Sign in or create a free account to save this Card to your Deck.
Practice Card · Evidence of progress
Define normal performance and exception conditions
YOU CAN'T DETECT AN EXCEPTION WITHOUT DEFINING NORMAL.
Exception detection depends on defining what normal performance looks like, which deviations matter and who responds. AI can flag anomalies but lacks the full reason without contextual investigation. Treat the practice as a repeatable method with a clear owner, inputs, output, trial condition, and review point. Completing the steps is not proof that the operating result improved.
Expanded teaching
Expanded teaching
What the practice is for
Exception detection depends on defining what normal performance looks like, which deviations matter and who responds. AI can flag anomalies but lacks the full reason without contextual investigation. This practice strengthens Reporting and Visibility by turning a recurring operating need into work that can be performed, checked, and improved.
Prerequisites
Reporting and Visibility makes current operational reality visible early enough for a responsible person to decide and act. The purpose is not more reporting. It is trustworthy evidence about work, exceptions, trends, owners, and decisions. Clarify the policy, criteria, authoritative source, owner, and exception rules before automating the work.
Work sequence
Create a small exception register and review signal quality after a trial period. Define the input, accountable owner, produced artifact, trial condition, acceptance criteria, escalation path, and review point.
How to evaluate the practice
Which thresholds or patterns have historically indicated intervention was necessary, and how often were they false alarms? Measure report timeliness, data freshness, metric accuracy, exception-detection lead time, false alarms, time to corrective action, hours spent gathering status, and whether each highlighted metric changes a management decision. Compare the result with the prior baseline and inspect whether the change reduces delay, variation, rework, risk, or unnecessary effort.
Where AI belongs
AI can consolidate approved feeds, summarize exceptions, flag trends, draft briefs, and help compare periods. It must preserve lineage, dates, uncertainty, and missing-data warnings. Managers retain interpretation and decision rights.
Completion versus effectiveness
A completed checklist or automated workflow does not establish improvement. The practice succeeds when observable operating performance improves and remains reliable under normal pressure and exceptions.
Diagnostic inquiry
Questions to ask, not conclusions
Is “Define normal performance and exception conditions” operating as a repeatable practice with a clear owner, inputs, output, exception path, and review point?
Questions that distinguish possible explanations
- Which thresholds or patterns have historically indicated intervention was necessary, and how often were they false alarms?
- What upstream condition, control failure, or missing information could produce the same symptom?
- If you take the proposed next action, what observable result will tell you whether it helped? Proposed action: Create a small exception register and review signal quality after a trial period.
Practical application
Practical application
Create a small exception register and review signal quality after a trial period. Use a recent, bounded sample instead of hypothetical data. Name one accountable owner, record the current baseline, define the acceptance or completion standard, and run the change through a normal operating cycle. At review, decide whether to keep, revise, expand, or stop the intervention based on evidence. Escalate immediately where privacy, security, legal, customer-harm, or material financial risks cannot safely wait for the test cycle.
Ask Atria
Talk through this Card
Talk it through with us
If this Card describes your week, a free conversation is the fastest way to see what it means for your company and what to do next.
Book a free conversation