Part 1: Start here

What does your first 30 days with AI look like?

Written lesson · about 3 minutes to read

The short answer

Your first 30 days are one small test: pick one task, one owner and today's numbers in week one, run it with a person checking every output in weeks two and three, then compare and decide in week four. The decision to keep, widen or stop is what turns a trial into a result.

Three things to remember

  1. 01Week one: pick one task, name one owner and write down today's numbers.
  2. 02Weeks two and three: a person checks every output, and you fix the causes of mistakes.
  3. 03Week four: compare with the starting numbers, then keep, widen or stop.

Week one: choose, name and measure

Pick one task that repeats every week and has a clear right answer, such as answering new enquiries, drafting quotes or chasing overdue invoices. Name one owner, a person who knows how the job is really done and has a few hours set aside, not a committee. Tell the people who do the task what the tool is for and what is not going to change.

Then write down today's number. Time the task on real jobs, count how often it goes wrong, and note how fast you answer. Take it from the records, not from memory, because memory bends toward whoever is paying. Also write down what a bad output looks like, and what result would make you stop.

Week two: try it, with a person checking every output

Set the tool up for a small group of real users on real work, including the messy jobs and not only the tidy examples. A person checks every output before it matters. Keep a simple list of every correction: what was wrong, and why you think it happened. The list matters more than the drafts, because it tells you what to fix.

Week three: fix the instructions and the inputs

Read the correction list. Most mistakes come from two places. Either the instructions were unclear, so add the rule, an example, or what the tool must never say. Or the inputs were poor: an out-of-date price list, a missing field, a scan nobody can read. Fix those, then run it again. The share of outputs a person has to correct, called the correction rate, should fall. If it does not, that is worth knowing.

Week four: compare and decide

Put the new numbers next to the starting ones: time, errors, response time and how often people fixed the output. Check whether the people who have the tool actually use it each week, and ask them plainly how it is going. Hours saved is the weakest evidence. Better signs are quotes going out the same day, invoices paid sooner, or an enquiry answered at nine at night.

Then decide, in writing: keep it, widen it to more people or another task, or stop. Stopping is a good result, and it cost you a month, not a year. Put a review date in the calendar.

What to do today

Choose the task and name the owner. Then write down how long that task took the last three times it was done, and who checks it. That single number is the start of your 30 days.

Next step

Where does this fit your business?

Six quick questions about your business, then your AI plan: what to start with, what a person still checks, and the lessons that go with it. Or bring it straight to a free thirty-minute call.