Credit underwriting for thin-file borrowers
Saarathi's ML model reads financial signals bureau scores miss, analyzing behavioral indicators from verified sources to score an SME in under three minutes.
Four capabilities beyond manual cash-flow review
Each capability addresses a defined NBFC underwriting task that has traditionally consumed hours of analyst time.
Bureau-led scorecards were built for salaried workers and established consumer credit histories. Many Indian micro-businesses work informally, show uneven income, and have never borrowed formally. A bureau score cannot evaluate this segment.
Saarathi's model starts with the data these businesses do generate: account activity, tax compliance, digital payments, and trade invoices. It does not merely augment a bureau score. It provides an alternative underwriting method for borrowers absent from bureau records.
Alternative data capture
Bank statements, GST filings, UPI histories, and invoice records come from verified sources through the Account Aggregator framework.
Behavioral credit model
The model examines 90 days of cash-flow behavior, income stability, overdraft frequency, and payment timing to assess repayment capacity.
Credit narrative output
Each score comes with a written account of its strongest contributing signals, using language familiar to your credit committee's appraisal memos.
Adjustable thresholds
Set scoring weights around your credit policy. NBFCs serving different market segments can calibrate the platform to their own requirements.
How Saarathi handles a credit application
Three layers turn raw financial data into a structured credit decision in under three minutes.
Data Capture
Bank statements enter through Account Aggregator after consent is captured. GST data comes from the GSTN portal, while UPI history is retrieved from verified AA-linked accounts. Every source is timestamped and checked before model processing.
Feature Building
The feature layer derives cash-flow consistency, overdraft frequency, GST filing regularity, UPI density, and invoice receivable aging. Over 200 features are computed and cross-validated before they reach the model.
Decision Result
The model returns a credit score with a written narrative naming the top five contributing signals. From submission, the pipeline targets a response in under three minutes, based on benchmark runs using typical application datasets.
Credit decisions your committee can defend
RBI digital lending guidelines require NBFCs to record the basis for credit decisions. Saarathi formats its output for that requirement from the outset.
Written signal account
Each score includes a structured, plain-language explanation of its leading signals, ready to place in a credit appraisal memo.
Adjustable score weights
NBFCs can set the relative importance of cash-flow, GST compliance, and invoice behavior signals according to their credit policy priorities.
Memo mapping
Output fields correspond to standard NBFC credit appraisal memo sections, cutting down reformatting before committee review.
Built with the DPDP Act in mind
Borrower data handling follows the consent and purpose-limitation principles of India's Digital Personal Data Protection Act 2023.
Consent before use
Your LOS records borrower consent before data reaches Saarathi. Processing does not begin without a consent record that can be verified.
Purpose-led data reduction
Saarathi extracts and retains only signals needed for credit assessment. Raw source data is not kept beyond the assessment processing window.
Assessment audit record
Every assessment creates a timestamped record of its data sources and computed signals, available for RBI inspection on request.