Work and processes
What Is Exception Handling: Designing Exceptions That Decide the Success of Automation, A Complete Guide
Korea Deep LearningPublished
Exception handling is the process of pulling cases that cannot be judged by fixed rules out of the normal flow, sending them down a separate path, and returning them to the flow after a person has checked them. The stability of automation is decided here.
Why exceptions decide the success of automation
The speed of normal cases hits its limit quickly.
When automation is evaluated, attention usually focuses on normal cases. People ask how fast they are processed and what the accuracy is. Once it is in operation, however, the time and staff are consumed by exception cases.
If the definition of an exception is broad, the effect of automation disappears. With a setup that sends everything to a person if there is any doubt, staff workload does not go down. Conversely, if exceptions are defined too narrowly, cases that should have been filtered out get through. In the end, the success of automation depends not on how quickly normal cases are processed, but on how precisely exceptions were defined.
The precise definition of exception handling: how it differs from error handling
Three axes that separate error handling and exception handling
First, the nature differs. An error is a state in which the system has failed. A file that will not open, or processing that times out, belongs here. An exception is a state in which the system worked normally but lacks enough grounds to make a judgment.
Second, the response differs. For errors, you retry or fix the cause. For exceptions, you hand the case to a person to make the judgment. No amount of retrying will produce a value when the grounds for it do not exist.
Third, the management approach differs. The goal for errors is to lower their occurrence rate. For exceptions, the goal is not only to lower the rate but also to filter accurately. An exception rate of zero may not be a good state.
The two are not alternatives. Both are cases that drop out of the normal flow, but they must go to different places and be handled in different ways.
Five situations that create exceptions
Five situations that occur in real work
First, a judgment is needed but the value is missing. Cells containing only two periods, cells covered by a white bar, and rows where only a printed unit remains actually occur. There is no basis within the document to tell whether the entry is not applicable or was left blank by mistake.
Second, there are multiple values and one cannot be chosen. There are cases where a fax retransmitted several times had different date notations overlapping, leaving no basis for deciding the receipt date.
Third, information conflicts between documents. If the name or account number on an application differs from that on a supporting document, a person must judge which one is correct.
Fourth, the form cannot be identified. A newly seen revised form, or a document whose top has been damaged, belongs here.
Fifth, a value was extracted but the confidence is low. This frequently happens with items where handwriting varies widely or the print is faint.
How to design exception handling in practice
Divide the types of exceptions and build a path for each
If all exceptions pile up in one queue, staff have to find the cause every time. Separating them by type makes it clear where to look.
The check method differs for a case where a value is missing and a judgment is needed, a case where documents do not match, and a case where the form is unknown. The screen has to make this distinction for the processing speed to pick up.
Always build the return path
A commonly missing part of exception handling design is the return path. After staff correct a value, the result has to flow back into the process and be registered in the system.
Without this path, staff check the exception screen and then enter the data into the business system separately. That leaves the situation no different from before automation.
Exception handling in the Korean context
Korean finance and public-sector work has many processes with deadlines. In reviews with a statutory payment deadline, or in accident reports with a golden-time window, the time an exception case sits in the queue becomes a risk in itself.
For this reason, a setup is needed in which a person is notified as soon as an exception is identified, and the order can be adjusted according to urgency. In addition, the reason for the exception and the outcome of its handling must remain as a history to meet audit requirements.
The key is to define exceptions narrowly
The success of automation is decided not by how fast normal cases are processed, but by how narrowly exception cases were defined. If exceptions are broad, the review burden stays as it is. If they are narrow, cases that should be filtered out get through.
In practice, the criteria fall into three groups. The first is when the confidence of a value falls below the standard. Where this threshold is set directly determines the share of exceptions. The second is when cross-checking between documents finds a mismatch. These are judged automatically, and the more the item affects the work, the stricter the criterion. The third is when business rules require human judgment, such as amount bands or document requirements that are set by policy.
Reviewing the share and causes of exception cases periodically provides grounds for refining the criteria. If exceptions keep occurring for the same item, it is time to revise that item's extraction standard or validation rules.
Frequently asked questions
Is a lower exception rate always better?
Not necessarily. Lowering the rate while letting through cases that should have been filtered out is dangerous. Look also at the share of exception cases that were actually corrected after being sent for review.
How can I reduce exceptions?
Break down the types of occurrence. If they cluster in a certain form, revise the item definitions. If they cluster in certain items, revise the extraction standards or validation rules.
Is a missing item an exception?
It is, when a judgment is needed. If there is no basis within the document to tell whether it is not applicable or was left blank by mistake, do not fill it in arbitrarily. Passing it on is the safer choice.
How do I manage the processing speed of exception cases?
Manage waiting time as a separate metric. In work with deadlines, the waiting time of exception cases determines the total lead time.
What is an appropriate share of exceptions?
It depends on the work. One case replaced full manual cross-checking with a structure that reviews only exception cases in pension screening, and a similar approach has run in credit screening.
How are exception cases handled?
They are passed to the staff member's screen. The reason the case became an exception and the relevant location in the original should be shown together to shorten review time.
Tell us how you handle exceptions today, and we will review a separation design by exception type with you.
Contact us



