard keeps showing the previous published version. Warning-level errors block nothing; they are recorded with the source rows so you can fix them.

Should I delete duplicate rows in the source file myself?

Fix them at the source or in a rule-based cleaning step, not by deleting by hand in the original file. Keeping the original file untouched means you can always explain where last week's number came from.

What cost coverage rate is enough?

There is no universal number. For shop-level decisions, an error of a few percent may be acceptable; for SKU-level decisions (discontinuing, raising prices), you need the cost for that exact SKU.

Can data checks replace an audit?

No. Data checks catch technical errors (duplicates, gaps, mismatches). They do not replace the reconciliation of documents and invoices that accountants and auditors do.

Read next: Cancellation rate: standardizing order statuses · How to calculate true profit on Shopee, TikTok Shop and Lazada · When Excel is no longer enough

Related posts

Verified dashboards from your own data, this afternoon

Free plan with no time limit · No card required · Sample workspace ready in minutes