The goal is proof, not a grand plan
The first month should answer one question: can a small AI system improve a real piece of work here?
It is not the month to connect every tool or study the whole company. Clear limits make the result easier to trust.
Week 1: Choose the work
Pick one job that:
- matters to a named person;
- happens often;
- has source material you may use;
- has a result a person can check;
- gives the saved time somewhere useful to go.
Keep a recent example. Note how long it took, where it waited and who made the final call.
Week 2: Build the smallest useful version
Give the system the facts, one good example and the checks the owner already uses.
Make the first version narrow. It may prepare a brief, sort a set of notes or bring the right facts into view. It does not need to run the whole job.
Show where the AI stops and the person steps in.
Week 3: Use it on live work
Run it several times. Keep the weak outputs as well as the good ones. Ask the owner to note:
- what saved time;
- what still needed care;
- which fact was missing;
- when they did not trust the answer.
Change the system from this evidence, not from a wish list.
Week 4: Keep what worked
Compare the before and after examples. Record the time returned, the quality change and what the owner did with the time.
Leave behind:
- the working system;
- the source files and examples;
- the checks and limits;
- a short record of what changed;
- the next test, if there should be one.
The decision at day 30
Continue only if the work is useful, the owner wants it and the next gain is clear. Stop if the system needs more care than it returns.
That is a successful first month either way. You have replaced hope with evidence.