Skip to content

Where to Start With AI in a Small Business: The 5-Question Test

By USA AI Corp · · 7 min read

The short answer

Start with one chore, not one tool. List the repetitive tasks that eat your week, then score each on five questions: how often it happens, whether it follows rules or needs judgment, whether the information is already digital, what a mistake would cost, and who will own it. The top scorer with cheap, easy-to-catch mistakes is your first AI project.

Imagine the owner of a four-truck plumbing shop walking in from a regional trade show with a tote bag full of flyers. One promises an AI dispatcher, another an AI estimator, a third an AI that writes his social posts. Every demo looked impressive. On Monday morning he has the same question he had on Friday: which of these, if any, deserves his money and his office manager's attention?

Starting from the tools is backwards. A product answers a question you haven't asked yet. The question that matters is which piece of work in your business is the best first candidate for automation, and you can answer it yourself in an afternoon with a pen, your team and the five questions below.

Step one: list chores, not tools

Before scoring anything, write down every task that annoyed you or your staff in the past two weeks. Phrase each one as a verb and an object, the way you'd hand it to a new hire: “answer calls after 5 PM,” “send appointment reminders,” “retype the intake form into the case system.” Leave out tool-shaped entries like “get a chatbot.”

Ask the people who do the work to add to the list. Office managers, front-desk staff and dispatchers know where the minutes leak far better than the owner does. Aim for ten to twenty items. Many will be dull, and dull is good: dull work tends to repeat, and repetition is what software handles well.

The five questions, and how to score them

Give each chore 0, 1 or 2 points per question, for a maximum of 10. Don't agonize over a score. You're sorting a list, not forecasting revenue.

1. How often does it happen?

Monthly or less often scores 0. Weekly scores 1. Daily, or many times a day, scores 2. Frequency matters because setting a system up takes roughly the same effort whether the task happens five times a month or five hundred, and only one of those pays back quickly.

2. Does it follow rules, or does it need judgment?

If you could write the steps on an index card (“get the name, address and what's leaking, then book the next open slot”), score 2. If it needs some judgment but a person can check the result, score 1. If it's mostly judgment, like deciding whether a patient needs a root canal or whether a firm should take a case, score 0. AI can help around judgment work, but it makes a poor first project.

3. Is the information already digital?

Score 2 if everything the task needs lives in software, email or a shared calendar. Score 1 if some of it sits on paper or in someone's head but could be written down. Score 0 if the task depends on filing cabinets, sticky notes or one person's memory of every customer.

4. What does a mistake cost, and would you catch it?

Score 2 if an error is cheap and visible within hours, such as a reminder text sent at the wrong time. Score 1 if it's annoying but fixable. Score 0 if one error could hurt someone, lose a client or create a legal problem. Score the chore as you'd actually scope it: if every emergency will go straight to a person, rate the routine part that's left.

Don't count on blaming the software if it gets something wrong. At least one tribunal has refused to let a business blame its chatbot: when Air Canada's website chatbot gave a grieving customer the wrong bereavement-fare rules, a British Columbia tribunal ordered the airline to compensate him and rejected the argument that the chatbot was responsible for its own words (Moffatt v. Air Canada, 2024). Start where errors are small and easy to spot.

5. Who will own it?

Score 2 if a named person will check on it every week and has the authority to change how it works. Score 1 if someone is willing but stretched thin. Score 0 if the honest answer is “everyone,” which in practice means no one. Projects without an owner rarely fail loudly; they just stop being used.

Scoring sheet: copy this onto paper and score each chore
Question0 points1 point2 points
How often?Monthly or lessWeeklyDaily or more
Rules or judgment?Mostly judgmentSome judgment, easy to checkFits on an index card
Information digital?Paper or memoryPartlyAll in software or email
Cost of a mistake?Could hurt someone or lose a clientAnnoying, fixableCheap and caught fast
Owner?“Everyone”Willing but stretchedNamed, with authority to change it

Flowchart

Running one chore through the test

  1. StartPick one chore from the list, e.g. “answer calls after 5 PM”
  2. StepScore it 0 to 2 on frequency, rules, data, mistake cost and ownership
  3. DecisionDid it score 0 on mistake cost or on ownership?
    Yes
    A personKeep it with a person until the risk drops or someone takes ownership
    No
  4. DecisionDoes it total 7 or more out of 10?
    No
    StepPark it, noting which question scored low and what would change that
    Yes
  5. StepAdd it to the shortlist with its score and a one-line reason
  6. A personOwner picks the top scorer and names who will check on it weekly
  7. DoneOne first project, chosen on evidence instead of a demo
Two zero scores act as hard stops: a chore that could hurt someone, or that nobody will own, stays manual however well it scores elsewhere.

A worked example: the plumbing shop with the tote bag

Here's how the test might play out for that four-truck shop. The scores are illustrative, the kind the owner, office manager and a senior plumber might agree on in half an hour.

Hypothetical scores for a four-truck plumbing shop
ChoreOftenRulesDigitalMistake costOwnerTotal
Answer calls after 5 PM and on weekends222129
Follow up on unsold estimates122218
Draft repipe estimates from photos111126
Post on social media112206
Diagnose problems from customer videos101013

After-hours calls win. They happen every night, the questions are the same each time, the calendar is already digital, and the office manager has agreed to read every call summary for the first month.

Notice how the mistake-cost score was reached. The team scored the job as it would be scoped, with every emergency paged to the on-call plumber, so the 1 reflects the routine calls that remain. Without that routing rule the chore would score 0 and stop at the first gate, which is exactly what the gate is for.

Diagnosis from customer videos scores lowest, and that's the useful part of the exercise. It was the most exciting flyer in the bag, but it's judgment work where an error is expensive. It can wait until the shop has seen how a simpler system behaves. Unsold-estimate follow-up, in second place, becomes next quarter's project.

Another business would get a different winner. A small law firm often finds that gathering the basic details a conflict check needs from web inquiries scores high, while anything resembling legal advice scores zero on mistake cost. A dental practice usually finds recall reminders near the top: frequent, rule-based, already in the practice-management system, and a misfired text is cheap.

This test picks the task. Whether the business itself is ready (logins, software that can connect, time for a supervised start) is a separate check, covered in our AI readiness signs.

Before and after

Starting from a demo vs. starting from the test

Starting from a demo

  1. 01Pick the tool with the best trade-show demo
  2. 02Sign up, then hunt for a task it might fit
  3. 03Nobody is assigned to set it up or check it
  4. 04It handles the easy cases; staff route around it

ResultA monthly charge for a tool nobody opens

Starting from the test

  1. 01List twenty chores with the people who do them
  2. 02Score each one on the five questions
  3. 03Pick the top scorer with cheap, catchable mistakes
  4. 04Name an owner and a weekly check-in
  5. 05Only then compare tools that do that one job

ResultOne working system, and a shortlist for next time

The same owner and the same Monday, with two ways of choosing a first AI project.

Turning the winner into a first project

A score isn't a plan yet. Before you talk to any vendor, write half a page about the winning chore:

  1. 01The job in one sentence. “Answer every call between 5 PM and 8 AM, book routine work and page the on-call plumber for emergencies.”
  2. 02What's out of scope. “No prices beyond the service-call fee. No troubleshooting beyond the basic safety steps we approve.”
  3. 03What it connects to. The phone line, the calendar or field-service software, and the on-call phone.
  4. 04How you'll know it worked. One number you can count today, such as after-hours calls that became booked jobs last month.
  5. 05Who owns it, and when they check. A name and a weekday.

That half page is the brief you hand to any vendor or consultant. It also makes quotes comparable, because everyone is pricing the same job. If you'd like a second pair of eyes on your scores, that's the kind of thing an AI opportunity review covers, but the test works fine on your own.

Guardrails for a first project

  • Keep people on anything that could hurt someone. Medical, legal and safety judgment stays with licensed people. Software can collect the facts and get them to the right person faster.
  • Tell customers when they're dealing with automation. It costs nothing and spares them the feeling of being tricked.
  • Run it alongside the old way first. For the first few weeks, have the owner review everything the system did, so problems surface before customers notice.
  • Don't start two projects at once. A second project splits the owner's attention just when the first one needs it most.
  • Write down why you parked the others. Your shortlist is next quarter's plan, and the reasons tell you what would have to change.

Medical and dental practices have one more rule before anything starts: HIPAA requires a signed Business Associate Agreement with any vendor that will handle patients' protected health information. Take that as general information, not legal advice; state requirements can go further, so confirm the details with your own advisers.

Sources

Common questions

Quick answers.

What is the easiest way to start using AI in a small business?

Pick one frequent, rule-based chore where mistakes are cheap and easy to catch, such as after-hours call answering or appointment reminders. Name one person to own it, run it alongside your current process for a few weeks, and expand only once it works.

Should I start with ChatGPT or a custom AI system?

Neither, until you know the job. Plenty of first projects turn out to be covered by your existing practice or field-service software, or by a focused off-the-shelf product. Custom work is worth it mainly when the way you do the job is genuinely different or crosses several systems. Decide the job first, then the tool.

How long does it take to pick a first AI project?

The scoring takes an afternoon: an hour to list chores with your team, an hour to score them, and a short conversation to name an owner. Writing a clear half-page brief for the winner takes a little longer, and it saves time with every vendor you talk to.

What should a small business not use AI for first?

Anything where one mistake could hurt someone or cost you a client: clinical advice, legal opinions, safety instructions beyond the basics, pricing promises and final financial decisions. AI can gather the facts for those tasks, but a person should make the call.

Put it to work

AI consulting in New York

AI consulting for small and mid-sized businesses: we map your workflows, rank what's worth automating and hand you a written plan, with us or without us.

See how it works

Most relevant for law firms, plumbers, dentists.

Keep reading

Ready when the work is.

Show us what is slowing your business down.

Book a call

Not ready to talk?

Written brief

Skip the call.
Send a brief.

A few sentences is plenty. We read every brief and reply by email within one business day.

Try it free. If you love it, keep it. If it's not for you, we take it back, no questions asked.

We only use this to reply to you. Privacy policy