This feature is currently in beta for customers with access to the 'AI-Powered Accounting Add-On'.
The outlier detection functionality highlights invoices that look unusual compared to your organisation’s typical spending patterns. It’s meant to help reviewers spot invoices that may contain mistakes, unexpected charges, or unusual details that deserve a second look.
What counts as “unusual”?
A flagged invoice does not mean the invoice is wrong. It means: “This looks unusual, please take a closer look.”
An invoice can be flagged when parts of it don’t match what’s normal for your organisation, for example:
The amount is higher than typical
The currency is unusual for a supplier
The expense account, VAT code, cost center, or cost carrier differs from what’s usually used in similar cases
What data is being used for outlier detection
Outlier detection is organisation-specific. It learns what “normal” looks like based on your own invoice history (not other companies’ data).
It uses:
Time window: invoices from the last 12 months
Included invoices: invoices that were not rejected or deleted
What it compares:
Amount
Key invoice/accounting fields like currency, supplier, expense account, VAT code, cost center, and cost carrier
Outlier detection in the ‘Review’ step
The feature only flags invoices when it detects something unusual compared with your organisation’s spending patterns. It only appears during the ‘Review’ step and can only be seen by the reviewer of the invoice.
If an unusual invoice got detected, a note will be displayed under the invoice detail that has caused the outlier detection.
A couple of important implications:
No flag doesn’t mean “everything is normal.” It simply means the system didn’t detect anything unusual.
Because this is still in early beta, it may not catch every unusual invoice yet.
Suggestion-only: no blocking
Outlier detection is advisory. It does not block invoices, prevent approvals, or enforce changes. It’s simply a signal to support human review.
Beta and improving over time
This is an early version of the feature released to gather real-world usage data. We will continuously adjust the scoring and sensitivity (calibration) as we learn which flagged invoices were truly unusual in practice, so the alerts become more accurate and better aligned with your organisation over time.