How to Get Started With AI in Your Small Business: First 7 Steps

Start small and specific: log where your week goes, pick one repetitive writing or admin task that takes at least two hours a week, set up one company-owned AI account, give it a short brief about your business, and trial it on that task for two weeks with a person checking every output. Budget $0 to $25 a month.

The seven steps below take about a month from start to decision, and the last one chooses your second task. The biggest risk at this stage isn't the technology. It's starting with a task that's too big, too sensitive or too vague to judge, and concluding after three weeks that "AI doesn't work for us". Each step here is designed to stop that happening.

Follow me on Instagram@sagnikteaches

To make it concrete, I'll follow an illustrative yoga studio through every step: the owner, four part-time teachers and a front-desk assistant, with a timetable of about 30 classes a week.

Connect on LinkedInSagnik Bhattacharya

Step 1: Log where the week actually goes (five days, 10 minutes a day)

Owners are usually wrong about where their time goes. Before choosing anything, keep a rough log for five working days. At the end of each day, note what you spent time on in blocks of 15 minutes or more:

Subscribe on YouTube@codingliquids

The yoga studio owner's week came out like this: about 4.5 hours on enquiries (beginners' questions, prices, private sessions, the teacher-training course), 2 hours on the newsletter and social posts, 1.5 hours on timetable changes and cover messages, 1 hour chasing unpaid class packs and 2 hours on bookkeeping. She'd have guessed social media was the biggest drain. It was enquiries by more than double.

Watch for work that never fills a 15-minute block. The front-desk assistant kept a log too, and hers looked quiet until the owner asked her to add a tally mark for every message that took under five minutes. Over five days there were 61: "Is the 6pm class full?", "Can I bring a friend?", "Where do I park?". At roughly 3 minutes each, that's just over 3 hours a week hidden in the gaps between other jobs. Short, repeated questions like these are often better first candidates than the big blocks, because the answers barely change from one week to the next.

Keep the log. It's also your baseline, the "before" figure you'll compare against in step 7. Setting a baseline before you introduce AI goes further if you want harder numbers.

Step 2: Pick one task that passes four tests (30 minutes)

From the log, pick the single task that passes all four of these:

1. It repeats at least weekly. Monthly tasks don't give you enough practice runs in a two-week trial.
2. It's mostly words. Reading, writing, summarising, rewording and sorting are where today's AI assistants are strongest.
3. You can judge the output in under a minute. If checking an AI draft takes as long as writing it, there's no saving.
4. A mistake gets caught before a customer sees it. For the first task, a person approves everything.

Scored against her log, the studio owner's five candidates came out like this:

Enquiry replies passed all four for the studio: dozens a week, all text, easy for the owner to judge, and nothing goes out without her pressing send. Bookkeeping failed test 3 (checking numbers is slow) and test 4 (errors can hide for months), so it waited. If several tasks pass, take the one with the most hours; what a small business should automate first has more on ranking them.

Tasks that usually pass in a small business: replying to the enquiries you get every week, turning rough notes into a tidy email, drafting a newsletter from bullet points, summarising long supplier emails, and rewording a policy in plain English. Tasks that usually fail at this stage: anything that relies on calculations, legal wording you'd depend on, and anything where the AI would act before you've seen what it's done. Those can come later, once you know how the tool behaves.

The same tests give a different answer in a different business. In an illustrative three-person landscaping firm, the owner's log showed 3 hours a week on quote follow-ups and 4 hours on pricing jobs. Pricing was the bigger number, but it failed test 3 (checking a price means redoing the measurements) and test 4 (a wrong quote reaches the customer as a commitment). Follow-up emails passed everything: weekly, all words, easy to judge, and nothing sends itself. So the firm started with the smaller task, and pricing went onto the "later" list with a note to revisit once it had a price book the AI could read.

Step 3: Set up one company-owned account (1 hour)

The right account depends on the office software the business runs on now. List prices in USD, as of September 2026:

Whichever you choose, sign up with a business email address rather than a personal one, pay with the business card, and turn on two-factor authentication. The studio already used Google Workspace Business Starter, so it started with Gemini in Gmail at no extra cost.

Then spend five minutes confirming the setup did what you think it did. Open the privacy or data settings and check the model-training switch: on a business plan it should be off by default, and on Plus or Pro you switch it off yourself. Sign out and back in with the business address, so you know which account the chats are saved under. The studio owner found her first Gemini session had opened under her personal Gmail, because that was the account signed into her browser. Every trial chat would have sat outside the business, under consumer terms, if she hadn't looked at the account picture in the corner before starting.

Step 4: Write a one-page brief about your business (1 hour)

Out of the box, an AI assistant knows nothing about your prices, policies or tone, so its drafts are generic and often wrong. A single page fixes most of that. Save it wherever your tool keeps standing instructions (a project, a saved prompt, or custom instructions), or paste it at the start of each session.

ABOUT US
[Business name] is a [type of business] with [number] staff.
We serve [who your customers are].

WHAT WE SELL
[Service 1]: [price], [what's included], [who it suits]
[Service 2]: ...

POLICIES CUSTOMERS ASK ABOUT
Cancellations: [rule]
Refunds: [rule]
[Other frequent questions]: [answers]

HOW WE WRITE
Tone: [e.g. warm, plain, no exclamation marks]
Sign off as: [name]
Always: [e.g. suggest the beginners' class for first-timers]
Never: [e.g. promise a specific result; quote a price not listed above]

IF YOU DON'T KNOW
Say "[Name] will confirm this" rather than guessing.

The last section matters most. Telling the assistant what to do when the answer isn't in the brief cuts invented details sharply. Giving AI your business context shows how to extend this page as you add tasks.

You don't have to start from a blank page. Paste in your website's prices page and FAQ and ask the assistant for a first draft:

Below is text from our website. Turn it into a one-page brief using
these headings: ABOUT US, WHAT WE SELL, POLICIES CUSTOMERS ASK ABOUT,
HOW WE WRITE, IF YOU DON'T KNOW. Use only facts that appear in the
text. Where a heading has no information, write MISSING rather than
filling it in.

[paste website text]

The MISSING instruction earns its place. When the studio owner first tried the prompt without it, the draft came back with "Refunds: full refund within 14 days of purchase" (illustrative), a policy the studio has never had. Her website said nothing about refunds, so the assistant supplied a plausible one. With the line added, the same section came back as "Refunds: MISSING". That's the answer you want: a gap you fill in yourself, rather than a confident guess you might not spot when you skim the draft.

Filled in, the studio's brief looked like this (figures illustrative). It took the owner 50 minutes, most of it spent finding the current cancellation wording:

ABOUT US
[Studio name] is a yoga studio with one owner, four part-time
teachers and a front-desk assistant. About 30 classes a week.
Most students are adults new to yoga or returning after a break.

WHAT WE SELL
Drop-in class: $18. Any class on the timetable.
10-class pack: $150, valid 4 months, not shareable.
Beginners' course: 4 Saturday mornings, $60, starts first
  Saturday of each month. Best first step for newcomers.
Private session: $70 for 60 minutes, booked with the owner.

POLICIES CUSTOMERS ASK ABOUT
Cancellations: free up to 12 hours before class; later than
  that, the class is used from your pack.
Mats: free to borrow. Bring water.
Injuries: ask the student to mention it to the teacher before
  class. Do not give advice about the injury.

HOW WE WRITE
Tone: warm and brief. Short paragraphs. No exclamation marks.
Sign off as: [owner's first name]
Always: suggest the beginners' course to anyone new to yoga.
Never: say yoga will fix a health problem.

IF YOU DON'T KNOW
Say "I'll check and come back to you today" rather than guessing.

Notice the injury line. It's there because the owner knew beginners often mention back pain, and she didn't want a cheerful AI reply implying a class would cure it.

Step 5: Trial it for two weeks with a checking rule

For two weeks, do the chosen task with AI every time, and check every output before it's used. Keep a simple trial log alongside: the date, how long it took, whether the draft needed no edits, light edits or a rewrite, and any factual error.

Keep the checking quick and consistent. For each draft, run a 30-second check in the same order: facts first (every time, price and date matches your source), then promises (nothing offered that you don't offer), then tone (would you sign it as it stands?). If a draft fails on facts or promises, log it as an error even though you fixed it before sending, because those are the patterns you need to spot across two weeks.

Here's one enquiry from the trial and what came back. The enquiry read: "Hi, I've never done yoga and I'm really stiff, plus I've got a dodgy lower back. Which class should I try and what does it cost?" With the brief loaded, the draft was (illustrative output):

Hi, thanks for getting in touch. The best place to start is our
beginners' course: four Saturday mornings for $60, starting on the
first Saturday of each month. It's slower-paced and the teacher
explains each pose, so being stiff is completely fine.

Yoga can be really helpful for lower back pain. Do mention your
back to the teacher before the first session.

Mats are free to borrow. Would you like me to book you onto the
next course?

[Owner's first name]

Price, day and mat policy all matched the brief, so it passed the facts check. It failed the promises check on one sentence: "Yoga can be really helpful for lower back pain" is exactly the health claim the brief said never to make. The owner deleted it, logged "light edit, promise", and after the same slip appeared twice more she moved the injury rule higher up the brief and reworded it as "Never suggest yoga helps any medical condition, even in general terms." It didn't recur. Her trial log for those three enquiries read:

The studio's two weeks, illustratively: 90 enquiries. Average handling time fell from about 6 minutes to about 2.5 minutes, including checking. Seventy-eight drafts needed light edits or none. Ten needed rewriting, and all ten were about the teacher-training course, which wasn't in the brief; adding one paragraph fixed it. Two drafts stated a class time wrongly, both caught before sending, because the brief linked to an old timetable.

The paragraph she added was short, and it deliberately left the details out:

TEACHER TRAINING
Runs twice a year. Dates and price change each intake, so never
quote them: link to the training page instead. Say the owner offers
a free 30-minute call to anyone considering it, and offer to book one.

Pointing to the source instead of copying the dates in is the same lesson as the old timetable. Anything that changes more than a couple of times a year goes stale inside a brief, so give the assistant a rule and a link rather than a fact it will repeat long after it stops being true.

One more thing about timing: two weeks is a guide, not a rule. What you need is enough outputs to see a pattern, roughly 30 to 40 of the chosen task. If one of your two weeks is a holiday week and only a handful come in, extend the trial until you reach that number rather than deciding on a thin sample.

The wrong class times are typical. Most early errors come from the information you gave the tool rather than from the tool, which is why the trial log records errors separately from edits. If your drafts come out sounding unlike you, keeping your own voice in a yoga or Pilates studio deals with exactly that.

Step 6: Agree three rules before anyone else joins in (30 minutes)

Before you widen the trial to staff, write down three rules:

1. What never goes in. For the studio: members' health notes, injuries and card details. Classifying your business data gives you a quick way to decide your own list.
2. Who checks what. Every customer-facing message is read by a person before it's sent, for now.
3. Where good prompts live. One shared document, so a prompt that works for one person works for everyone.

Three rules are enough for now. A fuller policy can wait until more people and more tasks are involved.

For rule 3, a shared document with one entry per prompt is plenty. The studio's first entry read:

NAME: Beginner enquiry reply
USE WHEN: someone new asks which class to try or what it costs
PASTE IN: the enquiry, with any health details removed
PROMPT: Using the studio brief, draft a reply that recommends the
beginners' course, gives the price and start date, and offers to
book them on. Under 100 words.
CHECK: price, start date, no health claims
LAST CHANGED: day 9 (added "no health claims" to CHECK)

The CHECK line is the part a new person needs most. It tells them which of the 30-second checks catches the mistakes this particular prompt tends to make, so they don't have to rediscover the health-claim slip for themselves.

Rule 1 was tested within a week of the front-desk assistant joining. A member emailed to pause her class pack, explaining in two sentences that she was pregnant and had been told to rest. The assistant pasted the whole email into Gemini to draft a kind reply. Nothing leaked, because it was the company account, but the health detail was exactly what the rule kept out. The owner didn't make a drama of it. She showed the assistant the version that works just as well: "A member wants to pause her 10-class pack for three months. Draft a warm reply confirming the pause and saying her remaining classes will be kept." The reply was as good, and the member's reason never left the inbox.

Step 7: Decide at day 30, then pick the second task

Compare the trial log with your baseline and make one of three decisions:

• Keep if it saves at least an hour a week after checking time, errors are rare and caught, and you actually used it most days.
• Fix if the saving is there but the drafts need too many edits. That usually means a better brief or a narrower task, not a different tool.
• Drop if you stopped opening the tool after the first week. That's a sign the task was wrong, so go back to step 2.

The sum behind the studio's decision fits on the back of the trial log. Before: about 45 enquiries a week at 6 minutes each, 270 minutes. During the trial: the same 45 at 2.5 minutes including checking, about 113 minutes. That's 157 minutes saved, roughly 2.6 hours a week, from a tool that cost nothing extra on her existing Workspace plan. Even at $20 a month the hour-a-week threshold would be cleared easily, so the time figure is the one worth checking honestly, not the subscription.

"Fix" is the verdict people skip, and it's often the right one. The landscaping firm's follow-up trial saved only about 40 minutes a week, because the drafts didn't know which job each quote was for and kept needing rewrites. Instead of dropping the tool, the owner changed the input: each request now started with three lines from the quote (the job, the price, the date sent). Edits fell to a sentence or two, and the second fortnight cleared the one-hour bar.

"Drop" happens too, and the signs usually show early. An illustrative mobile hairdresser chose Instagram captions as her first task because it was the job she disliked most, not because it was the biggest. Her log put it at 25 minutes a week. By day six she'd stopped opening the tool: writing a caption herself took about four minutes, while describing the photo to the assistant and editing its effort took about three, so the saving per post was one minute. The task had failed the two-hours-a-week rule before the trial began. Back at step 2, she picked appointment reminders and rebooking messages to regular clients, which her log showed at about 2 hours a week, and that trial was worth running.

The studio kept it: roughly 2.5 hours a week back from enquiries. Its second task, taken straight from the step 1 log, was the weekly newsletter plus timetable-change messages. It only needed a new section in the brief, because the account and rules were already in place. That's the pattern to aim for: each new task is cheaper to start than the last.

What the first month costs, in money and hours

Allow five to seven hours of your own time across the month. If you'd rather spend an afternoon on quick experiments before committing to a trial, twelve AI quick wins you can set up this week is the lighter-weight route.

Early signs your start is drifting

• You haven't opened the tool since week one. The task is wrong or the brief is too thin. Don't blame the software yet.
• You edit more than you'd have written. Narrow the task. "Reply to beginners asking which class to try" works better than "handle all enquiries".
• Someone pastes in data from the "never" list. Rewrite rule 1 in plainer words and show people a real example of what's off limits.
• You've signed up for three tools in the first month. One account is enough until the first task is settled.
• You can't say how much time it saves. Restart the trial log. A saving you can't measure is a saving you can't defend when the renewal comes round.

Questions before you take the first step

Can I start with a free AI account?

For learning, yes: try a free plan on public material such as your website copy to see what it can do. Once the work involves customers' names, messages or bookings, move to a company-owned account, either a business plan or a paid individual plan with model training switched off. Free plans also have tighter usage limits, which tend to bite in the second week of a real trial.

Do I need to tell customers I'm using AI?

If a person reads and approves every AI draft before it goes out, most small businesses treat it like a spell-checker and don't announce it. If AI replies to customers without a person checking, or a customer could reasonably think they're talking to a human, tell them. Rules on disclosure vary by where you trade and what you sell, so check if you're unsure.

My staff already use AI on their own. Do I start again?

No, start by asking them. Find out which tools they use and for what, because that's a ready-made list of tasks that already work. Then bring that use onto a company account, apply the three rules from step 6, and pick your trial task from what they've found useful rather than guessing.

Further reads

• AI for Small Business Owners: A Plain-English Beginner's Guide — The plain-English background, if the terms in this tutorial are new.
• What Is the Cheapest Way to Start Using AI in a Small Business? — How to run the same first month for the lowest possible spend.
• How to Run Your First AI Pilot Project in a Small Business — A fuller method once your second or third task is bigger.
• How to Write an AI Usage Policy for Your Small Business — Grow step 6's three rules into a policy as the team joins in.
• How to Review an AI Tool After 90 Days: Keep, Fix or Cancel — The keep, fix or cancel review to run after three months.
• Example AI Roadmap for a 12-Person Business, Month by Month — What the following year can look like, month by month.
• Who in Your Team Actually Needs a Paid AI Licence? — Deciding who else gets a seat when you widen the trial.
• Do I Need an AI Consultant? 8 Signs It's Time to Get Help — Eight signs a small business needs outside AI help, each with a test you can run this week, plus the signs that it doesn't need a consultant yet.
• Do I Need AI in My Business? A Decision Guide for Owners — A decision guide for owners: the three situations where AI is worth it, when waiting is sensible, a ten-question score and a 30-day test.
• AI Myths That Stop Small Businesses Getting Started — Ten beliefs that keep small firms from trying AI, what is actually true about each, and a small first step that tests it within a week.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: facts on plan prices from OpenAI, Anthropic, Google Workspace and Microsoft 365 pricing pages. Checked September 2026.