The AI Was the Easy Part

Something happened recently that changed the way I think about AI implementation.

I was working on improving a relatively simple business process. The information was being managed through a spreadsheet, and the objective seemed straightforward: make it easier for people to understand what was happening and manage the work more effectively.

Initially, it looked like a technology problem.

It wasn't.

Most of the work happened before the AI

Before I could build anything useful, I had to understand the process.

What was the spreadsheet actually telling us? What information mattered? What patterns existed? What did people need to know? What happened at each stage? What triggered the next action? What information needed to be structured? What was simply there because it had always been there?

That took time.

There was investigation, questioning, reasoning, restructuring and quite a lot of going backwards before moving forwards.

Gradually, though, the mess became a process I could understand.

Once I understood the process, I could design the information around it properly and move it onto technology the organisation already owned.

And then something interesting happened.

The AI became the easy part.

Give AI something good to work with

Once the underlying information had structure and context, applying AI to it became remarkably straightforward.

Suddenly we could start asking questions of the information conversationally. We could summarise it, reason across it and surface things that previously required somebody to manually work through rows and columns.

And we'd barely scratched the surface of what was possible.

What struck me wasn't simply what the AI could do.

It was how much more useful AI became because of the work we'd already done.

I spent far more time making the information understandable than I spent making AI useful. Once the first was done, the second happened remarkably quickly.

Maybe that's where the real work is

We talk constantly about implementing AI, but perhaps some of the most important AI work doesn't look like AI work at all.

It's understanding a process.

Cleaning up information.

Creating structure.

Deciding where information belongs.

Questioning why something is done a particular way.

Using the enterprise technology already available to create a better foundation.

None of that is particularly glamorous compared with an impressive AI demonstration.

But once that foundation exists, AI has something meaningful to work with.

And that's when things can become really exciting.

Sometimes the hardest part of implementing AI has nothing to do with AI.