Payroll credit: digital journey, Open Finance and the predictive AI tipping point

Thesis: in Brazilian payroll credit, the borrower journey is (or converges to) fully digital — proposal, consent, formalisation and registration. Open Finance increases that fluidity: by authorising data sharing, the customer enables the proposing institution to know, under regulated granularity, the terms of the outstanding loan. That unlocks competition via informed counter-offers. The tipping point is AI applied to decisioning: identifying the best option for the customer, making the choice more transparent, and anticipating conditions that would lead to a rejected transfer — with profile models and synthetic agents that validate, ex ante, scenarios such as a migration with a rate reduction on the order of 5%.

1. Brazil’s structural differentiator: a fully digital journey

In many markets, payroll-linked credit still depends on in-person channels, paper or physical intermediation. In Brazil — especially INSS and segments with electronic registration — the borrower typically completes proposal, underwriting, acceptance and formalisation through digital channels, with an audit trail and integration to registrars.

This is not merely UX: it is market infrastructure. It lowers origination cost, compresses decision cycles and creates the base for Open Finance and AI layers without reintroducing analogue friction mid-funnel.

2. Product fundamentals

A personal loan with instalments deducted automatically from payroll, pension or formal income. Typical monthly rates for INSS between 1.6% and 2.0%. Consignable margin often up to 40% of benefit or salary after recent regulatory adjustments. Structurally low delinquency (~2.3%) versus unsecured lines — which sustains competitive appetite for the book.

3. Portability and competitive dynamics

Portability (CMN Res. 5.057/2022) lets customers move the debt with no fee, mandatory CET disclosure and a short response window for the holding institution. Classic dynamics: attack (lower rate, possible cash-out) versus counter-attack (refinance and retention). The acceptance term evidences customer intent and reduces legal dispute.

4. Open Finance: from consent to readability of the outstanding contract

Customers never type bank passwords in WhatsApp or third-party channels. The correct flow: minimal identification in the origination channel → notification at the holding institution → explicit consent in the customer’s authenticated environment → transfer of authorised data (contract, outstanding balance, rate, CET, instalments, consignable margin, maturities).

That consent is the condition for an informed counter-offer: the proposing institution stops operating on a “generic market rate” and prices against the real contract — including enabling the holder to respond with calibrated retention. Without Open Finance, competition is opaque; with it, information asymmetry shrinks under the data subject’s control.

5. Tipping point: AI in customer decisioning and transfer predictability

Open Finance resolves data asymmetry; it does not, by itself, resolve decision quality. The AI layer — models trained on customer profiles, agent orchestration and, where applicable, synthetic populations — shifts the problem from “offer a lower rate” to “offer the right option, at the right time, with a known probability of success”.

5.1 Technical objectives of the AI layer

• Option ranking: order offers by CET, instalment, residual tenor and customer preferences.
• Assisted, customer-friendly decisioning: explain trade-offs in clear language (monthly saving, total cost, margin impact) without hiding risk.
• Early rejection detection: estimate, before formalisation, the probability of registrar refusal / eligibility / margin failure — avoiding journeys that end in frustration and trust churn.
• Audit trail: every recommendation emits evidence (features, model, version, consent) for compliance and human oversight.

5.2 Profile model + synthetic agents: ex ante validation

A supervised (or hybrid) model learns customer characteristics — income/benefit, available margin, credit behaviour, rate elasticity, propensity to port — and estimates acceptance and successful registration probability.

Synthetic agents (virtual populations calibrated to real segments) allow stress-testing the offer before presenting it to the customer. Operational question example: “For this profile, does a counter-offer with a rate reduction on the order of 5% versus the outstanding contract raise conversion without raising the transfer-rejection rate?” Simulation returns an outcome distribution — take-up, drop-off, holder retention, registration failure — and only then does the digital journey materialise the proposal to the customer.

5.3 Operational controls

• Triage and enrichment with human oversight in risk bands
• Decision proposals with audit trail and model versioning
• Integration to core / registrar APIs and events — no opaque black boxes for compliance

6. From strategy to execution: demonstrable MVP

MRemittance builds a demonstrable MVP — a fully simulated ecosystem — to show that Open Finance, portability and digital formalisation already translate into runnable software:

1. Simulated consent (WhatsApp + Open Finance webview)
2. Portability proposal (rate, instalment, cash-out) with read of the outstanding contract
3. Recommendation / eligibility score layer (early rejection detection)
4. Formalisation (PDF term, credit note, digital acceptance)
5. Retention counter-offer by the holding institution
6. Settlement (close old contract, activate new one)

The differentiator is business maturity: state machine, regulatory artefacts and institution-versus-institution contest — not messaging alone.