You do not fix a broken process with an AI model

I took this photo in Indonesia last week.From an AI angle, it reads like a confession.

This is what most AI projects look like in mid-market companies.

In the strategy slides (Londry) you see:

• "AI agent for end-to-end automation"
• "Autonomous decision engine"
• "Smart workflow orchestration"

In reality you have a bin with:

• No clear process owner
• Workarounds in Excel and email
• Legacy systems nobody has permission to touch
• Data quality no one names in meetings

Then AI use cases get painted on top, like the misleading arrow on the sign.

It's for the optics.No structural change.And everyone involved knows it - they just hope the model will somehow bypass the mess.

You do not fix a broken process with an AI model.You automate the breakage - faster, harder to see, harder to govern.

A few weeks in, someone in Operations sends an email:"No one knows why it's recommending this. Do we override it or trust it?"

That is when the project goes quiet.

The sequence that works - when it works:

Clean the bin first.• Map the real process, not the one in the manual.• Remove the waste, the email workarounds, the tribal knowledge that lives in two people's heads.

Fix the flow.• Clear ownership. Clean inputs. Documented decisions.• Boring, slow, occasionally humiliating when you realize how long the workaround has been running.

Then add AI - only where it actually solves something.

This takes longer than your deck promises.It requires admitting what is broken.Most leadership is not structurally ready for that conversation.

The question is not whether your leadership team is smart enough to see the bin.

It is whether they have the structural freedom to admit it is there.