LO
Loan officer / credit officer — tasks, one by one
The unit of analysis is the task, not the job title. Each one below carries its direction, whether the judgement rests on evidence or on platform inference, the reasoning, and what it does not establish.
Every task on this page#
Scoring the application
Automating✓ Evidence-backedTurning income, history and collateral into a decision about the ordinary case that fits the policy.
This has been automated since credit scoring arrived in the 1960s, and the current wave is an extension rather than a beginning: the inputs are structured, the outcome is measurable years later against default rates, and the model can be retrained on that outcome. It is one of the oldest completed automations in white-collar work.
Sixty years of automating this task did not remove the occupation, and that is the most useful fact on this page — but it did change who gets in and how many. Read a high exposure here as a statement about the ordinary case only: the policy defines what ordinary means, and everything outside it still arrives at a person.
Arguing the exception
Still human-led≈ Platform inferenceThe self-employed applicant whose income is real and does not look it, the business with one bad year for a reason — building the case for lending outside the model.
An exception is by definition a case the model scored wrong, so the model cannot be the thing that finds it. What makes the case is context that is not in any file — knowing the trade, the local market, why the year was bad — and a person willing to attach their name to the judgement.
The exception surviving is not the same as the exception being permitted. Many lenders have narrowed or removed override authority entirely, which deletes this task without automating it — and where that has happened the stated reason is consistency rather than cost. A task can be irreplaceable and still be abolished by policy.
Gathering the file
Automating✓ Evidence-backedChasing documents, verifying them, filling the forms, and getting the package to a state where somebody will look at it.
Document collection and verification is a rules-based workflow with a checkable outcome — the document is present and matches, or it is not — which is the profile that lets a system retry unattended. Open banking and direct data feeds removed a large share of it before models arrived.
This is the task that employs the junior half of a lending team, so automating it removes the entry route rather than the work — the judgement needed above it was learned by handling a few hundred files. Nobody has said what replaces that, and the consequence arrives as a shortage of experienced officers five years later.
Being the name on the file
Still human-led≈ Platform inferenceApproving within your limit, and answering for the decision at the credit committee and, later, to the regulator.
Accountability for a lending decision is assigned to a person in every regime we can verify, and several markets now add an explicit right to human review of an automated decision. That is held in place by law rather than by skill, which makes it the most durable task here and the most contingent.
A signature is cheaper to keep than a decision, and a bank can keep the name on the file while moving everything that produced it. Human review that consists of clicking approve on a model's output is human review in name only, and no measurement distinguishes the two from outside.