Exception Handling
What a workflow does when a case falls outside what it was built for: a malformed input, a low-confidence result, a connected system that is down. The right answer names an owner, a queue or a notification, and says whether the action is reversed or held.
What it means in practice
The workflow article sets out the controls that make a process trustworthy: low-confidence results go to a person, exceptions have an owner and a recovery path, and an audit trail keeps the inputs, the model’s response and the approvals. The running cost article warns that a workflow trying to handle every exception on day one is the most expensive shape to run, so it should start narrow and widen with evidence.
How we use it
The running cost article’s worked example sends every mismatched invoice amount to a named person in finance while a clean match is only sampled, and keeps a log of who approved each exception. The output checking article adds the customer-facing case: a draft that promises a refund, a deadline or an exception the business never authorised is something a reviewer has to catch and someone has to own. Finding the exceptions in the first place is part of an AI audit.
FAQ
Questions about exception handling
How do I test how an AI Workflow handles errors?
Ask for the failure path. The workflow article says to ask what happens when the input is malformed, the confidence is low, a connected system is down or a case falls outside what the system was built for. A vague answer about accuracy is a failed test.
Should an AI Workflow handle every exception from day one?
No. The running cost article calls a workflow that tries to handle every kind of document and every exception on day one the most expensive shape to run. It recommends starting with the narrowest process worth automating and widening it once figures from real use are in.
What it means for a Business
Handling an exception means deciding in advance who is told, where the case waits and whether the action is undone or held. A vague answer about the model being very accurate is not a plan, and nor is a recovery path nobody owns.
Ready to put this to work?
Tell us where your team is with AI and we will tell you honestly what would make the biggest difference.