How It Works

From raw transaction data to a scored decision in seconds.

Panthera's pipeline has three stages: data ingestion, behavioral feature extraction, and scored output with reason codes. Each stage is designed for production latency and regulatory explainability.

Step 01

Ingest alternative data from consented sources.

The lender collects data consent from the applicant at origination. Panthera connects to the data sources the applicant has authorized through tokenized API calls or secure batch uploads. No bureau pull is required. The ingestion layer normalizes formats, currencies, and date ranges across markets before anything enters the feature pipeline.

  • Mobile wallet transaction logs (90-day lookback minimum, 365-day preferred)
  • Utility billing records with payment timestamps and clearance status
  • E-commerce purchase and installment history from platform partnerships
  • Payroll transfer metadata from consented bank account connections
# Data ingestion: consent token flow

POST /v1/consent/init
{
  "applicant_id": "ap_7k2m9p",
  "market": "vietnam",
  "sources": [
    "momo_wallet",
    "viettel_utility",
    "shopee_commerce"
  ]
}

# Response: consent URL returned to
# lender for borrower authorization
{
  "consent_url": "https://consent.pantherahq.com/...",
  "expires_at": "2026-06-09T14:00:00Z"
}
Step 02

Extract behavioral features across four families.

Panthera's feature engineering layer converts raw transaction timestamps and amounts into 200+ behavioral features spanning four families: cash-flow velocity, repayment regularity, income seasonality, and obligation clearance. Feature availability adapts to the data sources actually present for each applicant.

Cash-Flow Velocity
Average monthly inflow/outflow ratios, rolling balance volatility, peak-to-trough range across 30/60/90-day windows.
Repayment Regularity
Small obligation clearance rate, day-of-month payment consistency, delay distribution histogram, installment-to-income ratio.
Income Seasonality
Payroll regularity index, seasonal amplitude modeling, regional holiday pattern adjustment, gig income burst identification.
Obligation Clearance
Utility on-time rate, telecom renewal cadence, marketplace installment completion, peer transfer reciprocity patterns.
Step 03

Score and explain. Not just a number.

The API response includes the Panthera Score (0-1000), a band designation (A through E), and up to five reason codes that reference the actual model features contributing to the decision. These are the features with the highest SHAP values, expressed in plain language tied to the data inputs.

scoreinteger 0-1000
Panthera Credit Score. Higher = lower predicted default probability. 750+ is equivalent to prime credit in most market contexts.
bandstring A-E
Score band where A = lowest risk (800+), E = highest risk (below 400). Band thresholds are configurable per lender policy at onboarding.
reason_codesstring[] max 5
Human-readable descriptions of the model features with highest SHAP contribution to the decision. Tied directly to actual input features, not generated rationalizations.
feature_countinteger
Number of features actually computed for this applicant, based on available data sources. Useful for assessing score confidence in thin-data cases.
latency_msinteger
Server-side processing time in milliseconds, excluding network transit. P99 target is 800ms. Typically 250-450ms for standard single-source requests.
# Full score response: Indonesia market

{
  "request_id": "req_92nkp1",
  "applicant_id": "ap_7k2m9p",
  "market": "indonesia",
  "score": 748,
  "band": "B",
  "reason_codes": [
    "R1: High GoPay wallet velocity (90d)",
    "R2: Consistent PLN bill clearance",
    "R3: Stable payroll deposit cadence",
    "R4: Low outflow volatility index"
  ],
  "feature_count": 187,
  "data_sources_used": [
    "mobile_wallet",
    "utility_billing",
    "payroll"
  ],
  "latency_ms": 312,
  "model_version": "id-v4.2",
  "scored_at": "2026-06-09T09:14:22Z"
}
Next Steps

See the platform in your market context.

Talk to our team about sandbox access, data integration requirements, and what approval lift looks like in your specific borrower segment.

Request Sandbox Access Explore the Platform