Your First AI Project: Start With One Task You Do Every Week

Your first AI project does not need a budget line, a consultant or a new subscription. It needs one task you already do every week, a stopwatch, and four weeks of attention.

Most small businesses have not started, so you are not behind where it counts

The U.S. Census Bureau’s survey of businesses found that 19.8% of businesses used AI as of May 3, 2026. Among firms with 250 or more employees the figure was 37%, and among firms with fewer than 20 employees, use did not change significantly between December 2025 and May 2026. If you run a small shop and have not started, you are in good company: Census found that fewer than one in five of the smallest firms, those with four or fewer employees, used AI.

The Federal Reserve Bank of San Francisco heard why from owners themselves. In its July 2026 brief on the 2024 Small Business Credit Survey, nearly 40% of respondents said they used or planned to use AI. The barriers named by owners interested in AI were tool costs, limited time and capacity for staff training, and knowledge gaps about how to put AI to work. None of those barriers is solved by buying more software. A small, timed project addresses all three: it costs little, it fits in an hour a week, and it teaches you what the tool can and cannot do.

The right first task repeats weekly and is quick to check

Choose a task that meets four tests. You do it at least once a week. The result is something you can read and judge yourself. A mistake costs you an edit, not a customer. And it does not require pasting in private customer details such as account numbers, health information or payment data.

Good candidates are the weekly customer email, a first draft of a job posting, a summary of a long supplier thread, a reply to the question customers ask you most, or a task list built from your meeting notes. If two candidates tie, take the one that annoys you more; repetitive work is where a first draft helps most. These tasks match what workers already do. In a Census Bureau analysis published in August 2026, the most common ways U.S. workers used AI were searching for information or technical help (37%), writing communications, documentation or instructions (32%), generating ideas (32%), and interpreting, translating or summarizing information (31%).

Time the task first, then run the project in four weeks with these steps

Before you open any tool, time the task as you do it today. Time it a few times if you can, and write down the minutes. That baseline is the only way to tell whether AI helped or only felt fast.

Expect a range, not a guarantee. Among workers who used AI in the previous week, Census found that 25% saved less than an hour, 31% saved one to two hours, 15% saved three to four hours, and another 15% saved more than four. Ten percent saved no time, and 4% said AI added time. Those figures are self-reported by workers, which is exactly why your own stopwatch beats them.

Pick one task you repeat every week, time it before you start, and read every word the tool writes before anyone else does.

Start this week. Each step is small enough to finish in one sitting.

1. Write the task in one sentence, such as “Draft Friday’s customer email from my notes.” Time it by hand and record the minutes.
2. Pick one tool. Use the AI feature in software you already pay for, or a general-purpose assistant on a free plan. Do not buy anything yet.
3. Strip out private details. Replace customer names, account numbers and anything confidential with placeholders such as [Customer] before you paste anything in.
4. Write your instruction once and save it. Say who the reader is, the tone you want, the length, and paste in one good example from your past work.
5. Set one checking rule and write it on a sticky note: you read every word, and you verify every name, number, date and price against your own records before it leaves the building.
6. Run the task the same way each week for four weeks. Log the minutes for drafting plus checking, and note any error you had to fix.
7. At the end of week four, compare your log with your baseline and make one decision: keep it, change the instruction, or stop.

Counting the checking time matters, and so does keeping the project small. One task, one tool and one rule leave you with a result you can explain to a partner or an employee in a minute. A draft that takes two minutes to write and ten to correct has not saved you anything, and your log will show that plainly.

Buy results, not promises, when someone offers to sell you the next step

Once you have a number, you are ready for the sales calls that follow. Treat a vendor’s income or savings claim as a question, not evidence. In March 2026 the FTC announced a proposed settlement with Air AI, a company it had sued in August 2025 for allegedly telling buyers they would earn substantial income and were covered by refund guarantees. The proposed order carries an $18 million judgment, largely suspended because the company and its owners could not pay the full amount, and bans the company and its owners from marketing business opportunities. The order takes effect only after a federal judge approves it.

You do not need to be a lawyer to use that lesson. Ask any vendor for the hours saved on a task like yours, in writing, from a customer your size. Then compare the answer with your own four-week log. If the claim and the log disagree, trust the log.

Start with one task, count the minutes, and let your own four-week log tell you what to buy next.