Thin-File Credit Agent AI

Invoked when: SME or self-employed applicant has thin or no bureau footprint but is VAT-registered with at least 6 months of filing history

• –Reads UAE FTA / ZATCA VAT data for the applicant — VAT returns, net VAT liability, and filing regularity — and extracts a revenue picture: what is the average monthly taxable turnover, is the turnover trajectory growing, stable, or declining, and is the borrower filing regularly or showing lapses that suggest business irregularity or avoidance.
• –Computes a VAT-based income proxy: for a trader or service business, VAT turnover is the top line; applying a sector-specific EBITDA margin estimate (available in the NTC policy corpus by CN code) produces a rough income proxy that allows DBR (Debt Burden Ratio) computation even without verified income declarations. The proxy is labelled explicitly as an estimate with the margin assumption stated — it is not presented as verified income.
• –Cross-checks the VAT turnover against the open banking / data aggregator bank credit flows — the bank credits should be broadly consistent with the VAT turnover over the same period. Where they are consistent, the corroboration strengthens the income proxy's reliability. Where they diverge significantly, it flags the discrepancy rather than using the higher figure.
• –Flags VAT-specific risk indicators: a business registered at a residential address without plausible commercial activity, a VAT registration less than 12 months old (short vintage risk), or a sudden turnover spike in the most recent 2 months ahead of loan application (potential income inflation for credit purposes). These are flags for L3 human review, not automatic declines.

Invoked when: open banking / data aggregator data includes AANI / instant payment scheme transaction history that can supplement or contextualise other thin-file signals

• –Reads the AANI / instant payment scheme transaction data from the open banking / data aggregator feed — inflow volume and frequency, outflow volume and frequency, counterparty diversity (number of distinct AANI / instant payment scheme IDs transacted with), average transaction size, and the regularity of inflow patterns — and extracts behavioural signals about the borrower's economic activity and financial management.
• –For a small trader whose primary revenue channel is AANI / instant payment scheme collections, the AANI / instant payment scheme inflow pattern is the closest available proxy for business revenue frequency and stability: a borrower receiving 200+ inflows per month from 80+ distinct counterparties over 12 months is demonstrating a live, active trading business that the bureau cannot see. The stability and growth of this pattern over time is a meaningful signal.
• –Identifies AANI / instant payment scheme behavioural patterns associated with financial stress that are relevant even without bureau data: a sharp drop in AANI / instant payment scheme inflows in the last 60 days before application, a shift from diverse counterparty inflows to a few large irregular transfers (potentially personal borrowings disguised as business income), or AANI / instant payment scheme outflows to known finance company or MFI AANI / instant payment scheme IDs suggesting undisclosed borrowing obligations.
• –Does not use AANI / instant payment scheme transaction data to infer lifestyle, spending preferences, or any attribute that is not directly relevant to the borrower's capacity and willingness to repay the proposed loan. AANI / instant payment scheme analysis is bounded to income proxy derivation, financial management signals, and obligation identification. Lifestyle-based credit assessment raises fairness concerns that are inconsistent with CBUAE / SAMA's fair lending guidelines.

Invoked when: open banking / data aggregator bank statement data is available for at least 6 months, serving as the primary or supplementary income verification source

• –Reads the full 12–24 month open banking / data aggregator bank statement data and constructs a structured cash flow picture: average monthly net inflows (total credits minus intra-account transfers and identifiable loan disbursements), average minimum monthly balance, end-of-month balance trend, bounce rate on outgoing payments, and the seasonality profile of inflows across months — distinguishing businesses with predictable seasonal patterns from those with genuinely irregular cash flows.
• –Computes a cash flow-based repayment capacity estimate: what instalment amount could this borrower service while maintaining a minimum acceptable end-of-month balance and a buffer for irregular expenses? This is the thin-file equivalent of DBR (Debt Burden Ratio) — derived from observed cash flows rather than declared income, and therefore more directly reflective of actual financial behaviour than stated income figures.
• –Identifies the quality of inflows: salary-like credits (regular fixed amounts from a consistent employer or client) are higher reliability than irregular merchant credits; confirmed digital income receipts are higher reliability than cash deposits that appear as bank counter credits. The cash flow proxy rates each income stream by its reliability and weights accordingly.
• –Flags structural cash flow risks: average monthly balance below one month's proposed instalment (insufficient buffer), a pattern of using the overdraft facility as a regular operating resource rather than an emergency bridge, or a declining average balance trend over the most recent 3 months suggesting the borrower's financial position is deteriorating ahead of the loan application.

Invoked when: all available alt data signals have been processed and a synthesised NTC credit assessment is required for the underwriting decision

• –Synthesises the VAT profile, AANI / instant payment scheme signal summary, and cash flow proxy into a single structured NTC credit assessment that covers four dimensions: capacity to repay (derived from cash flow proxy and VAT income estimate), stability of income (regularity and trend of inflows, VAT filing consistency), financial behaviour (bounce rate, balance management, existing obligation discipline), and business health for SME borrowers (VAT turnover trend, business vintage, counterparty diversity).
• –Applies the NTC-specific credit policy — retrieved via RAG — to the synthesised assessment: eligibility criteria for thin-file products (minimum months of AA data, minimum VAT vintage, minimum average monthly balance), the product-specific limits for NTC borrowers (ticket size ceiling, maximum tenure, collateral requirement at different ticket levels), and the escalation rules that define which NTC cases can be approved at L1 versus which require L2 or L3 review.
• –States the confidence level of the assessment explicitly — and honestly. An NTC assessment built on 6 months of AA data, 8 months of VAT filing, and no AANI / instant payment scheme history has a different confidence profile than one built on 24 months of AA data, 18 months of VAT, strong AANI / instant payment scheme signals, and verified salary/tax records. The confidence level drives the escalation routing and the recommended ticket size within the policy range.
• –Produces a plain-language explanation of the assessment for the credit officer — what signals are available, what they indicate, what signals are absent and what that absence means, and what the key risk factors are for this specific borrower. The explanation is designed for a credit officer who needs to understand the basis of the assessment, not just the output, so they can exercise informed judgment in the L2 or L3 review.