PollGuard Verify

Confirm every voter is real, eligible and unique.

A layered identity check built for African identity documents — and reviewed by your own administrators before anyone is approved.

PollGuard Verify — member onboarding
Personal details
Scan ID document
Selfie with ID
4Liveness check
5Fingerprint identity
6Face enrollment
7Review & submit
face detectedturn head slowly →36 / 48 tickskeep eyes on screen

The layers

Seven checks. One decision — made by a person.

Each layer produces a score your administrators can see. No member is approved or rejected by the model alone, and every decision is attributable.

L1

Document capture & authenticity

Live capture with glare, moiré and screen-replay detection; edge, hologram and microprint forensics on the document itself.

auth score
L2

OCR & MRZ cross-validation

Fields read from the front are checked digit-by-digit against the machine-readable zone — Kenyan ID, Maisha Namba, passports, and regional documents.

field match
L3

Selfie ↔ document face match

Deep-embedding comparison between the live selfie and the document portrait. Thresholds published to your admins, not hidden.

cosine ≥ 0.45
L4

Liveness

A continuous head-sweep challenge plus a spoken-word check — a photo, a video or a screen can’t pass it.

live score
L5

AI second opinion

A vision-model review of document, selfie and liveness evidence produces an explained verdict for the reviewer — advice, not a decision.

explained verdict
L6

One person, once

Face embeddings are searched against the entire register so the same person cannot enrol twice under different names.

dedup index
L7

Human review

Everything lands in your administrators’ review queue with the full evidence trail. Approval is a named, logged action.

attributable

Your administrators decide

The review queue — evidence, not a black box.

Reviewers see every score, every image and every threshold, and their decision is recorded against their name.

admin.pollguard.africa/verification

Pending review — 7

ANAmina NjeriREVIEW
DODavid OtienoCLEAR
GWGrace WambuiCLEAR
JMJames MwangiREVIEW
FKFaith KeruboCLEAR

Amina Njeri — session evidence

Document authenticity0.96
Face match (selfie ↔ document)0.91
Liveness0.97
Duplicate searchno match

What we do not claim

We verify the document and confirm that the live person matches it. We do not currently perform a live cross-check against the government population register: our IPRS integration is sandboxed and is not a real government lookup. If your election requires a statutory register check, tell us before you commit and we will discuss what is possible.

Verification also involves sensitive personal data. Our data protection notice sets out exactly what is collected, how long it is kept and how a member can decline.

Asked by every electoral commission

Which identity documents are supported?

Kenyan national ID (including Maisha Namba), Kenyan passports, and regional documents for Uganda, Tanzania, Nigeria and South Africa — with per-country field extraction and MRZ validation.

Can a member be approved by the AI alone?

No. Every session ends in your administrators’ review queue. The models score and explain; a named person approves — and that decision is logged and attributable.

What happens to biometric data?

Face data is stored as mathematical embeddings for matching and duplicate detection, retained under your organisation’s policy and the Kenya Data Protection Act 2019. Members can request deletion through your administrators.

What if a member can’t complete the checks?

The flow supports assisted and in-person verification at supervised centers — the same evidence trail, captured by an official instead of the member’s own phone.

Verify your register before the ballot opens.

Advanced KYC is priced per verified voter — see the pricing page for the rate card and worked scenarios.