> ## Documentation Index
> Fetch the complete documentation index at: https://docs.amlok.io/llms.txt
> Use this file to discover all available pages before exploring further.

# Business underwriting research

> Evaluate business cash flow, obligations, operating performance, and evidence quality.

Business underwriting research should test whether the proposed representation improves assessment of a defined business and lending product. It should not infer a universal credit decision from one financial ratio or from a graph's visual appearance.

## Assessment dimensions

| Dimension | Evidence to combine | Proposed questions |
| - | - | - |
| Business identity and scope | Authorized institution records, verified identifiers, permitted registry data | Which legal entity, trading names, accounts, and related entities are actually in scope? |
| Revenue and demand | Commerce orders, invoices, payment activity, settlements, bank receipts | Are sales growing, seasonal, concentrated, refunded, or unmatched? |
| Profitability | Accounting statements, cost records, payroll, transaction evidence | Are margins supported and persistent? How much is accrual versus cash? |
| Repayment capacity | Scheduled debt, reconciled cash flow, financing proceeds, tax obligations | Which obligations are known, estimated, missing, or newly observed? |
| Liquidity | Cash, receivables, payables, inventories, reserves | Can the business absorb a defined delay or cost shock? |
| Leverage and security | Facility terms, balance sheet, permitted security-interest evidence | What is the supported debt structure? Which balances and priorities remain unknown? |
| Operational resilience | Customer/supplier dependence, fulfillment, disputes, system coverage | Which changes warrant analyst attention, and what alternative explanations exist? |
| Evidence integrity | Reconciliation, document provenance, cross-source consistency | Is there a discrepancy requiring investigation, rather than a proven allegation? |

## Unit of analysis

A full Insights run concerns an application and a frozen evidence snapshot. Evaluate related businesses and guarantors only with a documented association and permitted evidence. Keep the business as the billing/workload subject for included stages; research does not create standalone verification products.

Separate working-capital loans, invoice finance, equipment finance, merchant finance, and commercial mortgages in outcome reporting. Their repayment schedules, loss definitions, collateral, seasonality, and observation units differ. Pooling them into one “accuracy” figure can conceal poor performance on a material product.

## Proposed outcome tasks

Define a primary credit outcome such as a specified delinquency state within a stated horizon using the institution's independently supplied performance data. Evaluate default, first-payment default, fraud, and covenant breach as different endpoints. Record cures, refinancings, prepayments, restructuring, and censored follow-up.

Use time-to-event methods where observation lengths vary. Compare the bank-only, multi-source tabular, event-model, and graph-enhanced systems on identical borrowers and cutoffs. Report discrimination, calibration, expected decision costs, subgroup performance, and uncertainty. A raw public filing is not a substitute for repayment outcomes.

## Scenario analysis

For a fictional business with CAD 100,000.00 of eligible receipts, CAD 70,000.00 of eligible operating cash outflows, and CAD 15,000.00 of scheduled debt service, the cash-flow coverage proxy is 2.00x. A 20% receipts reduction with costs unchanged yields 0.67x. This is a mechanical sensitivity, not an estimated event probability or a validated forecast.

Specify which costs can vary, timing of collections, interest-rate exposure, currency, and contractual repayment structure. Never apply a rate shock to fixed-rate debt as though it reprices immediately. Avoid double-counting a shock in both the financial inputs and an unexplained risk overlay.

## Human and operational evaluation

Measure complete analyst handling time, corrections, evidence retrieval, overrides, and rework, as well as machine time. Use a controlled assignment of comparable files or an appropriately designed crossover study. An earlier risk signal does not prove a loss was prevented. Approval expansion or loss reduction requires a prospective design and observed outcomes, not simply a retrospective score comparison.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.