Most AI pilots in small practices track the cases that follow the happy path. Far fewer track the moments that almost went wrong, the exceptions that required extra work, or the occasions when the practice simply got lucky. Those quieter events are often the most useful evidence a practice can collect. They reveal where the process is fragile, where training is incomplete, and where the original decision record no longer matches reality.
This article describes a lightweight exception log designed for five-to-fifteen-person behavioral-health practices. The log is deliberately simple. Its purpose is to make the near-misses and lucky escapes visible so that the practice can learn from them before they become formal incidents. The method does not require specialized software or dedicated compliance staff. It requires only the willingness to treat imperfect moments as information rather than as failures to be minimized or ignored.
Why Exceptions and Near-Misses Matter More Than Success Metrics
Success metrics, such as number of cases processed or average time saved, tell the practice whether the pilot is producing the expected efficiency. They do not tell the practice whether the process is stable. An exception log answers a different question: under what conditions does the designed workflow break, stretch, or rely on informal heroics?
In our first administrative pilot we discovered that the most valuable entries were not dramatic failures. They were the small moments when a required field was missing, when a reviewer almost skipped the confirmation form, or when a staff member used a workaround that happened to produce a correct result. Each of those moments contained information the success metrics could not surface.
What Belongs in the Log
We capture four categories of events:
Formal exceptions: cases that could not follow the approved path and required a documented deviation.
Near-misses: cases in which an error was caught only because someone happened to look twice or ask an extra question.
"We got lucky" moments: situations in which the process produced an acceptable outcome despite a gap in design or training.
Process observations: notes about friction, confusion, or emerging patterns that do not yet rise to the level of an exception.
Each entry includes the date, the person recording it, a one- or two-sentence description, and a simple tag for the category. The form is paper or a shared digital document, whichever the team will actually use. Completeness matters more than sophistication.
Making the Log Usable Rather Than Theatrical
An exception log that no one reads is worse than no log at all. We therefore schedule a short weekly review. The operations lead spends ten minutes scanning the new entries and noting any patterns. Once a month the relevant team discusses the accumulated observations and decides whether any change to the decision record, the training, or the stop conditions is warranted.
The review is kept short on purpose. Long retrospective meetings tend to be postponed. A ten-minute scan that actually happens every week produces more learning than a quarterly deep dive that is repeatedly rescheduled.
Linking the Log to the Decision Record
When the log reveals a recurring gap, the response is not only to fix the immediate problem. The response is also to update the original decision record so that the written version of the process matches the lived version. If the team has begun handling a particular exception in a consistent informal way, that handling either becomes part of the approved workflow or is explicitly rejected. Leaving it informal creates a quiet second process that the practice can no longer fully explain.
The log therefore functions as both an early-warning system and a maintenance tool for the decision record. Over time the combination of the two documents becomes the practice's most reliable memory of how the pilot actually behaved. When a new staff member joins or a clinician asks why a certain step exists, the updated decision record and the accumulated log entries supply an answer that does not depend on the original designer still being present.
We have also found that the act of writing the entries improves frontline attention. Staff who know that near-misses are expected and valued become more willing to surface small problems early. The alternative, treating every deviation as a personal failure, encourages silence until the problem is large enough to force attention.
That's a judgment call, not a tool question. Recording exceptions, near-misses, and lucky escapes is how a small practice turns operational experience into durable knowledge. Without that record the practice is left with success metrics that look clean and a set of unexamined habits that may not survive staff turnover or external scrutiny.
Slow is not the same as behind. Taking a few minutes each week to capture the imperfect moments has consistently produced clearer process improvements than any amount of retrospective analysis performed months later from incomplete memory. The practices that treat the log as optional often discover its absence only when they need the evidence and find only optimistic summaries instead.