40+ engineered signals, not a dashboard
Line-item extraction is table stakes. What a lender actually uses is structured credit signals — bounce prediction, salary consistency, EMI aggregation, fraud indicators, and for MSMEs, four-layer synthetic financials. TransactIQ produces them as API outputs the credit model consumes directly.
Five signal families
Representative signals from each family. Exact outputs and thresholds are configurable per lender policy; the families below are the shape of what you get.
Bounce and default prediction
- ▸ Historical bounce incidence across rails (NACH, cheque, ECS)
- ▸ Mandate-level re-presentation patterns
- ▸ Insufficient-funds flag frequency trend
- ▸ Approximate days-past-due signal from EMI irregularity
Income and obligation
- ▸ Salary credit detection including variable-date patterns
- ▸ Salary consistency scoring across the statement window
- ▸ EMI obligation aggregation across identifiable lenders
- ▸ Debt-burden ratio approximation from EMI outflow share
Cash flow and volatility
- ▸ Monthly inflow and outflow trending
- ▸ Month-end balance trajectory
- ▸ Inflow source concentration and diversification
- ▸ Cash-flow volatility index for current-account MSME assessment
Fraud and anomaly
- ▸ Round-tripping pattern detection
- ▸ Mule-account indicator pattern signals
- ▸ Fake salary credit detection for NTC (new-to-credit) cases
- ▸ Counterparty concentration anomalies
MSME synthetic financials
- ▸ Personal vs business transaction separation
- ▸ Synthetic P&L — revenue by channel, inferred direct and indirect costs, indicative EBITDA
- ▸ Synthetic balance sheet — working capital view, visible borrowings, indicative net worth
- ▸ Synthetic cash flow — operating, investing, and financing cash inferred from bank activity
How signals are designed
Four design commitments the credit, risk, and regulator-facing teams can rely on.
Signals, not summaries
TransactIQ returns structured signals the credit team's existing model can consume — not a dashboard an underwriter has to re-interpret. Every output is consumable as an API field, not a human-readable paragraph.
Rule-traceable, not opaque
Every signal can be traced back to the underlying transactions that produced it. A regulator asking 'how did you conclude the salary is inconsistent' receives a line-item answer, not a model confidence score.
Configurable thresholds
Each lender's policy for what counts as a salary, a bounce threshold, a round-trip period, or a synthetic-P&L segmentation is a configuration. The engine is shared; the policy is per-tenant.
Measured, not assumed
Signal efficacy is measured against realised outcomes on production portfolios. A signal that stops predicting what it claims to predict is deprecated, not left in the output for backward compatibility.
Want the full signal catalogue?
Early-access partners receive the full field-by-field signal catalogue with example outputs and configuration guidance for lender policy mapping.
Request signal catalogue