Where automation fails in operations, and what to fix first

Most automation projects that stall share the same pattern: a capable tool wrapped around a process nobody had written down. Here is what we look for in the first week of an audit and the order we fix it in.

The undocumented exception

Every team has a step that one person handles "by eye". Automating around it produces a system that works 90% of the time and needs a human 100% of the time. Fix: write the exception down, decide whether it is a rule or a judgement, and automate only the rule.

Inputs that were never structured

PDFs, forwarded emails and screenshots of spreadsheets. Modern extraction handles them, but the confidence has to be surfaced, not hidden. Fix: extraction with a confidence threshold and a review queue for anything below it.

Success measured in demos, not hours

If nobody measured the time the process took before, nobody can prove the automation helped. Fix: a two-week baseline before a line of code.

Ownership that ends at launch

Automations drift as the business changes. Fix: a named owner, a monthly review of the exception queue, and monitoring that pages someone when volume or error rates move.

• Baseline the hours first.
• Automate rules, route judgement.
• Surface confidence; never hide it.
• Give it an owner past launch.

Do those four and the tooling choice becomes the easy part.