Success Rates, Bias & Diversity

Learn how user success is measured, and how FaceTec achieves cutting-edge performance even in the most diverse populations.

First-Time Mobile Device User Success Rates

Note: The FaceTec SDK must have Low Light Mode enabled and be running the latest version to achieve the below levels.

~85-90% - Prove Liveness on their very first Session.

~5%-10% - Perform an unsuccessful Session on their first try, are shown retry guidance, are able to collect additional data on their second session, and then succeed on their second session.

~3% - Perform another unsuccessful User Session, see more guidance, collect additional data, and then succeed in sessions 3-5.

The remaining 1-2% of the sessions consist of spoof attacks, sessions from users who refuse to improve their capture environment, sessions from users with damaged cameras, users who wear darkened glasses and do not take them off, users who are in challenging lighting and refuse to improve it, users who quit the process, etc.

With millions of real-world sessions performed, we have not detected any significant differences in performance when using low-end mobile devices.

Please Note:  Across Billions of 3D Liveness Checks Annually, in addition to the above typical aggregate User Metrics, FaceTec Customers and Partners consistently see 95-98% of Users Successfully Prove Liveness Within 3 Scans, and ~99% of Users Successfully Prove Liveness Within 5 Scans.

Your App may achieve different results if it is not running the Latest Versions of FaceTec SDKs, has incorrect String Customizations, has an unusually high amount of Spoof Attacks, or where Users are not sufficiently motivated to complete the process (and quit the process early).

First-Time Web Browser Success Rates

Users must be running the Latest SDKs, and have a compatible Web Browser and a webcam that can be secured with video feed encryption.  These requirements, along with some incompatible face angles from USB webcams, can cause a lower 1st-try success rate. However, real users still succeed during retries and end up with 98-99% 1st-time user success, just like on mobile devices. If real users have any behaviors with their hardware, they are prompted to fix them with suggestions directly in the FaceTec UI, and also on www.LivenessCheckHelp.com.

~85-90% - Prove Liveness on their very first Session.

~5%-10% - Perform an unsuccessful Session on their first try, are shown retry guidance, are able to collect additional data on their second session, and then succeed on their second session.

~3% - Perform another unsuccessful User Session, see more guidance, collect additional data, and then succeed in sessions 3-5.

The remaining 1-2% of the sessions consist of spoof attacks, sessions from users who refuse to improve their capture environment, sessions from users with damaged cameras, users who wear darkened glasses and do not take them off, users who are in challenging lighting and refuse to improve it, users who quit the process, etc.

With millions of real-world sessions performed, we have not detected any significant differences in performance when using low-end mobile devices.

Please Note:  Across Billions of 3D Liveness Checks Annually, in addition to the above typical aggregate User Metrics, FaceTec Customers and Partners consistently see 96-98% of Users Successfully Prove Liveness Within 3 Scans, and ~99% of Users Successfully prove 3D Liveness within 5 Scans.

Your App may achieve different results if it is not running the Latest Versions of FaceTec SDKs, has incorrect String Customizations, has an unusually high amount of Spoof Attacks, or if Users are not sufficiently motivated to complete the process.

What Does "No Observable Bias" Mean?

FaceTec 3D Face Matching and 3D Liveness Algorithms are exhaustively tested against sets in the many billions of matches. No Observable Bias means that no identifiable errors or patterns were detected in our trained evaluators for the misidentified individuals in all the tests performed.

While all systems that use visible light to capture biometric data will have some bias, at the thresholds we have published in our 3D Face Matching FAR and Bias in 2D Face Matching vs. 3D Face Matching white papers, the bias level for age, gender, skin-tone, device cost, country-of-origin, and eyeglass-wearer combinations are not apparent to our evaluators, and we consider them to be unobservable. While developing and improving our algorithms, FaceTec assesses the results from many billions of match pairs. However, because the error rate is so low, the number of misidentifications is also very low, and our trained evaluators manually review 100% of the misidentifications.

Usability with Light Skin Tones

Camera sensors need light to function, but if too much light is reflected back to the sensor, the camera will not work well. As a result of this, people with lighter skin tones and too much light present in their environment may appear "blown out" to the camera. This can occur in direct sunlight or other extremely bright environments. However, making a 90-degree turn will usually change the directional lighting enough for the sensor to function properly. With a camera sensor that is not over-saturated, FaceTec has no observable difference in performance over the spectrum of lightest to darkest skin tones.

Usability with Dark Skin Tones

Camera sensors need light to function, but if there isn't enough light reflected back to the sensor, the camera will not be able to collect digital data. As a result of this, people with darker skin tones must have more light present in their environment than those with lighter skin tones. However, since FaceTec uses its 3D Face capture technology, less light is required than 2D systems with the same camera, and the necessary light can be provided by the screen of the device the user is interacting with. In its large-scale testing, FaceTec has no observable performance difference over the spectrum of human skin tones.

Matching Identical Twins

In real life, identical twins and people who look similar or have submitted "Morphed" photos can use each other's passports/IDs. FaceTec SDKs can detect and reject "morphed" photos, and the 1:N can flag people trying to use multiple passports/IDs under different names. Over time, Identical twins are less identical as they get older, but can be very similar when young. For identical twins, FaceTec recommends using our 3D:3D APIs for increased accuracy when detecting identical twins and flagging them accordingly.

Maintaining a Real-World Perspective During Your Due Diligence

Typically, the first time anyone uses FaceTec, "it just works". It's easy, fast, and intuitive, and once the check is complete, the users move on with their lives. However, Developers immediately enter into "test mode" and start turning off lights, pointing the camera at their face in direct sunlight, trying different angles, blocking parts of their face, etc. This is interesting research, but it is not an accurate real-world use-case simulation.

Developers who are tasked with usability testing should push the limits of FaceTec capture UI, but they should not let those experiences incorrectly shape their overall perspective. Just as users would not try to type their complex passwords in a bumpy subway car, users would not try to scan their faces during a laser light show in a nightclub.

Real users just need FaceTec SDKs to work under their everyday conditions, so they can go through the verification process and continue with their lives. Almost everyone agrees that FaceTec SDKs are super easy to use and present much less friction than passwords, so please do not let extreme usability testing incorrectly skew your perspective. There are hundreds of millions of first-time FaceTec users who have successfully opened new accounts and can now securely access them daily without any unexpected behaviors. They never ever want to go back to a bank branch or have to enter a password.

Behaviors Related to Performing Liveness Checks on Testing Devices

Users may need two to five attempts to prove 3D Liveness if their environment is not typical. As these cases are often the result of difficult/uneven lighting, Users should make changes to their environment until they are directly facing the camera under normal lighting conditions.

If Users still cannot pass the 3D Liveness Check after adjusting their environment, then this may indicate a compatibility-related behavior. In these cases, we recommend Users to check their Device and Browser's Compatibility with the FaceTec Software.

In the rare case of it not being a Compatibility or Camera Environment Behaviors, we encourage Users to seek assistance from our Support Team by reproducing the behavior on the FaceTec Demo App and sending a report.

"I am Performing a Session in a Challenging Environment Repeatedly, and FaceTec is not working."

FaceTec SDKs are designed to work in the real world. FaceTec SDK on-screen instructions guide new users to perform successful sessions that prove their 3D Liveness. FaceTec SDKs achieve 97-99% real-world first-time user success because even if the user has to retry, they will likely pass seconds later on a subsequent attempt after seeing the Retry Screen and correcting their environment.

Real end users *want* FaceTec SDKs to work on themselves, and will follow on-screen help instructions.

Unlike this test case, Real end users DO NOT seek challenging environments and repeatedly try User Sessions from those environments without changing something to better their odds. FaceTec works in a wide variety of challenging environments, but if you are seeing the Retry Screen, you need to check out your photo next to the Good Example Photo. If your photo shows glare or data loss, you need to fix it, like a real user would.

Accessibility with Atypical Facial Conditions

FaceTec makes every effort to ensure its 3D Liveness Detection and Face Verification Software works perfectly for as many members of the global population as possible. Even though there are no observable biases in FaceTec's algorithms across the spectrum of typical human skin tones, because the AI models the user's face in 3D, and is trained on a large, global dataset (180+ countries, and millions of faces), the AI can still encounter challenges making 3D Liveness Decisions on atypical human faces. Atypical traits that may influence the AI include: objects that obfuscate the face (such as eye patches or prosthetics), birth conditions, partial or whole facial paralysis, scarring, incidents that have left a lasting effect on the facial structure, and skin conditions. Users who are wearing heavy makeup (thick foundation, concealer, eyeshadow, eyeliner, fake or drawn-in eyeliner, etc.) that cover or obscure the traits of human skin and features are also challenging conditions for Liveness AI, which is designed to determine if a Real 3D User is performing a scan and is not using a full or partial face mask. Our AI training sets do include some of these conditions; however, with unlimited combinations of these atypical traits, these are cases where our AI is unable to make a confident Liveness Decision due to these variables. Users with atypical facial features may experience a higher False Reject Rate and potentially the inability to successfully prove 3D Liveness with the FaceTec SDKs.

We recommend that our Customers provide a support channel with a manual review process specifically for users with atypical facial features and that human agents manually review the 2D audit trail images to make an informed decision about whether or not to allow the individual to proceed to their desired digital destination if, after numerous attempts, the FaceTec AI is unable to make a positive 3D Liveness determination.

Maintaining a Real-World Perspective In Your Production Integration

FaceTec may not provide a successful 3D Liveness for 100% of your Users because there will likely be a small percentage of Users using Your Application that are in one of the following Groups:

1. Fraudsters
2. Subset of Real Users who do not perform "legitimate" attempts -- i.e. performing "lazy" sessions, or otherwise not following guidance.
3. Subset of Real Users who perform legitimate attempts, but do not try "enough" -- Users must be allowed to try at least 5 times, and sometimes even more, to prove adequate data to Prove 3D Liveness (Note:  95-97% of Users Succeed within 3 tries, and ~99% within 5 Tries).
4. Subset of Real Users who have Atypical Facial Conditions in which the FaceTec AI cannot make a confident enough decision.
5. Subset of Real Users on incompatible devices -- this includes devices with insecure runtime environments or that have camera compatibility-related behaviors.

Important:  ~95-97% of FaceTec software users successfully prove 3D Liveness within 3 tries, and ~99% of Users successfully prove 3D Liveness within 5 tries.  These Success Rates include ALL User Groups, including the Groups listed above.

Investigating 3D FaceScan, 3D FaceMap, IDScan, or 2D Image Behavior

FaceTec may investigate specific User cases in an effort to improve the FaceTec SDKs. 

However, FaceTec does not perform analysis of Scans or Images "as a service", nor can FaceTec provide detailed reports about each and every Scan that is sent to FaceTec for analysis.

Since FaceTec is primarily a Security Product that relies heavily on Computer Vision, AI, Cryptography, and various methods of Attack Prevention, FaceTec cannot provide specific details about low-level computations and decisions.

Accessibility for Blind & Visually Impaired Users

FaceTec's iOS, Android, & Browser SDKs are designed to be accessible for blind and visually impaired users, both with and without the use of a Screen Reader.

Users who are severely visually impaired, either partially or fully blind, have multiple options available for successfully completing a 3D FaceScan:

1. Using Blind User Audible Assistance with a Screen Reader.
2. Using Vocal Guidance without a Screen Reader.
3. Using another person to hold the device for them, perform the button presses, and guide them through the 3D FaceScan capture process. The only requirement is that the visually impaired user must have their eyes open.

Blind User Audible Assistance is a feature that helps visually impaired users complete a 3D FaceScan while using a Screen Reader. The FaceTec SDK will automatically detect when the OS is using a Screen Reader and turn on the Blind User Audible Assistance feature for the user. The Blind User Audible Assistance feature provides the user with customizable, advanced face-alignment feedback and interface prompts to help guide the user in performing a successful 3D FaceScan. Please see the Blind User Audible Assistance info here.

To integrate Blind User Audible Assistance into Your App, please see the Blind User Audible Assistance (BUA) Integration Guide.

Vocal Guidance is a feature that helps visually impaired Users complete a 3D FaceScan without the use of a Screen Reader. When Vocal Guidance is enabled, audible sound files are played to give directions to the user at specific points during interaction or when trouble is perceived in proceeding through the 3D FaceScan. Please see the Vocal Guidance info here.

Note: When a visually impaired User is performing a 3D FaceScan without Blind User Audible Assistance (using Vocal Guidance or another person), the User is required to keep their eyes open to complete the Session. For Blind User Audible Assistance Mode, a User is allowed to complete the Session with closed/obstructed eyes and have their Session reviewed further.