Face match issues at sign in can occur for a number of reasons. Before raising a bug or escalating to the tech team, work through the guidance below to identify the cause and where possible, resolve it directly with the customer.
The below screenshot shows what the individual sees when a selfie fails.
⚠️ You must be signed in as System Settings to access the scoring icon, which is the best starting point for any face match investigation.
How to View the Face Match Score
Go to the Address Book.
Select the individual experiencing the issue.
Open the Scoring icon.
Here you will see two sections: Spoofing and Liveness.
Spoofing
The spoofing score appears as a decimal. For example, 0.09 represents a 9% confidence score, which is a fail. Thresholds are approximately:
Good — 50–70%
Very Good — 70–90%
Look at the photo captured and work through the scenarios below:
📷 The photo is the correct person, but their appearance has changed
Their likeness may have changed since their profile photo was taken.
→ Ask the customer to have the individual update their profile picture to a more recent photo.
📷 The photo is dark or the face is obstructed
A hat, sunglasses, or poor lighting will cause face match to struggle.
→ Advise the customer on best conditions for a successful face match:
Good, even lighting
No hats, sunglasses, or obstructions
Face and shoulders clearly visible
📷 The confidence score is -1
A score of -1 indicates a system error, not a face match failure. This may be caused by:
A high volume of users signing in or out simultaneously, overloading the system
A wider system issue
→ Check whether this is isolated or affecting multiple users. If it appears to be a capacity issue, escalate to the tech team — this may require increasing capacity with Microsoft.
📷 The photo is clearly not the same person
A very low threshold score (e.g. 0.09) where the image is of a different person entirely suggests spoofing.
→ Go back to the customer, let them know that spoofing has been detected, and identify the individual responsible.
Liveness Check
⚠️ Important: Liveness check should not be enabled on individual phones. Running liveness checks across multiple devices simultaneously puts significant strain on the system. Liveness is best suited to a site tablet, where it is used by one person at a time. If a customer specifically requires liveness on individual phones, this can be enabled on a case-by-case basis — but this should be avoided where possible.
📷 The liveness score is 0
A score of 0 means the system could not determine whether the image is of a real person.
If the image appears to be a photo of a photo — the system has flagged that the sign-in attempt may not be a live person. Go back to the customer and explain this.
If it appears to be a genuine person — poor lighting can sometimes cause the AI to incorrectly flag a real person as non-live. Advise the customer on best conditions for a clear, well-lit photo.
In the example below, overhead strip lighting confused the system into flagging the individual as non-live.
📷 The liveness score is -1
A score of -1 indicates a system error, the same as in the Spoofing section above. This may be caused by:
A high volume of users signing in or out simultaneously
A wider system issue
→ Escalate to the tech team if the issue cannot be resolved.