Faster to the questions, not the answers.

I turned a spreadsheet that had beaten a team for weeks into something useful in minutes recently, but the interesting part isn't the time it saved, it's what happened next.

An organisation I'm involved with needed to compare the hours it had delivered against the hours it had billed. Simple in principle, but the data was spread across multiple sites and ran to thousands of lines, with multiple entries for the same people and none of it formatted in a way you could work with, so it was almost impossible to do by hand.

The team had managed it once, for a single site and a single month, and even that took weeks of pulling figures off the system by hand. Replicating it across every site, and keeping it up to date, simply wasn't realistic.

So we tried AI on it, mindful of using it responsibly given it was real operational data. It took those thousands of lines, sorted them by location, and rebuilt them into the same format the team had created by hand, then did it for every site rather than just one. What had taken weeks took minutes.

Once the data was usable, it started flagging where the figures didn't line up, where the hours delivered and billed didn't quite match. Things worth a closer look that, buried in an unusable spreadsheet, nobody would ever have seen.

But this is where the human comes back in. When the team worked through what the tool had flagged, some of it was down to something no AI could have known. The invoicing ran in four-week blocks while the hours were recorded by calendar month, so many of the mismatches weren't problems at all, just two different ways of measuring time that were never going to line up. The tool could see the numbers didn't match, but it couldn't know why, because it didn't understand how the organisation actually works. That took people who did.

And that's the thing I keep landing on. The team would have got there eventually, but only after days just getting the data into a usable state. The AI didn't find the answers, it cleared away the work standing between the team and the point where the real thinking could begin.

Two things stay firmly with the human. The first is accuracy, because AI isn't right by default, it can misread or quietly get something wrong, so you have to stand behind what it produces before you rely on it.

The second matters even more. Data on its own is worth nothing, it's the decisions that come from it that create the value, whether that's investigating a mismatch, changing a process, or having a conversation with someone. The AI can lay it all out in front of you, but it can't make those calls and it doesn't carry the consequences, you do. You're accountable for the numbers you present, and more importantly, for what you choose to do about them.

Used that way, as something that gets you to the questions faster rather than answering them for you, it's one of the most useful things I've come across.