Underwriting — company 1526

one applicant in → one score, one decision, one limit out · your API key stays in this browser

Credit history (Зээлийн мэдээллийн сан)

Scored by the v4 trained model: will a genuinely new loan go bad within 12 months (Gini 0.47; class-history directions enforced monotone). Blank fields take neutral defaults and are listed in the result.

Borrower

Зээлийн эрх (the rule-engine limit) is now an output, computed with your own DSTI formula — income × 55% minus estimated payments on existing debt, annuitised at the loan's rate and term — and shown in the result. Supplying it via the API still overrides the estimate.

Loan being requested (optional)

Leave blank to underwrite the borrower without pricing a specific loan.
How the score works. Three checks — credit history, the borrower (age, income, tenure), this loan — each on 300–850. The universal score is the lowest, never an average: a strength can never hide a weakness. Thresholds come from Cash24's own economics, not legacy rules: blended APR 75.3%/yr → 6.3% interest per month; minus ~1.5% opex/funding; a default loses ~85% of principal (LGD). So lending breaks even between 5.6% (net) and 7.4% (gross) monthly default risk — against a book average of 4.48%, that is 1.25× and 1.65× average risk. The model's relative risk maps those to score cutoffs, shown with every result. Indicative until the shadow run re-anchors the probabilities; a borrower already in default elsewhere is a policy REFER regardless of score.

Collection score C1–C5

Separate product: for borrowers ALREADY in trouble (open loan class 02+). Predicts full cure in 6 months and carries the measured recovery / LGD per band. Uses the credit-history fields above — set "Worst open class now" to 02+ — plus the two fields below.
Service status at /health · model details at /modelcard (needs your API key)