The same spine, five different jobs.
Each of these starts from the same consented record. What differs is which pillars you turn on, which decision you are trying to defend, and who signs off on it.
Underwrite the applicant the bureau cannot see.
Your decline reasons are dominated by absence of data, not evidence of risk. Wynk scores from what the applicant does have — three months of statement or wallet history — and hands back a decision you can defend.
What you switch on
- — Ingestion of statements, wallet ledgers and repayment outcomes
- — Behavioural scoring with reason codes
- — Affordability decisioning under NCA Reg 23A
- — FICA CDD on the same subject, same call chain
What changes
- — A first-time borrower becomes scoreable on day one
- — Declines carry a reason a regulator accepts
- — Repayment outcomes you feed back sharpen the model
- — Consent revocation stops processing, provably
Extend the frontier without moving your risk appetite.
Wynk runs as an overlay, not a replacement. Applications your scorecard declines for insufficient data are re-presented with a behavioural score, a PD and an indicative limit — inside your existing policy, on your existing cut-offs.
Deployment shape
A scoring call on the decline path only. Your primary scorecard is untouched; Wynk answers the population it never had data for. Every overlay decision is separately auditable, so the inclusion lift is measurable rather than asserted.
What your risk committee gets
Champion and challenger scores side by side, model discrimination on the record, signed reason codes per decision, and a hash-chained processing log that survives an audit without a reconstruction exercise.
Price measured behaviour, not a declaration form.
Engagement data you already collect — steps, gym visits, screenings, adherence — becomes an explainable morbidity index, a hazard band and an actuarial view of expected claims and reserve, per member.
Shared value, made explicit
Members see what moves their band and why. Because the index is built from weighted, evidenced factors rather than a black box, an underwriting or reserving committee can interrogate each contribution.
Special information, handled properly
Health data is special personal information under POPIA §26 and §32. It is sealed under a segregated key domain, gated on its own consent purpose, and every unsealing is written to the audit chain.
Settlement data is an underwriting asset.
A merchant's card and QR settlement history is a better predictor of their working-capital risk than anything on their balance sheet. Wynk ingests it with zero personal information and returns a merchant scorecard.
Zero-PII ingestion
Point-of-sale ingestion takes settlement rows only — no cardholder, no consumer identity. The merchant is the subject; the consumer never enters the spine.
Bulk and single scoring
Score a merchant on demand or push a book through the bulk endpoint. Either way you get the same reason codes and the same metering line on your invoice.
Rank the book without trading personal information.
The propensity-to-pay endpoint returns a ranking signal over a book of accounts and exchanges no personal information in either direction. Where a recoveries-history signal is available it is consent-gated on its own purpose, cached, and sealed under its own key domain.
Consent is not assumed
A recoveries-history lookup requires an active grant for that specific purpose. Without it the score is produced from behavioural features alone and says so — it does not silently degrade.
Effort where it pays
A propensity ranking tells a collections floor which accounts justify a call. The signal is a model output, not a promise; it is priced per call and metered like everything else.
Before the first call.
How much history does a subject need to be scoreable?
Two months of observed signal and roughly ten transactions produce a usable feature vector. Below that the engine still returns a score but flags the vector as insufficient, and the observation-depth reason code carries a visible penalty — so a shallow file never quietly presents as a deep one.
Does Wynk lend, or hold a credit provider licence?
Wynk is infrastructure. The platform includes a lending book because some partners run their own origination through it, but the regulated activity — and the licence — sits with the credit provider. Nothing on this site is an offer of credit.
What happens when a data subject revokes consent?
The revocation is appended to the consent chain and the engines fail closed on the next call for that purpose: scoring raises a consent error rather than returning a stale score. The revocation itself is permanent evidence — the ledger is append-only, so a revocation cannot be deleted to make processing look authorised.
Can we run this in our own environment?
Yes. The platform is a zero-dependency Node runtime with a dual-driver data layer — SQLite for single-node and edge deployments, PostgreSQL with row-level security for multi-tenant production. There are also signed desktop builds that run the whole platform locally with no internet connection at all.
How is usage billed?
Per call, per product, at published rates — the same rates the metering engine actually charges. See pricing, which is read live from the running price list rather than typed into the page.