Credit Lenses Logo
Credit LensesRequest Demo

The Core Mechanism

How Credit Lenses Works

Our three-step pipeline turns raw bank data into actionable credit intelligence in seconds.

1

The Input

It ingests raw, unstructured bank statement data—messy merchant names, cryptic reference codes, and inconsistent formats—directly via Australia's CDR infrastructure.

2

The Engine

It uses a hybrid processing pipeline—combining lightning-fast caching rules with fine-tuned Large Language Models (LLMs)—to instantly clean, categorize, and analyze the data.

3

The Output

Within 5 seconds, it returns a structured "Smart Financial Identity" JSON payload or visual report that highlights true spending habits, flags hidden liabilities (like high-frequency Buy-Now-Pay-Later usage or undisclosed gambling), and calculates the probability of future cash flow breaks.

Under the Hood

A Closer Look at Each Stage

1

Data Ingestion & Normalisation

Credit Lenses connects directly to CDR-accredited data holders via secure, consent-driven APIs. When a consumer grants access, we receive up to 24 months of raw transaction history across all linked accounts—savings, credit cards, personal loans, and more.

  • Handles multiple account types and institutions in a single request.
  • Normalises date formats, currency codes, and merchant identifiers.
  • Deduplicates pending vs. settled transactions automatically.
  • Supports batch and real-time ingestion modes for flexible integration.
2

Hybrid AI Processing Pipeline

Our engine combines deterministic rule-based matching with fine-tuned LLMs to achieve both speed and accuracy. Common transactions (e.g., major retailers, utilities) are instantly categorised via a cached rules layer, while ambiguous or novel entries are routed to AI for contextual analysis.

  • Rules Layer: Handles ~80% of transactions in under 50ms using a continuously updated merchant database.
  • LLM Layer: Processes the remaining ~20% using contextual clues—transaction descriptions, amounts, timing, and frequency patterns.
  • Feedback Loop: Corrections from the LLM layer are fed back into the rules engine, improving speed over time.
  • Risk Scoring: A proprietary model scores income stability, expense volatility, and hidden liability risk on a 0–100 scale.
3

Structured Risk Intelligence

The output is a comprehensive "Smart Financial Identity" delivered as a JSON payload via API or as a visual PDF report. It gives decision-makers everything they need at a glance.

  • Income Verification: Confirmed salary deposits, frequency, and employer identification.
  • Expense Breakdown: Categorised into 50+ spending categories with month-over-month trends.
  • Hidden Liability Flags: BNPL commitments, gambling frequency, payday loans, and undisclosed debts.
  • Cash Flow Forecast: Probability of future cash flow stress based on historical patterns.
  • Fraud Indicators: Anomalous transaction patterns, round-tripping, and synthetic income detection.

Performance

Built for Speed and Scale

< 5s

Average processing time per applicant

99.2%

Transaction categorisation accuracy

50+

Spending categories supported

24 mo

Transaction history analysed