The 30-Day Rule: One AI Tool, One Metric at a Time

Most owners do not have an AI problem. They have a too-many-trials problem. Three free accounts, two browser extensions, one paid plan nobody remembers signing up for, and no clear answer to the only question that matters: did any of it give you time or money back?

The rule we use is short. Pick one tool. Put it on one job. Track one number for 30 days. At the end, keep it, fix it, or cut it. Only then add the next one.

This post explains why the rule works, how to run it, and what to do at day 30.

Why one at a time

The research on failed AI projects is mostly about large companies, but the failure patterns translate directly to a ten-person business.

RAND interviewed 65 experienced data scientists and engineers about why AI projects fail. Their report, The Root Causes of Failure for Artificial Intelligence Projects, names five leading causes. Two of them matter most for owners: teams misunderstand or miscommunicate what problem the AI is supposed to solve, and organizations focus on using the latest technology instead of solving a real problem for the people doing the work. One of RAND's principles for success is to focus on the problem, not the technology. Interviewees also described leaders who switch priorities every few weeks or months, so projects get discarded before they can show results.

Fortune's write-up of an MIT report on generative AI pilots (150 leader interviews, a survey of 350 employees and an analysis of 300 public deployments) found that about 5% of pilots achieved rapid revenue acceleration while most stalled with little measurable impact. Describing the young startups that were succeeding, the report's lead author said they "pick one pain point, execute well, and partner smartly." The same piece reports that the biggest returns showed up in back-office work, not the flashier sales and marketing tools that took most of the budget.

Small businesses describe the same friction in their own words. In Simply Business's 2026 SME Insights Report (1,842 UK owners), 39% said not seeing a clear use for AI was a barrier, second only to security and privacy concerns at 44%. In the Goldman Sachs 10,000 Small Businesses Voices survey, owners named difficulty choosing tools as one of the barriers to deeper use.

None of that is solved by trying more tools at once. It is solved by naming one job and checking whether the tool did it.

Step 1: Pick the job before the tool

Start from a task, not a product. Good first jobs share three traits:

They happen every week, so 30 days gives you four or more repetitions.
You can describe the output in one sentence ("a reply draft for every new quote request").
A mistake is cheap to catch before it reaches a customer.

If you do not have a shortlist yet, run the one-week audit first. It ends with a ranked list of the work costing you the most, which is exactly what this rule needs as input.

Then choose the tool that fits the job. Often it is one you already pay for. The rule is about focus, not about buying something new.

Step 2: Choose one number, and write down today's value

One metric keeps the trial honest. Pick something you can count without a dashboard:

Minutes spent on the job per week (time it with your phone for a week before you start)
Items handled per week (quotes answered, invoices sent, posts drafted)
Turnaround time (hours from request to reply)
Rework (how many outputs you had to redo or heavily edit)

Write down the starting value before day one. Without a baseline, day 30 becomes a feeling, and feelings about new tools run hot in week one and cold by week three.

Avoid metrics you cannot attribute. "Revenue went up" over 30 days has too many causes. "Quote replies now go out the same day" is something the tool either helped with or did not.

Step 3: Set the rules of the trial

Write these on one page and share them with anyone involved:

The job, in one sentence.
The tool, and who uses it.
The metric and its starting value.
What the tool may do on its own, and what needs a person to approve. For anything customer-facing, start with "drafts only, a person sends."
The end date, 30 days out, on the calendar.

That fourth line is where privacy concerns get handled in practice. Decide up front what data the tool can see. If it does not need customer records to do the job, do not connect them.

Step 4: Run it properly for 30 days

The most common way a trial fails is that it never really runs. The tool sits in a tab, gets used twice, and the verdict at day 30 is "it didn't help."

Three habits prevent that:

Use it on every instance of the job, not just the easy ones. You need the awkward cases to know whether it holds up.
Keep a running note of misses: what it got wrong, how long the fix took. Two lines a day is enough.
Adjust the instructions once a week, not every hour. Weekly changes let you see whether a change helped.

Do not add a second tool during the 30 days, even if a newer one launches. The priority-switching problem RAND's interviewees described applies at any size.

Step 5: Decide at day 30

Measure the same number the same way you did at the start, then make one of three calls.

Keep it if the number moved in the right direction and the misses are rare and cheap. Write down how the job now runs so it does not depend on one person remembering the setup.

Fix it if the number moved a little but the misses are eating the gain. Usually the fix is tighter instructions, better inputs, or narrowing the job. Give it one more 30-day round, no more.

Cut it if the number did not move. Cancel the plan, remove its access to your accounts, and write one line on why. That line saves you from re-trying the same idea in six months.

Then, and only then, pick the next job from your list.

What this looks like over a quarter

Three rounds in a quarter means three decisions you can stand behind, instead of a dozen half-used subscriptions. Some rounds will end in a cut. That is the rule working: you learned something in a month instead of paying for it for a year.

The pace feels slow at first. It is slower than signing up for everything. It is much faster than finding out a year later that nothing stuck.

Where we come in

Running these 30-day rounds alongside owners is the core of our embedded AI partnership: we help pick the job, set the metric, build the workflow and make the day-30 call with you. If you want help choosing your first round, book a free session and bring your list.

Related service: embedded AI partnership.

Get our notes on AI and how businesses run in your inbox: subscribe to The Beyond Brief.

The 30-day rule: one tool, one metric, before you add another