← Home

How to start using AI in your business

Start with one recurring task, not a tool. A step-by-step way for small and mid-size teams to pick it, change it, keep a person checking the result, and measure it.

The short answer

Start with one recurring task, not a tool. Pick something your team does every week, where a person can check the result in under a minute and a mistake is cheap to fix. Fix what slows that task down, change it the cheapest way that works, and run it on real work for two weeks while you watch one number.

That’s the whole method. The rest of this page is how to do each part.

Step 1: pick one recurring task

List five to ten tasks your team repeats, in your team’s own words. “Turn call notes into a proposal” is a task. “Improve communication” is not.

Then score each one on five things: it happens often, one person can decide to change it, its inputs live somewhere a computer can read, a person can check the result quickly, and a mistake is cheap. Start with the highest score.

The full scoring, with a prompt that interviews you about your team’s week: Pick your team’s first AI task.

Step 2: fix what’s slowing it down first

Before any AI, ask what really holds the task up. Usually it’s the process (work waits for an approval nobody needs), the data (the information sits in three places) or the ownership (nobody is sure whose job it is).

Fix that first, even if the fix is a shared checklist. AI on top of a broken process gives you a faster broken process.

Step 3: change it the cheapest way that works

Weigh four options in this order:

  1. Keep the tool you have and connect it.
  2. Change the process.
  3. Buy something that fits better.
  4. Build the missing part.

Most first wins come from the first two.

Step 4: keep a person checking the result

Decide where someone reviews the output before it reaches a customer or moves money. A draft that a person checks is a safe first step. A message sent to a customer with nobody looking is not.

Step 5: run it on real work and measure one number

Write one line before you start: time per task, errors caught late, or how long someone waits. Run it for two weeks on real work, then look at that number.

The last test matters most: can someone on your team keep it running without the person who set it up? If not, it isn’t finished.

Where the AI tools fit

Mistakes that stall a first project

  • Starting with a tool. You end up looking for a problem to fit it.
  • Starting with the biggest project. Automating “sales” or “onboarding” first usually stalls. A small, frequent task teaches you faster.
  • No one checks the output. If nobody can tell quickly whether it’s right, nobody will trust it.
  • No owner after launch. If it only runs while the person who set it up is around, it will stop.

Questions people ask

Do I need a developer? Not to start. Picking the task and trying it in a chat assistant needs no code. Building the missing part, step 4 in the list above, sometimes does.

Do I need a paid AI plan? Not to pick your first task. The free Claude and ChatGPT plans can interview you about your team’s week. Claude Code needs a paid Claude plan.

How long does a first task take? When I work with a team, the first task takes four weeks: see the work, build the change into the team’s workflow, test it on real work and hand it over. How the first four weeks go.

Should I bring in someone to help? If you do, ask them which task they’d start with and why, who on your team they’ll sit with, and what your team can run without them at the end. Seven questions to ask anyone you bring in.

Have a recurring task in mind?

Tell me how your team handles it today.

Work with me