Choose work you can observe

Start with one recurring point of friction: a report assembled from scattered updates, a meeting that requires too much preparation, or a commitment that gets lost between conversations. Choose something frequent enough to test within thirty days and narrow enough that one person can own it.

Write down the trigger, the information needed, the output, and who uses it. Describe what happens today before deciding where AI belongs. “Improve our operations” is too broad. “Prepare a weekly exception report from these three approved sources” gives you something concrete to build and check.

Week one: define and measure

Complete the workflow as you do it today. Record the gathering time, the checking time, the missing information, and the corrections. Choose one measure of improvement and one boundary you will not cross. Confirm which information the system may access and who will review its output.

Week two: build the smallest useful version

Use a small set of representative examples to create the first version. Keep the scope to one input, one useful output, and one review step where possible. Run it alongside your existing process so a failed attempt does not interrupt the work.

Week three: test normal and awkward cases

Check the output against the source material. Include missing updates, conflicting notes, and unclear requests. Record mistakes and make the system flag uncertainty instead of guessing. Test the fallback as well as the successful path.

Week four: decide what has earned a place

Compare the complete effort with your starting point—including review, corrections, and maintenance. Decide whether to keep, adjust, or stop the workflow. If it helps, document its owner, sources, review routine, and fallback. Expand only after this version has earned your trust.

The Margin Finder workbook gives you space to work through one point of friction and map these thirty days. A calendar is a planning aid; a dependable result matters more than meeting an arbitrary deadline.