Automation
Automation only works if you design the exceptions
The happy path is easy to draw. The process people actually run includes the cases that do not fit. Those cases decide whether automation helps or creates another inbox.
May 21, 2026 · 5 min
Teams often automate the version of a process that appears in a slide: a request arrives, a system updates, a confirmation is sent. The live process is less tidy. A field is missing. A customer replies in the wrong channel. Two records almost match.
If those cases have nowhere to go, people create a shadow process. The automation still runs, but the real work returns to email.
Map the work that already happens
Sit with the people who run the process and watch the exceptions. Count them. Name the reasons they occur. You will usually find a small set of patterns, not infinite chaos.
Those patterns become first-class paths: a review queue, a clarifying question, a fallback owner. Automation should make the exception visible, not hide it.
Use AI only where the rule is not enough
Language, documents and classification are good homes for AI. Status updates, identity checks and system-to-system movement are often better as rules.
The result is a quieter operation: fewer handoffs, clearer ownership, and a process that still works on a difficult Tuesday.