Verify a face,
not a document
people can fake.
FaceTrust AI links a NIN or BVN registry identity to a real, live face — then reuses that same proof to identify people 1:N at a merchant desk, bank branch, or front desk, in under a second.

Name-and-number identity checks stop at a lookup.
Most identity checks in Nigeria today stop there: type in a NIN or BVN, get back a name, and trust that the person standing in front of you is that name. Numbers get shared, stolen, and typed in by someone else entirely. A registry match on its own proves the number is real — it proves nothing about who is holding it.
Name + number lookup
Proves the NIN or BVN exists in the registry. Says nothing about who is holding it right now.
Face-proof link
Every identity link is only confirmed after a live face capture matches the registry's own photo. No match, no link — no name, no number, no unlocked account.
From a document to a verified face, in three steps
NINBVNChecking registry…Validate
A NIN or BVN is checked against the national registry. Only a masked name and status come back — no photo, not yet.
Face matchLive98.7%
similarity · threshold ≥ 80%
Prove with a face
A liveness check runs first, then a live capture is compared against the registry photo — a fixed threshold decides, not a person eyeballing two photos.
Match foundChecked against 42,318 enrolled faces
✓ Liveness passed
✓ Confidence above threshold
Identify, anywhere, 1:N
At a merchant desk, one capture is compared against every enrolled face — a match, a no-match, or a flagged near-tie, in under a second.
A facial query, not a text-based one
A NIN or BVN only has to be typed in once, at enrollment. After that, a face is the query — FaceTrust AI's API looks it up against your own enrolled population and returns the linked profile, with no number to key in at the point of service.
Face-only at the point of service
Once someone has linked a NIN or BVN with a face proof, every later verification only needs a camera — no form, no number to recite or key in.
Vector-indexed Faceprints
Each enrolled Faceprint is stored in a vector index, built for high-dimensional similarity search rather than a traditional text-matching database.
Millisecond 1:N lookup
A live capture is compared against the enrolled index by cosine similarity — a 1:N lookup against your own population, not a linear scan, which is what keeps it fast at scale.
Built for institutions, not open lookup
The index is scoped to the people who have actually enrolled with your institution through a consented NIN/BVN-plus-face flow — not a general-purpose search over anyone's face.




Pricing built around your institution
Verification volume, branch count, and integration depth vary a lot between a single pilot branch and a nationwide rollout — so every plan is quoted, not shelf-priced. Talk to us and we'll size it to what you actually need.
Compare plans in detail
Pilot
One branch, prove it works
- Sandbox + production API access
- 1 branch, up to 3 integrated desks
- NIN and BVN face-linking
- Standard email support

Growth
Multiple branches, one institution
- Everything in Pilot
- Unlimited branches on one API integration
- Institution Console — KYB, staff, compliance
- Branch-scoped analytics
- Priority support

Enterprise
Full platform, dedicated support
- Everything in Growth
- Custom SLA and onboarding
- Dedicated integration engineer
- Security review and audit-log exports
- Named account manager
Built for speed, not just a login screen
These are architecture facts and targets, not marketing round-ups — FaceTrust AI is early, so we'd rather show you how it's built than invent a usage number.

The Institution Console analytics screen — the source for the outcome breakdown below.
512-d
Face embedding size
one canonical Faceprint vector per person
<1s
1:N identify target
capture to match, no-match, or near-tie
4
Registries checked
NIN, BVN, CAC, and PVC
<0.001%
Target false acceptance rate
engineering target for the current scale-up — not yet independently audited
Target latency budget per stage
Design targets for a single 1:N identify call — not a measured production average.
~440ms total, under the <1s target
Outcome breakdown
Example from the Institution Console analytics screen shown above — one branch, 12 calls.
Capture & matching
Engineered for scale, tuned for the real world
A roughly 3-second capture is enough to enroll or verify a face. The current engineering target is scaling that pipeline to 5,000,000 verification scans over a 2-month rollout, while holding the false acceptance rate below 0.001% — a target, not a figure from independent audit.
01
Capture
A ~3-second live capture, checked for liveness before anything else runs.
02
Landmark extraction
68+ geometric facial landmarks are detected and used to align and normalize the face.
03
Faceprint
The aligned capture is converted into the canonical 512-dimensional (ArcFace) Faceprint used for every match.
Built to handle real-world variation
A face rarely shows up to a camera under ideal conditions. These are the conditions the matching pipeline is engineered to account for, not edge cases it ignores.
- Illumination
- Adaptive correction for uneven or low light.
- Pose variation
- Matches across head yaw, pitch, and roll, not just face-on captures.
- Expression
- Focuses on landmarks that hold steady across smiles, frowns, and neutral faces.
- Occlusion
- Works from partial data when glasses, hair, or a mask cover part of the face.
- Aging
- Feature extraction designed to hold up as a face changes gradually over time.
- Image quality
- Flags captures too low-resolution to assess confidently, rather than matching on weak data.
- Background clutter
- Face detection tuned to separate the subject from a busy background.
- Look-alikes
- Matching tuned on the subtle details that separate very similar faces, including twins.
- Cosmetic changes
- Normalizes for makeup, hairstyle, and facial hair differences between captures.
- Data diversity
- Trained and continuously updated across demographics, lighting, and pose to limit bias.

Security
Built to be trusted with a face
Handling biometric data is a different kind of responsibility. Here's what that means in practice, not just in a policy document.
Read how we protect biometric data- Photo access is gated
- The registry photo is never handed to a client before a live face proof succeeds — validation and photo access are two separate, gated steps.
- One embedding per person
- Every face is stored as a single canonical embedding per person, not a photo album — old embeddings are archived, never silently overwritten without a trail.
- Liveness before matching
- Liveness checking runs before any face comparison, so a printed photo or a screen replay is rejected before it ever reaches the matching engine.
- Every check is logged
- Every comparison, match, and mismatch is logged — accuracy, volume, and dependency health are all visible on live dashboards, not reconstructed after the fact.
- Layered staff access
- Password, one-time codes, a fresh face check, and — for the most sensitive platform actions — hardware security keys plus a second person's approval.
- Biometrics deleted first
- Deleting an account removes the biometric record first, before anything else, so no face data outlives the consent that created it.
Early feedback
What pilot teams are telling us
FaceTrust AI is still early — these are real reactions from pilot desks, anonymized by role rather than attributed to a specific name or company we can't yet name publicly.
Verification that used to take a minute at the counter now takes less than a scan. Customers notice the difference immediately.
Branch Operations Lead
Pilot microfinance bank
An API-first platform
Every surface FaceTrust AI ships is a thin client over one documented API: identity linking, 1:N identification, profile and audit management, disputes, operator and institution administration, and analytics are all real, versioned endpoints — not screens with logic baked in behind them.
Full API reference, authentication flows, and integration guides are available for any institution or merchant integrating directly.
Read the API docsQuestions we get asked a lot
Is FaceTrust AI a replacement for NIMC or NIBSS?
No. FaceTrust AI is a verification layer that sits on top of the existing NIN and BVN registries — it validates a number against the registry, then adds the face proof step those registries don't do themselves.
What happens if my face doesn't match the registry photo?
The link or check is refused — no name, no number, no unlocked account. Near-tie cases are flagged for manual review rather than auto-approved or auto-rejected.
Do you store my photo?
No photo album. A live capture is compared against the registry photo at proof time, then converted into a single canonical face embedding for your account — the raw registry photo isn't retained by FaceTrust AI after that.
Which institutions can integrate?
Banks, microfinance institutions, and other licensed businesses can onboard through a KYB flow backed by real CAC registry validation, then manage branches, desks, and staff from the Institution Console.
How fast is a 1:N identify check?
Under a second end-to-end is the design target, from camera capture to a match, no-match, or flagged near-tie — see the latency breakdown above.
Bring FaceTrust AI to your business
Whether you're a bank, a microfinance institution, a hotel, or a retail chain looking to replace card-and-PIN checkout with a face, we'd like to hear what you're building.
Request enterprise / institution access
For banks, MFBs, hotels, and retail chains ready to pilot FaceTrust AI at a branch or network level.