AI ideas are often symptoms, not opportunities

When you ask a team where AI might help, the answers can tell you a great deal about what is frustrating them. People want help checking documents, finding information, preparing reports or moving the same details between systems.

Those are useful starting points. I’m less convinced that they should go straight onto a delivery roadmap.

The distinction I keep coming back to in AI diagnostic work is between the task someone wants help with and the reason they are having to do it.

Imagine a team asking for AI to check documents before they go to clients. That might be a good application. But suppose the errors begin with an intake form that lets essential information go missing, or with two systems holding different versions of the same client details. The checking request has revealed a problem further upstream. Improving the inputs could remove some of the work entirely.

This is why I think AI ideas are often symptoms. They show us where people are compensating for something that isn’t working, although the solution may still involve AI.

There is a strong practical objection. Organisations inherit systems they cannot easily replace, responsibilities they cannot quickly reorganise and processes they are obliged to follow. Helping a team check documents this month may be more valuable than proposing a redesign that takes a year.

I don’t think the answer is to insist on fixing every underlying problem before doing anything useful. I do think we should understand what we are buying: a temporary reduction in effort, a lasting improvement to the process, or another dependency that someone will have to maintain.

That changes how I would assess the opportunity. Alongside the possible time saving, I’d want to know whether the task should still exist and what would need to change for it to disappear.

When is automating a workaround sensible pragmatism, and when does it make a broken process harder to replace?