Before putting AI into a process, there's a more important question

The question people usually ask is where to apply artificial intelligence. It's the wrong question to start with — or, at least, a premature one.

The question that's usually missing is simpler and more uncomfortable: is this process, as it exists today, good enough to deserve being faster? Or does it have flaws that, once sped up, will just happen more often?

Automating doesn't fix. It amplifies.

AI and automation do one thing very well: run a process faster, with less human effort. What they don't do is decide whether that process was right to begin with.

A process with a poorly designed step, an unclear decision point, or an inconsistent rule keeps that same problem after being automated — except now it happens more often, faster, and with fewer people around to catch it before the mistake reaches the customer.

The most common pattern

• The slowest or most repetitive process gets picked to automate first — not necessarily the best-defined one
• Automation gets built on top of the existing process, exceptions and inconsistencies included
• Mistakes that used to be rare and caught in time become frequent and reach the end of the line

None of this means automating is a mistake. It means order matters: automating a confusing process doesn't make it clear, it just makes the mistake happen faster.

Automating isn't always the first step. Removing what's unnecessary — and fixing what's poorly defined — usually is.

This isn't an argument against AI or automation. It's an argument for the right order: observe the process as it actually happens, find where the inconsistency lives, decide what's worth keeping human — and only then decide what to speed up.