A Self-Employed Income File the Decision Engine Refused to Clear

What it does

An underwriting exception on a mortgage file reaches the Credit Analyst when the decision engine cannot clear it, for example self-employed income. In Banking the file cannot simply wait, because a stalled offer puts the rate lock, the completion date, and the broker relationship at risk. Supporting material arrives as uploaded payslips, structured bureau data, and branch notes.

The AI gathers the file, tests the exception against current credit policy and the delegated authority matrix, and sets out the next permitted steps. It requests the missing bank statements, updates the status, drafts the note explaining the exception, and logs every action it takes. When bureau data contradicts the declared income, a credit approval has expired, or the exposure passes the analyst mandate, it stops and hands over.

The Credit Analyst reviews the execution trail, releases the files that met every rule, and reworks those where affordability needs judgment. The Risk Manager enters only when a named threshold is crossed, such as an exposure limit, a policy waiver, or a regulated affordability breach. Decision authority stays with the accountable human, and each override reason is kept for the credit committee.

Where the value comes from

An exception that sits in a queue costs more than analyst minutes, because the rate lock and the agreed completion date keep moving. Lower handling time per exception is the benefit, since the agent completes the repeated policy checks, document chasing, and status updates on eligible files. For a lender that shows up as offers issued within the service promise rather than after it, which is what the broker channel actually measures.

The client measures three inputs before claiming anything: annual exception volume, baseline minutes per case, and fully loaded hourly cost. Those figures, adjusted by benchmark reduction, scope, adoption, and confidence factors, drive the `transaction_based_time_saving` model that sits behind the `handling_time_reduction` benchmark. Logs that separate autonomous completions from handoffs to the Risk Manager prove which share was actually earned.

The remaining streams stay outside the headline. `cycle_time_reduction` measures the same analyst minutes from a different angle, so counting both inflates one saving. `error_rework_reduction` does have `error_rework_cost_avoidance` available, yet pricing it requires an independent rework population. `control_quality`, `user_experience`, `auditability`, and `exception_quality` have no monetization path and no client inputs, which keeps them as control and adoption measures.

How work changes

Credit analyst spends less time on repetitive evidence checks, status updates, standard requests, and case note drafting. Routine loan exception workflow steps decline when the task plan meets policy conditions. Manual chasing also falls because the agent records missing inputs and sends standard requests.

Credit analyst gains structured supervision work. The role monitors execution audits, resolves exception handoffs, tunes thresholds, and reviews reopened cases. This work requires stronger process judgment than routine queue handling.

Risk manager participates only when the workflow crosses defined authority, risk, or expertise thresholds. The secondary role reviews the exception handoff, approves or rejects the proposed action, and records the reason. They do not become an equal actor in routine case completion.

Required skills include policy interpretation, audit review, exception triage, and control design. Accountability remains with the business owner because the agent executes within delegated limits. Freed hours are monetizable when the organization reduces overtime, avoids hiring, lowers backlog, or reallocates capacity to measured work.