A few minutes into my conversation with Anders Dræge, I realized something slightly embarrassing.

He has won awards for his AI work.

The last award I remember winning was for school electronics, somewhere around eighth grade.

And the difference between us isn’t the technology.

Anders leads AI development at Frende Insurance in Norway. His AI systems run in production, and together they save over 400 human hours a month.

His answer was almost disappointingly simple.

• Ask the people who do the task today how long it takes, once. Minutes, sometimes seconds.
• Count how many times the process runs.
• Multiply.

That’s the whole method. He borrowed it from the company’s robotic process automation team (RPA: software robots that click through the steps a person would), who have counted saved hours that way since 2014.

Some tasks are tiny but happen thousands of times. Others are slow because a person has to dig through several systems to find one answer. Both add up, and both end as one number a leader understands: hours per month.

It took me straight back to a meeting from my consulting years.

We were building email classification for an insurance client, and we told leadership it would be hard to get the AI to 100%. We were aiming for around 90.

One of the senior people almost jumped out of his chair. “If you get it to 90%, ship it today. Our manual workflow is at about 65.”

Both mistakes come from the same place: we compare AI to a promise, when we should compare it to how the work gets done today.

Here’s the uncomfortable part: Anders’ closing advice wasn’t about AI at all.

Think value creation, he said, and then be able to communicate it. When leaders see that a system saved them a number of hours every single month, AI stops being an interesting experiment and becomes a priority.

That one landed, because communicating the value is exactly the step I skip.

If you can’t show the hours, even your best AI project looks like a hobby.

Hours are the easy part to count, though. The harder call is which initiatives deserve the hours in the first place. Anders also lets AI make some decisions entirely on its own, which goes against most of the advice you’ll read on LinkedIn, so I asked him why.

That’s where the full conversation goes next.

You can watch the full conversation here: