It’s not magic: it’s systematic checking
When volumes grow, going through a VAT Transaction Report of tens of thousands of transactions row by row becomes impossible. Anomaly detection is a set of techniques — not necessarily “strong AI” — that help you:
- spot unexpected patterns over time;
- highlight out-of-scale values (outliers);
- flag sales/VAT distributions inconsistent with the past.
It doesn’t replace your tax advisor, but it helps them focus on the truly critical points instead of manually reviewing thousands of rows.
Types of anomalies: reference table
In the context of VAT for marketplaces and OSS, the most useful anomalies to look for fall into these categories:
| Type | Concrete example | Severity | Typical cause |
|---|---|---|---|
| Off-standard rate | 0% VAT on a physical good in Germany | 🔴 High | Incorrect Amazon configuration or wrong product category |
| Country disappeared | France absent in Q3, present in Q1 and Q2 | 🔴 High | Incomplete VAT Report file or marketplace not included in the export |
| Unexpected country | Sales in Romania not expected in the model | 🟡 Medium | FBA warehouse automatically activated by Amazon |
| FC_TRANSFER spike | €50,000 of stock in one day to Poland | 🟡 Medium | Real FBA movement but not covered by local VAT |
| Currency anomaly | GBP amounts treated as EUR | 🔴 High | CSV parser error or marketplace setting |
| Concentrated credit notes | 80% of refunds in the last week of the quarter | 🟡 Medium | Seasonal returns campaign or product quality issue |
| Quarter-on-quarter change > 50% | From €10,000 to €1,500 in one quarter | 🟡 Medium | Paused campaigns, stockout, or missing file |
| Calculated VAT of 0 on positive taxable amount | 200 transactions with taxable amount > 0 and VAT = 0 | 🔴 High | OSS threshold not configured or aggregation bug |
A numerical example: how an anomaly shows up
Imagine you have this extract from the Q2 VAT Transaction Report:
| Country | Taxable amount | VAT declared | Implied rate |
|---|---|---|---|
| Germany (DE) | €12,400 | €2,356 | 19% ✅ |
| France (FR) | €8,100 | €1,701 | 21% ⚠️ |
| Poland (PL) | €3,200 | €0 | 0% ❌ |
| Netherlands (NL) | — | — | absent ⚠️ |
Three immediate signals:
- France 21% — France’s standard VAT rate is 20%, not 21%. Possibly a rounding error or misconfiguration.
- Poland 0% on €3,200 of taxable amount — if you’re selling physical goods in Poland, this is a critical anomaly to check before filing the OSS return.
- Netherlands absent — if you had NL sales in Q1, their total disappearance in Q2 suggests a missing file or a marketplace not included in the export.
Without anomaly detection, these three problems would go unnoticed until filing time — or worse, until a tax audit.
How anomaly detection is applied to OSS data
A typical workflow in VATManager:
- You import data from Amazon, eBay, Shopify into the system;
- The system generates the aggregated table by country/rate;
- It automatically compares it against previous quarters;
- It flags deviations beyond configurable thresholds (e.g. a change > 40% compared to the previous quarter);
- It shows the list of specific transactions causing the anomaly, for a targeted review.
The advantage is catching errors before filing the OSS return, when fixing them is still simple.
What to do when you find an anomaly
The response depends on the type and severity:
Rate anomaly (0% on a physical good)
- Check the product category on Amazon Seller Central;
- Check whether the product was mistakenly classified as exempt;
- If the error is confirmed, fix the configuration and recalculate VAT for the affected transactions before filing the return.
Country missing compared to the previous quarter
- Re-download the VAT Transaction Report for the period from Amazon Seller Central;
- Check that all active marketplaces are included in the export (Amazon.de, Amazon.fr, etc.);
- If the country really was inactive that quarter, document the reason.
Unexpected FC_TRANSFER spike
- Check Amazon’s FBA panel to see whether there was an actual stock movement;
- Check whether that country has a local VAT registration threshold you might have exceeded;
- Consult your accountant if the volume exceeds local materiality thresholds.
Anomaly discovered after the return has already been filed
- For minor errors: you can wait for the next quarter and add an adjustment line;
- For significant errors: consider voluntary disclosure via the Agenzia delle Entrate’s OSS portal, with your accountant’s support.
📎 Learn more: Amazon VAT Transaction Report: how to read it without mistakes — Common mistakes in Amazon sellers’ VAT management — How to read a per-country VAT report
Frequently asked questions
Does anomaly detection replace my accountant’s review? No. It’s a pre-check tool that identifies the critical points worth focusing on. The final judgement on what to do remains with your accountant or tax advisor.
What are the most frequent anomalies in Amazon VAT reports? The most common are: a 0% rate on physical goods (possible configuration error), a country missing compared to previous quarters (possibly an incomplete file), and spikes in FC_TRANSFER not justified by actual logistics.
What do I do if I find an anomaly after the return has already been filed? It depends on the scale. For minor errors you can wait for the next quarter and correct it with an adjustment line. For significant errors, consider voluntary disclosure via the OSS portal with your accountant’s support.