What to automate first: a one week audit you can run yourself

Every few weeks someone messages me with the same opening line: "We want to automate everything." I understand the feeling. When a team is buried in copy and paste, everything looks like a candidate.

But the first automation matters more than the next ten. If it saves real time, people trust the second one. If it breaks or saves twenty minutes a month, the whole idea quietly dies.

So before we build anything, I ask for one week of notes. No software, no workshop. Just a log.

The log

Give everyone who touches the work a shared sheet with five columns:

1. Task, in plain words. "Copy new orders from email into the stock sheet."
2. Minutes, roughly, each time it happens.
3. How often, per day or per week.
4. Tools involved, every app the task passes through.
5. What goes wrong, if anything. Typos, late replies, forgotten follow ups.

People forget the small things, so ask them to add a line the moment they finish a task, not on Friday afternoon.

Score it

At the end of the week, multiply minutes by frequency. That gives you hours per week for each task. Then sort the list.

The top of that list is rarely the task people complained about most. It is usually something dull that happens forty times a day and takes three minutes each time. Nobody notices three minutes. Everyone notices two hours.

Next to the hours, give each task a simple yes or no on three questions:

• Does it follow the same steps every time? If a person has to make a judgement call on most runs, it is a job for an AI agent with an approval step, not a plain workflow.
• Do the tools have a way in? An API, an export, an email inbox, even a shared folder. If the data only lives on paper, start with the paper.
• Would a mistake be cheap to fix? For the first automation, pick something where an error is an annoyance, not a refund.

The task with the most hours and three yes answers is your first build.

What it usually turns out to be

In our own projects the winner tends to fall into one of a few families:

• Moving data between tools: orders into a spreadsheet, form leads into a CRM, invoices from email into accounting.
• Reminders and follow ups: quotes nobody chased, invoices nobody nudged.
• Reports someone builds by hand every Monday from four different dashboards.

None of these are glamorous. That is the point. They are predictable, so they can run by themselves, and they are frequent, so the time adds up quickly.

Set a payback line before you build

Once you know the task, work out what it costs today. Hours per week multiplied by the hourly cost of the person doing it. Then compare that with the price of building it; our pricing page lists what typical workflows cost.

If the automation pays for itself within a few weeks, build it. If it takes a year, either the task is smaller than it looked or the build is bigger than it needs to be. Both are worth knowing before anyone writes a line of code.

Keep a person in the loop at first

For the first month, let the automation do the work and let a person glance at the result. Not because it will fail, but because the team needs to see it working before they stop checking. After a few weeks of clean runs, the checking stops by itself.

If you run this log and want a second pair of eyes on it, send it over. The automation audit is free, and you will get a written plan and a fixed price either way.