When AI is not the answer: fixing workflows first

The signs that AI is the wrong tool

A workflow that is slow for structural reasons will stay slow with AI added to it. The signals are usually visible within a few conversations: two teams own the same step, approvals are undefined, the same information is entered in three places, or the process changes depending on who is on duty.

In those situations the problem is not a lack of intelligence. It is a lack of clarity, and clarity cannot be generated by a model.

What usually works better

The fixes are unglamorous and they tend to hold.

• Remove a step that exists because it always has.
• Agree on one source of information instead of reconciling several.
• Make ownership explicit: who decides, who approves, who is informed.
• Automate the predictable handoff with a simple rule before considering anything smarter.
• Standardise the few cases that repeat, and leave the exceptions to people.

A quick test before choosing technology

Ask three questions. Can you describe the workflow as it actually happens today, including the exceptions? Can you point to where the time or errors concentrate? Would the process still make sense if the tool you are considering did not exist?

If any answer is no, the first investment belongs in the workflow itself. Once the process is stable and understood, AI becomes a much smaller, safer decision, and it becomes obvious where it would genuinely help.

Why we are comfortable saying no

Recommending a simpler fix costs us a project in the short term and earns trust in the long term. More importantly, it is what the evidence from the workflow points to.

Technology should solve a real problem and fit the way people actually work. Sometimes the honest answer is that the problem is not technological yet.