Where to start with AI: the two-hours-a-day problem

Ask someone what they want AI to do and they usually name a tool. They have seen a demo, a competitor mentioned a product, someone on the team has a login they like. So the project becomes "roll out the tool," and three months later the tool has eleven seats and two people who actually use it.

That order is backwards. The tool is the last decision, not the first.

Start with the work, not the software

Most people we talk to lose somewhere around two hours a day to work a machine could do. Not the interesting parts of their job. The other parts. Copying numbers between a spreadsheet and a system that will not talk to it. Rewriting the same status update for three audiences. Reading a long thread to pull out the one decision that got made. Formatting a document so it matches the last one.

None of this is dramatic on its own. Twenty minutes here, fifteen there. But it adds up to a real number, and that number is where the value is. You cannot find it by looking at AI products. You find it by looking at how people actually spend their week.

So before anything else, do the boring thing. Have everyone keep a rough log for a week. What did you do, and how long did it take. You are not trying to be precise. You are trying to see the shape of where the time goes.

Score the work two ways

Once you have a list, two questions sort it fast.

First: how much time would this save if it went away? A task that happens twice a year is not worth automating no matter how annoying it is. A task that happens forty times a week is worth a serious look even if each instance is small.

Second: how hard is it to hand off? Some work is messy in ways that are easy to describe and easy to check. Pull these five fields out of an email and put them in this format. A person can verify that in seconds. Other work depends on context you cannot easily write down, or carries a cost when it goes wrong that you would not want a machine deciding on alone.

Plot your list against those two axes and the order picks itself. High time saved, easy to hand off, low cost of error. That is where you start. Not because it is exciting, but because it works, and a thing that works builds the trust you need for everything after it.

Why assessment beats tool-shopping

When you lead with a tool, you are betting that the thing you bought happens to fit a problem you have not defined yet. Sometimes it does. Usually it half fits, and you spend the next quarter bending your process to match the software instead of the other way round.

When you lead with the work, the tool becomes obvious. You know exactly what the job is, how often it runs, what good looks like, and what happens when it goes wrong. Picking software for a problem that specific is a much easier decision, and you can tell within days whether it is working because you defined the target before you started.

There is a quieter benefit too. The first win sets the tone. If your first project is the flashy one everyone is excited about and it stumbles, people decide AI is not ready and go back to their old way of working. If your first project is the unglamorous one that quietly saves an hour a day and never breaks, people start bringing you the next idea themselves. Momentum is worth more than ambition in the early going.

The one thing to do this week

Pick a single task. Make it one you can describe in a sentence and check in a glance. Run it for two weeks alongside the way you do it now, so you can compare. Keep the human in the loop on every result while you learn what good looks like.

You are not trying to transform anything yet. You are trying to prove, to yourself and the people around you, that this works on something real. Everything else gets easier once you have done it once.