The Pipeline

From behavior to a credit decision

Three stages, consent at each step, score delivered in milliseconds. Here is what occurs between the API call and the number your underwriter receives.

Data Pipeline

Three-stage pipeline

Stage 1

Borrower Data Signals

Your system sends borrower identifiers and permissioned data access points in one API call. Your side captures and logs consent first. Supersymm receives only data the borrower authorized.

Available signal categories include mobile recharge, utility payments, geolocation, peer transfer cadence, and app engagement, based on borrower consent and the data your integration provides.

Stage 2

Supersymm Engine

The engine extracts features, then normalizes, weights, and combines indicators across categories. Signal types decay at different rates, so a mobile top-up from yesterday counts more than one from four months ago.

The scoring model uses an ensemble approach, trained on alternative-data populations and calibrated against bureau-scored benchmark cohorts. An explainability layer records the signals driving each borrower score.

Stage 3

Lender Score Output

The API response arrives in the same sync window as your bureau pull, with median latency under 800ms. It includes the numeric score (0-1000), confidence band, top three signals with relative weights, and a data freshness flag.

Send the score into your existing loan origination system. No model rebuild is required. Partners may blend it with a bureau score where both exist, or use it alone for no-file applicants.

The Model

How the scoring model works

A plain-language methodology overview for lenders.

Ensemble approach

The model combines signal types with gradient-boosted decision trees. Mobile, utility, and location data pass through separate feature pipelines, then combine at scoring. A score remains possible when categories are missing, with a related confidence adjustment.

Fairness validation

The model is tested against bureau-scored benchmark cohorts for systematic disadvantage across demographic groups. Validation compares predicted default rates by score band with observed defaults in early-access pilot data. Methodology is available to lenders on request.

Explainability layer

Each response lists the top three signals contributing most to the borrower's score. If a lender declines an application, these data categories can support its explanation. No black-box scoring.

Signal decay rates

Behavioral signals receive recency weights. A recharge from last week carries more predictive weight than one from last quarter. The 90-day window is the default; shorter windows are available for rapidly changing profiles. Each response reports freshness.

Data and Privacy

How borrower data is handled

Borrower permission is required for every score. Supersymm processes only signals the borrower explicitly permits the lender to share.

  • Supersymm stores no raw behavioral data. It processes derived features only.
  • Score goes only to the requesting lender, never stored or pooled.
  • Derived features are purged 12 months after the borrower's last lender query.
  • The lender-side consent flow is documented and auditable.

LGPD principles

Brazil's Lei Geral de Protecao de Dados (LGPD) principles guide Supersymm's handling of personal data, including lawful basis, purpose limitation, minimization, and access rights. Lenders remain responsible for borrower disclosure in their loan origination flow.

Consent flow

The lender shows the consent disclosure before sending data to Supersymm. Every scoring request must include the consent_confirmed field. Requests without confirmed consent are rejected by the API.

Want a deeper look at the methodology?

Our data team can review model validation, signal categories, and integration specifics before you choose a tier.

Talk to Our Data Team