KYC onboarding cut to under ten seconds

A BNPL lender in India lives or dies on how fast a new borrower can be brought through identity checks without the check itself becoming the reason the person abandons the funnel. The operational problem is not really 'verify identity' in the abstract; it is doing document capture, face match, liveness and field matching fast enough that a first-time user on a patchy mobile connection completes onboarding, while still catching the impersonation and forged-document attempts that a credit product attracts.

The framing that matters here is that speed and assurance pull against each other, and the interesting design decision is where you put the boundary between the flow that completes on its own and the flow a human reviews. HyperVerge's own angle for ZestMoney foregrounds this boundary rather than hiding it, which is the honest way to describe KYC: some share goes straight through, the rest is queued for a person, and the split itself is the thing to manage.

HyperVerge, the publisher, reports that ZestMoney's onboarding time fell from roughly 10 minutes to under 10 seconds after adopting its automated C-KYC workflow covering document-quality validation, face match, liveness detection and field matching. HyperVerge states the flow reached about 70% automation, which it pairs with a corresponding manual-review share of roughly 30% rather than claiming full automation.

HyperVerge additionally reports a transaction increase of around 300% during 2021 in connection with the deployment. No publication date is stated for the case study, so these figures carry no verifiable time anchor and should be treated as undated vendor-reported claims that an evaluator would want to re-verify directly with the parties before relying on them.

An Indian SMB should ask what the 70/30 split actually means in operation: what triggers a case into manual review, who staffs that queue, what the review turnaround is, and whether the 30% is stable or spikes with fraud waves and new document types. A sub-10-second automated path is only useful if the manual tail does not quietly become the real onboarding time for a meaningful slice of applicants.

Then inspect the evidence trail. Ask whether the system produces an audit log that can reconstruct any single decision after the fact: which document was captured, what the liveness and face-match signals were, what field-matching rules fired, and who overrode what. Also ask what biometric and document data is retained, for how long, and on what stated consent basis, and confirm applicable rules, consent, retention and escalation design with qualified counsel or compliance owners rather than assume.

This is a lending use case at a scale and risk appetite that most SMBs will not share. A large BNPL book can absorb the cost of a tuned model and a staffed review desk, and its fraud pressure justifies aggressive liveness. A smaller operator copying the same thresholds may see either a much larger manual queue it cannot staff or a false-accept rate it cannot afford, so the reported automation share does not transfer as a target.

The applicable rules also differ by product, and whether any onboarding obligations attach to a given SMB's product is not something to infer from a lender's setup; confirm applicable rules, consent, retention and escalation design with qualified counsel or compliance owners. Regional ID-document coverage and consent handling likewise need re-testing for the SMB's own applicant population.