Where should a small business start with AI?

You can do this without settling your company's entire AI strategy. Briu's existing starting guide describes two routes: practical coaching for people learning the tools, and implementation for a workflow whose rules are already understood. Either can stand alone. This exercise is for the person who wants a useful first attempt.

The takeaway is a short task description and a review rule you can use at work. The example below is a practice scenario, not a customer case or a promise of saved time.

Choose something you can judge

Imagine you coordinate work for a small service business. Each week, you turn internal project notes into a handoff for the next person. For this exercise, choose one completed project and create a fictional version of its notes. Remove identifying details and replace the facts with invented ones before using a tool approved for that material.

Your task is to prepare a handoff draft. The result needs to show the work requested, what is finished, what remains unresolved, and who needs to make the next decision. You already know what a good handoff looks like, so you have a basis for checking the attempt.

You could instead choose a document comparison or a draft answer to a recurring question. Keep the first exercise small enough that you can compare the entire output with the supplied material. If reviewing it requires expertise you do not have, choose another task or involve the person who does.

The operating leverage map offers a broader set of places to look. For this first attempt, use three questions: does this task recur, does improving it matter, and can someone clearly own the result? Treat the answers as a way to select an experiment, not as proof that automation belongs there.

Describe the result before writing the prompt

Write a sentence that a colleague could act on. For the practice handoff, try: "Prepare an internal handoff from these fictional notes so the next person can see completed work, open questions and decisions needing review."

Then describe the shape of the output. Ask for four short sections with those headings. Require a reference to the relevant note beside each factual statement. Tell the tool to mark a missing detail as unknown and to place conflicting details together for review.

Choose a stopping point. In this exercise, the work ends at a draft in the conversation. It does not include changing a project record, assigning another person work, or sending a message. That boundary makes the task understandable to the person practicing and to the person reviewing it.

You now have something more useful than "help with project management." You have a result, source material, output format, and endpoint. Keep these in a small task card so another person can repeat the same exercise.

Choose the information deliberately

Before putting real work into any tool, establish which account and information your company permits. The coaching page describes using approved material, redacted examples, or synthetic data within company rules. Follow that approach for the exercise, and resolve uncertainty with the person responsible for your company's information. See how Briu prepares coaching sessions.

Do not make account selection part of a guessing game during the attempt. Record which tool and account you use and what material belongs in it. If that decision is unresolved, prepare a wholly invented practice example and keep the real records outside the exercise.

For the fictional handoff, give each note a simple label such as A, B, or C. Include one deliberately missing date and two notes that disagree about whether an item is complete. You are designing a practice case that gives the reviewer something specific to inspect.

Keep the source set short. A first exercise should let you examine the relationship between the input and every important statement in the output. More documents can come later if the task earns another attempt.

Make the review rule specific

"Check the answer" leaves too much for the next person to interpret. Write a rule that names a reviewer, a check, and a consequence.

For this exercise: "The project owner checks every completed item, date, and proposed next action against the supplied notes. Missing or conflicting facts stay open. The draft is not used as a team handoff until that review is complete."

Make the reviewer practice the check. Can they find the note behind each statement? Does the draft describe a suggestion as though someone has agreed to it? Does it call work complete when the notes only say it is planned? Does it make the missing date visible?

Record corrections beside the draft. If the same correction appears again, revise the task instructions or source material and try the practice case again. Avoid treating a polished paragraph as the acceptance standard. The exercise succeeds when the output meets the checks you chose.

Run one comparison you can explain

Before the assisted attempt, prepare the fictional handoff yourself. Record the time you spend reading, organizing, and checking it. Then run the same practice task with AI assistance and record preparation, generation, review, and correction separately.

These are exercise instructions, not a research design that establishes a general productivity result. Working on the same material twice introduces familiarity. Keep that limitation beside the result. The comparison helps you decide what to investigate next; it does not establish what your business will save.

Write down what the assisted draft gets right and what requires attention. Perhaps the headings are useful but the reviewer has to rewrite the next actions. Perhaps the output is accurate but supplying the notes takes longer than making the handoff yourself. Record either outcome without adjusting the standard to reward the tool.

Briu's starting guide recommends evaluating quality alongside time and other workflow measures. Use that discipline here: a faster attempt does not pass if it fails your review rule. See the measurement approach.

Decide what earns a second attempt

After the comparison, choose a concrete next action. You might keep the task card, revise the source instructions, try another fictional case, or stop using AI for this task. Write down why so the next attempt has a purpose.

If you continue, choose a case that differs in a meaningful way. For the handoff exercise, try notes with no open questions, then notes with an unresolved contradiction. Keep the review rule consistent. These are proposed practice cases, not a claim that passing them certifies a production workflow.

When the exercise moves to approved real work, keep the same owner and record the extra preparation and review it requires. Define the trial's scope before expanding to more people or connecting systems. The owner should be able to describe what is accepted, what needs correction, and what stops the work.

A useful repeated task may deserve better instructions, a template, or a software connection. You do not need to decide that during the first session. Bring the evidence from actual use to the next decision.

Leave with a task and a review rule

Before you close the exercise, complete your task card: the outcome, owner, permitted inputs, output shape, review checks, stopping point, and next trial. Add the date and the questions that remain open. Another person should be able to read it and understand what you intend to try.

For help practicing on your team's work, explore AI coaching. Coaching can focus on a recurring task, checking results, and agreeing the next step. A software build is optional.

If you organize a business group, the practical AI session resource explains the available formats. Use this exercise to frame the discussion: each participant chooses one task and writes the review rule before leaving.