The first automation sets the tone for everything after it. Pick one that pays for itself quickly and your team will ask for the next one. Pick one that is too ambitious and AI becomes the project nobody wants to talk about.
Start with a list, not an idea
Ask the people who do the work to list every task they repeat each week. Not the managers, the people doing it. You will get a long, unglamorous list: copying order details, answering the same five questions, chasing unpaid invoices, building the Monday report. That list is where the value is.
Score each task on four things
Give each task a score from 1 to 5 on each of these:
- Volume. How many times a week does it happen? More is better.
- Time. How long does each one take? Longer is better.
- Rules. Could you explain how to do it well to a new hire in a page? Clearer is better.
- Risk. What happens if one is done badly? Lower is better for a first project.
Add them up. The top three are your candidates.
Try it with your own tasks: rename a row, move the sliders, and the best candidate rises to the top. The example rows are for illustration.
Check the data is reachable
A great candidate is useless if the information is locked away. For each of your top three, ask where the inputs live and where the outputs need to go. If both ends are tools with an API, like a CRM, an inbox or accounting software, you are in good shape.
Start small, then widen
Automate the most common version of the task first and send everything unusual to a person. A system that handles most cases and hands off the rest is worth far more than a perfect one that never ships. Once it runs, look at what it handed off and teach it the next most common case.
Decide how you will measure it
Before you build, write down how you will know it worked: hours saved per week, response time, error rate. Measure the current numbers now, because nobody remembers them later.
This is exactly what our one week audit does, with your team and your real data. See how we work.