Do Small Businesses Need AI? A Five-Question Test for Owners

Probably yes for at least one job, and no for the business as a whole. Score five questions: weekly repeatable admin hours, cost of an error, data sensitivity, staff skills and budget. A job scoring 8 or more out of 10, with no zero on error cost or data, is worth a trial now; 5 to 7 needs fixing first.

The test looks at jobs because "does my business need AI" is too vague to answer, while "should the Monday quote follow-ups go through AI" has an answer. Most owners find that one or two tasks pass easily, several fail for good reasons, and the rest depend on a decision only they can make, such as how much risk a wrong email carries.

Follow me on Instagram@sagnikteaches

The five questions and how each one scores

Pick one job. Score each question 0, 1 or 2. Add them up to a total out of 10. Two questions, error cost and data sensitivity, also act as brakes: a 0 on either means do not hand this job to AI as it stands, whatever the total says.

Connect on LinkedInSagnik Bhattacharya

Read the score like this. 8 to 10: run a two-week trial now. 5 to 7: a pilot is reasonable, but fix the weakest answer first, and expect to supervise closely. 0 to 4: leave it for now, or change the job so it scores better. And again: a 0 on question 2 or 3 stops the job, whatever the total.

Subscribe on YouTube@sagnikteaches

Each question deserves an explanation, because the thresholds look arbitrary until you see where they come from.

Question 1: weekly repeatable admin hours

"Repeatable" is the word that matters. AI saves time on work that has a similar shape every week: summarising enquiries, drafting standard replies, turning notes into a report, sorting invoices. It saves little on work that is different every time and needs judgement, such as negotiating a contract.

Count honestly. For one week, jot down every task that starts with the same trigger and ends in the same kind of output. Note the minutes. People usually guess they spend an hour on something that, once timed, takes twenty minutes, so measure before you decide.

Why do 3 hours a week score 2? A job that takes 3 hours a week is about 150 hours a year. If AI trims even a third of that, you recover about 50 hours, which pays for most tools several times over. A 30-minute weekly job saves, at most, a day or so a year, which is rarely worth a new tool and a new routine.

Say a tutoring centre sends a weekly progress note for each of 24 students. Each note takes 8 minutes to write from scratch, so that is 192 minutes, just over 3 hours. Score 2. A yoga studio that writes one newsletter a month spends perhaps 2 hours a month, or half an hour a week. Score 0.

Question 2: what one error costs

AI output is confident whether it is right or wrong, so the question is not "will it make mistakes" but "what happens when it does and nobody spots it". Think through the worst realistic case, not the average one.

• Score 2: an internal draft that a person reads before it goes anywhere, such as a meeting summary or a first version of a social post.
• Score 1: something customers see but can easily forgive and correct, such as a reminder with the wrong time that you can re-send.
• Score 0: a quoted price, a legal or tax statement, a health or safety instruction, or anything a client will rely on without checking.

A removals firm drafting a "your move is booked for Thursday" message from its diary scores 1: a wrong day is awkward, but a human glance catches it. The same firm asking AI to calculate the insurance value of a customer's goods scores 0, because a wrong figure creates a dispute.

A 0 here is not a permanent ban. It tells you to change the job's shape: let AI prepare a draft and let a person decide, so the error has to pass someone qualified before it counts.

Question 3: data sensitivity

What goes into the AI tool? If it is your own product descriptions, public information or figures with names removed, the data score is 2. If it includes customer names, emails and order histories, the score is 1, and you should use a business-grade plan and a written rule about what may be pasted. If it includes health records, financial account details or confidential client files, the score is 0 until you have a plan, a contract and a policy in place.

The list prices matter for this question. Business plans such as ChatGPT Business ($20 per user per month billed annually, $25 monthly) and Claude Team have two-seat minimums, and the major business plans do not train on business content by default, while free and consumer plans may work differently. Check the privacy page of whichever tool you pick, because settings differ by plan. For more on getting your records in order before any trial, the checklist on preparing business data for AI goes through it step by step.

Notice how a data score of 0 can be fixed: strip names and identifiers, or use only your own notes. Often the job survives anonymisation. A landscaping company that wants AI to draft quote wording can paste the job description (a 40-square-metre patio, two raised beds) without the client's name or address, and score 2 where the same job with full client details would score 1.

Question 4: who checks the output

This is the question most owners skip, and it is the one that most reliably decides whether a trial works. A tool is only as good as the person who reads what it produces. If nobody has the time or the knowledge to judge the result, the task fails quietly.

Name a person. Not "the team", a person. Ask whether they could write the output themselves, because someone who could do the job can spot a wrong figure or an odd phrase in seconds. If they would need to look everything up, the checking is slower than doing it.

Then ask about skill with the tool. Nobody needs to be technical, but they should be able to write a clear instruction, notice an answer that reads well but lacks substance, and know when to stop and ask. A one-hour practice session on a harmless task is enough to find out. If your people are nervous, the tutorial on what stops small businesses adopting AI covers how to deal with that.

Question 5: budget, measured in money rather than feelings

Put two figures side by side. The first is what the trial costs: tool subscription, plus the time of the person setting it up and checking it. The second is what the saved hours are worth. Use a loaded hourly figure, meaning wages plus the on-costs you pay, not the number on the payslip. If you have no figure, $25 an hour is a reasonable stand-in for illustration; replace it with your own.

Take a two-seat team plan at $25 per seat monthly, so $50 a month, and an owner who spends two hours a week on a job AI will cut by half. That is one hour a week, or about 4.3 hours a month, worth about $108 at $25 an hour. The tool costs $50, so it returns roughly twice its price before counting setup time. That scores 1: positive, but not by a wide margin once setup and checking are added. If the job took three hours a week and AI halved it, the same tool would return about three times its cost, and the answer moves towards a 2.

Do not forget the free options. A free plan from one of the major assistants is enough for a trial of a low-volume job, which makes the budget score 2 for any job where the free allowance is not a limit. Free tiers restrict usage (ChatGPT's free plan limits messages with uploads, for example), and OpenAI's $8 Go plan says it may include ads, so a free trial answers "does this work", not "will this scale". The guide on how much to budget for AI turns the answer into a monthly figure.

One full worked score: a tutoring centre's progress notes

Here is the whole test on one job, with illustrative figures. A tutoring centre with 24 students writes a short progress note to parents each week.

1. Hours. 24 notes at 8 minutes each is 192 minutes, about 3.2 hours a week. Score 2.
2. Error cost. A note that says the wrong topic was covered is awkward and easily corrected, and a tutor reads each one before sending. Score 1. (A note containing a made-up grade would score 0, so the tutor must supply the facts.)
3. Data. The note mentions a first name and a topic. Using initials or "[first name]" and leaving out surnames and school details makes the data low-risk. Score 2 with those rules, 1 without.
4. Checker. The student's own tutor reads the note before it goes. They know the child and the lesson. Score 2.
5. Budget. A two-seat business plan at $50 a month, against about 2 hours a week saved. At $25 an hour, 2 hours a week is roughly $215 a month. Score 2.

Total: 9 out of 10, with no zero on the brakes. Verdict: run a two-week trial. The prompt below shows how the tutor feeds in facts instead of letting the tool invent them:

Write a progress note of 70 to 90 words for a parent. Use only these facts:
Student: [first name]. Topic this week: fractions, adding with different
denominators. Went well: set out working neatly. Needs more practice:
finding the common denominator. Homework: 10 questions, worksheet 4.
Do not mention grades or comparisons with other students. Friendly,
plain tone, no exclamation marks.

A sample output (illustrative) might read: "This week [first name] worked on adding fractions with different denominators and set out the working very neatly. Finding the common denominator is the step to practise, so worksheet 4 has ten questions on exactly that." The tutor then fixes anything off, such as a stray claim like "great progress", which the facts did not support, and sends it.

Five more jobs scored, with different verdicts

Seeing a range of scores helps more than a single success story. All of these are illustrative.

A landscaping company drafting quote follow-ups: 8

Hours: about 2 a week chasing 15 open quotes, score 1. Error: a wrong date or a cheerful nudge to someone who already declined, score 1. Data: first names and job descriptions only, score 2. Checker: the owner reads each message, score 2. Budget: a free plan covers it, score 2. Total 8. Verdict: trial, with a rule that nothing sends without a read.

A removals firm calculating insurance values: stop

Hours might be high, but error cost is 0. A wrong valuation creates a dispute and a possible claim. The brake applies whatever the total. Alternative: AI drafts the inventory list from the surveyor's notes, and a person sets the values.

A yoga studio replying to booking questions: 6

Hours: 2 hours a week of repeat questions about class times and cancellations, score 1. Error: telling someone the wrong time is irritating, score 1. Data: customer names, score 1. Checker: replies are only glanced at now and then, score 1. Budget: a free plan covers the volume, score 2. Total 6. The easiest answer to improve is the checker, so the fix is to have AI suggest a reply for staff to approve, not send automatically. That lifts the checker to 2 and the total to 7. See what to automate first in customer service for how to stage this.

A tutoring centre handling parent complaints: stop for now

Hours are low and error cost is high: a poorly phrased reply to an upset parent can lose a family. Data includes a child's circumstances. The score is 0 on two brakes. Keep this human, and possibly use AI only to tidy the owner's own draft.

A removals firm tidying its weekly job notes: 9

Crew leaders send rough voice notes after each move. Turning them into a clean log takes 4 hours a week, score 2. Errors are internal and caught by the office, score 2. Data is addresses, so anonymise or use a business plan, score 1. A named office manager checks, score 2. Budget fits, score 2. Total 9. This is a classic quiet win: low risk, high repetition.

A mistake in scoring that makes everything look like a yes

The most common failure is scoring the tool's best case instead of the realistic case. An owner marks "hours" as 2 because the job "could" take 3 hours, when it takes one on an average week. Another marks "checker" as 2 because "someone will look", without naming who. The score then reads 9, the trial starts, nobody checks, and a few weeks later an embarrassing email goes out.

Guard against that with three habits. Use measured minutes, not recollection. Write the name of the checker on the scoring sheet. And ask someone else, a business partner or manager, to challenge any 2 you gave yourself. Scores that survive a challenge are the ones worth acting on.

Turning a score into a two-week trial

A pass is a reason to try, not to commit. Keep the trial small and the measures plain.

1. Day 1. Time the job manually for three normal runs and write down the minutes and the number of corrections you usually make.
2. Days 2 to 3. Set up the prompt, with the facts the tool may use and the things it must not do. Test on five old examples where you know the right answer.
3. Days 4 to 10. Run the job live, with the named checker reading every output. Log minutes taken and corrections made.
4. Days 11 to 14. Compare. If time is down by at least a third and errors caught did not rise alarmingly, keep it. If not, change the prompt once, then stop.

The measure is minutes saved per run, corrections per run and whether anything wrong reached a customer. Three numbers are plenty. If you want a fuller scoring list of your whole business after this, the AI readiness checklist takes about twenty minutes.

When the honest answer is not yet

Some businesses score low everywhere, and that is a legitimate result. A one-person trade with little paperwork, no repeat messages and mostly physical work may save an hour a week at best. Spending a month setting that up is a poor trade. The sensible move is to note the score, put a reminder in the diary for six months, and spend the effort on something with a clearer return.

If you score high but feel uneasy, trust the unease enough to find its source. Is it data? Is it that nobody will check? Often the unease points at the exact question where you scored a 1 that should have been a 0. Fix that and the job may well pass. Where you would like a second opinion, that kind of job-by-job scoring is part of what my AI implementation consultation covers, and you can reach me through the contact page.

Doubts owners raise after scoring the test

What if I score high on admin hours but low on data safety?

Treat the job, not the business, as the unit. A job that touches sensitive client data may fail while another job passes. Pick a task that uses your own notes, public information or anonymised figures, and revisit the sensitive job once you have a business-grade plan and a written rule about what staff may paste in.

Is not using AI a risk to my business?

Only if a competitor's saved hours turn into lower prices, faster replies or better service that you cannot match. For most small firms that effect is gradual. A sensible position is to test one job a quarter, so the choice stays yours rather than being forced on you.

How often should I rerun the test?

Rerun it for each new job you consider, and for the whole business once or twice a year. Tools, prices and your own staffing change, so a no this spring can become a yes later. Keep the scored sheet so you can see what changed in your answers.

Can the score replace professional advice?

No. It decides whether an AI trial is worth your time, not whether it is lawful for your trade. If a job involves regulated data, contracts or health, safety or financial advice, use the test to choose the pilot, then ask a qualified adviser before going live.

What if my team is against the idea?

Treat it as a skills and trust score, not an obstacle. Start with a job the team dislikes, such as meeting notes or chasing invoices, let them judge the output, and keep a person responsible for checking. Resistance usually falls when the first task removes work rather than adding monitoring.

Further reads

• Is AI Worth It for a Small Business? How to Work Out Your Answer — Work out the money case once a job has passed the test.
• AI Myths That Stop Small Businesses Getting Started — Clear up the beliefs that bias owners towards yes or no.
• How to Audit Your AI Subscriptions and Cut Wasted Spend — Keep spend in check once you have signed up for tools.
• Is a 1:1 AI Consultation Worth It for a Small Business? — Decide whether outside help would shorten your trial.
• 10 AI Implementation Mistakes Small Businesses Make — Mistakes owners make once a trial turns into a rollout.
• AI KPIs for Small Businesses: 12 Metrics Worth Tracking — Measures to track after the two-week trial.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: ChatGPT Business pricing page, ChatGPT plans page and Claude pricing page (list prices, seat ranges and training defaults, checked 5 November 2026).