Count your overrides by reason, or you are managing a screening policy you cannot see

Your screening criteria are written down. Your overrides are not.

An override is the decision to approve an applicant your own criteria declined, and in most operations it is recorded as a checkbox with an optional free-text note. Which means the set of exceptions to your policy, the part carrying most of your risk, is the part you have no report on.

The rate difference is large enough to change how you staff this

Across our own portfolio, applicants approved by override default at roughly 4 times the rate of applicants approved on the standard.

That number is not an argument for eliminating overrides. Overrides are frequently correct, and a policy with no exception path either declines good residents or gets bypassed informally, which is worse because it becomes invisible.

It is an argument for counting them, and the reason counting matters is that 4x is an average over reasons that are not equally risky. Some override categories almost certainly perform near the standard. Others are much worse than 4x. Without a category breakdown you cannot tell them apart, so you are either tightening all overrides or none.

Six categories, each a different bet

These cover most of what actually happens:

Income shortfall with a guarantor. Declined on ratio, approved because a guarantor signed. Risk depends almost entirely on whether the guarantor was verified, which is usually the weakest link.

Income shortfall with an increased deposit. Declined on ratio, approved with additional security. The deposit caps your loss rather than reducing the probability, and those are different things.

Credit score below threshold, no derogatory history. A thin file or a low score with no collections, judgments, or charge-offs. Frequently the best-performing override category, because the score was measuring absence of history rather than bad history.

Credit with adverse history. Collections, charge-offs, or prior housing debt. Different risk from a low score with a clean record, and it gets collapsed into the same bucket constantly.

Prior rental history problem. A previous eviction filing, a balance owed to a prior landlord, or a negative reference. This is the category with the most predictive content and the one most often overridden on a sympathetic explanation.

Document or identity flag cleared manually. A verification flag a person reviewed and decided was benign. Worth tracking separately because it measures how well your flags are calibrated as much as it measures applicant risk.

Six categories is enough to act on and few enough that a leasing agent will pick correctly. A free-text field produces forty variants of the same reason and no report.

Who can override is a policy decision that is usually a default

Two things worth setting deliberately.

Authority level. If a leasing agent can override, override volume will be high and quality will vary by agent. If a regional must approve, volume drops and so does leasing speed. The common middle is agent-level for one category and manager-level for the rest, and the categories should be chosen by measured performance rather than by intuition.

Rate ceiling. An override rate above roughly 15% of approvals means the written criteria do not match how the property actually operates, and the honest fix is to change the criteria rather than to keep excepting them. A rate under about 2% suggests either unusually well-matched criteria or an informal process happening off the record.

Neither threshold is a rule. Both are worth knowing your own number against.

Structured reasons are also the fair housing answer

This connects to compliance in a way that makes the reporting case stronger.

An override is a discretionary decision, and discretionary decisions are where disparate treatment happens without intent. Two applicants with similar files, one approved and one declined, is exactly the pattern a complaint is built on.

Structured override reasons with a required selection, a recorded decision-maker, and a timestamp produce a defensible record: this category of exception is granted under these conditions, consistently, and here is every instance.

A free-text note reading "good applicant, approved" produces no defense at all. It is also, separately, useless for the outcome analysis, so the compliance fix and the reporting fix are the same fix.

What ProofUp provides and what it does not

Verdicts come back as Pass, Flag, or Fail with the specific signal behind every flag, and every attempt is recorded in an audit trail. So the input to an override decision is documented: what was flagged, why, and what the reviewer saw.

Where we stop, and this is a real gap rather than a caveat:

We do not hold your rent ledger. Joining override records to delinquency outcomes twelve months later requires your property management system, and it is work rather than a report. The 4x figure comes from that analysis being done, not from a screen in the product.

We do not set your override categories or your authority levels. Those are policy.

We do not decide whether a flag should be overridden. A human decides, and the platform's job is to make sure the decision and its basis are recorded rather than to make it.

Do the join once, on last year

Take approvals from twelve to eighteen months ago. Split them into override and standard. For each group, pull the share that went 30 days delinquent at any point.

Then split the override group by reason, even if you have to read notes by hand to categorize them.

That afternoon of work tells you which of your exceptions are fine and which one is carrying your losses. It is the most useful thing available in screening and the fact that it requires manual effort is why almost nobody has done it.

What is your override rate, as a share of approvals?