Audit before tools: how to decide what to automate first

I watched a team buy three AI tools before a single person had written down what they actually did all week.

Six weeks later they were paying for all three and using none of them properly. Nobody had decided what the tools were for.

This is the most common way AI money gets wasted, and it costs nothing to avoid. The fix is an exercise that takes about thirty minutes and needs no software at all.

Step 1: List every task you did more than twice last week

Open a blank page and write them down. Be unglamorous about it: "pull the numbers into the Monday deck", "chase four people for approvals".

Not "strategy". Not "comms". Those are job titles, not tasks.

You're looking for work that has a shape, because shape is what a machine can copy.

Step 2: Put a time against each one

Rough is fine. Round up rather than down, because everyone underestimates the small stuff.

Most people get to somewhere between fifteen and twenty hours of genuinely repetitive work before they run out of page.

Step 3: Sort every task into three buckets

This is the part that does the work.

• AI-only. The task has a fixed input, a fixed output, and you can tell within seconds whether it came back right. Formatting, first-draft summaries, pulling fields out of documents, sorting incoming messages.
• Human + AI. You still make the call, but a machine does the fetching, drafting or sifting. Most research, most first drafts, most analysis.
• Human-only. Judgement calls, relationships, anything irreversible, and anything you'd have to read line by line to trust.

The third bucket is not a failure. Protecting it is the point.

The sort is where it gets uncomfortable, and that's the useful part. The tasks that feel busiest are usually the automatable ones, and the small ten-minute tasks are often the ones that actually needed you.

What the sort looks like in practice

Here's a typical list from a marketing manager's week, to show the shape of it:

The two AI-only rows add up to three and a half hours. That's the starting point, and neither needs anything more advanced than the tools most teams already pay for.

Step 4: Pick three, not fifteen

Take your three biggest AI-only items by time. That's your list.

Not the tool everyone on LinkedIn is talking about this week, and not all fifteen things at once. Start with the one you do most often, because frequency beats size almost every time.

If you build one of them and it saves you forty minutes a week, you've already done better than most companies that bought licences first. They spent a lot more than you did.

Doing this for a team? Add a fourth column

On your own, three buckets are enough. Across a team or a whole organisation, one more column saves a lot of pain later: a risk flag.

For each task you plan to automate, note whether it touches:

1. Sensitive data, such as customer details, financial records or employee files
2. Anything a customer sees before a person has checked it
3. A regulated process, if you work in finance, health, legal advice or the public sector

A flag doesn't mean "don't automate". It means a person stays in the loop, and someone decides in writing how the output gets checked.

This is the column most teams skip, and it's the one compliance asks about first. Writing it down at the start is far cheaper than adding it after the first awkward email.

What to do this week

Do the list before you look at a single tool. Thirty minutes, a blank page, a time against every line.

You'll know your top three before you reach the bottom of the page.