No. 02

AI for Small Business, Minus the Hype: A 30-Day Pilot That Won't Waste Your Money

A practical, four-week plan for small business owners to test AI on one real problem, measure if it helps, and decide with evidence. Includes what to try first, what to never hand over, and how to measure results.

The Tea the short version

  • Only about one in five U.S. businesses use AI, and the smallest businesses aren't catching up yet.
  • Don't 'adopt AI.' Pick one annoying, repeatable task and test AI on that.
  • Measure before and after: time spent, errors, and how customers feel.
  • At day 30, make one of three calls: keep it, change it, or kill it.

Let's skip the hype and talk like business owners.

You don't need "an AI strategy." You need your Tuesdays back.

Where small businesses actually are

Here's the real picture. In early 2026, about one in five U.S. businesses reported using AI in any business function, according to the U.S. Census Bureau's Business Trends and Outlook Survey. Between December 2025 and May 2026, AI use grew among firms with 20 or more employees, but didn't change significantly among firms with fewer than 20 (U.S. Census Bureau).

And among businesses that do use it, most keep it narrow. A Census Bureau working paper found that 57% of AI-using firms use it in three or fewer business functions, most often sales and marketing, and 66% use it only to help people do tasks, not to replace them (Census Bureau working paper).

That's actually the right instinct. Start small. Help people. Expand only what works.

Before you start: is the floor ready?

AI works best on top of organized information. If your customer list lives in three places and nobody agrees on the numbers, start there first. I wrote a whole guide on that: Get Your Data House in Order Before You Buy AI.

If your basics are in decent shape, let's run a pilot.

Week 1: Pick one task and measure it

Choose one task that is:

  • Repeated at least weekly
  • Annoying enough that someone complains about it
  • Low-risk if the first draft is imperfect, because a human will review it

Good first candidates I see in my classes:

  • Drafting replies to common customer emails
  • Turning messy meeting notes into a clean summary and to-do list
  • Writing first drafts of product descriptions or social posts
  • Summarizing long documents, like a lease, a grant notice or a vendor contract, for your own understanding (not as legal advice)
  • Cleaning up a spreadsheet of contacts

Then measure the "before." For one week, track:

  1. Time: how long the task takes now
  2. Errors: how often it needs fixing
  3. Feeling: a simple 1-to-5 rating from whoever does it, and from customers if they're involved

Write it down. This is your baseline. Without it, you're just guessing.

Week 2: Set your guardrails

Before anyone types anything into an AI tool, agree on the rules. Write them on one page:

Never paste in:

  • Customer Social Security numbers, card numbers or bank details
  • Health information about customers or staff
  • Passwords or login codes
  • Anything a customer told you in confidence

Always:

  • A human reads it before it goes out
  • Check facts, names, prices and dates. AI makes things up confidently
  • Use a business account with data-training turned off when the tool offers that setting

Who's in charge: One person owns the pilot and decides what's okay.

Week 3: Run it for real

Now use the tool on your chosen task, every time it comes up, for a week. Keep tracking the same three numbers: time, errors, feeling.

Tips that make a big difference:

  • Give it context. "Write a reply" gets generic mush. "Write a warm, short reply to a regular customer whose order is two days late; offer 10% off next time; sign it from Maria" gets something useful.
  • Save your best prompts. When something works, save it in a shared doc so the whole team gets the same quality.
  • Note the failures. When it gets something wrong, write down what. Patterns will appear.

Week 4: Decide with evidence

Compare week 1 to week 3. Then make one of three calls:

If...Then...
Time dropped, errors stayed flat or improved, people felt betterKeep it. Write it into your normal process
Time dropped but errors went upChange it. Add a checklist step or narrow the task
No real time savings, or people hated itKill it. That's a win too. You just saved yourself a subscription

Then pick the next task. One at a time.

What not to hand to AI (yet)

  • Decisions about people. Hiring, firing, pay, who gets service. Keep a human there, always. More on why in Your Boss's Algorithm Now Needs a Human.
  • Money movement. Don't let any tool send payments or approve invoices on its own.
  • Anything with legal or medical stakes, without a professional reviewing it.
  • Unsupervised access to your accounts. If you're tempted by AI agents that act on your behalf, read Who Holds the Kill Switch? first.

The real goal

The point isn't to use AI. The point is to run a business you love, with more time for the parts that only you can do: the relationships, the craft, the community.

If AI helps with that, great. If it doesn't, you're not behind. You're being smart.

Pour me another: What's the one task you'd hand off tomorrow if you could? Tell me through the contact form. I might build a class around the most common answers.

Sip slow, stay curious.D

Receipts 2 sources cited in this post
  1. U.S. Census Bureau census.gov
  2. Census Bureau working paper census.gov
Dei’Marlon “D” Scisney

Your host

Dei’Marlon “D” Scisney

Founder & Principal Data Engineer, The Data Guys™

Data engineer, teacher and civic technologist. Vice-Chair of Seattle’s Community Technology Advisory Board. Raised by his grandmother in Evansville, Indiana; building community technology centers in Seattle’s Central District and back home in Evansville.

I am not a lawyer, and nothing here is legal advice. Facts were current when this was written, and every claim links to its source. Spot something that has changed? Tell me and I’ll update it.

Another cup?

Keep sipping.

Pour me another

Get the tea before everyone else.

New posts on data, AI, privacy and the people tech forgets, plus The Tea Report, straight to your inbox. Written by me, never sold, never shared.

Free. No spam, ever. Unsubscribe in one click. Prefer LinkedIn? Subscribe there.