How to Write a One-Page AI Strategy for Your Business

A one-page AI strategy answers seven questions on a single side: what the business must improve this year, which two or three jobs AI will be tried on, what it won't be used for, the data rules, who owns it, the budget, and how you'll judge it at 90 days. Most owners can draft one in about two hours.

The line owners are most tempted to skip is what AI won't be used for, yet a sentence such as "no customer names or card details in free AI tools" settles more day-to-day questions than any goal on the page. The main caveat: this is a decision document, not a vision statement. If a line wouldn't change what anyone does next Monday, cut it.

Follow me on Instagram@sagnikteaches

Why one page? A small business rarely needs a 20-page plan that nobody opens again. It needs a filter. When a supplier offers an "AI-powered" upgrade, when a member of staff asks to pay for a tool, or when you read about something new over the weekend, the page should tell you in under a minute whether it fits.

Connect on LinkedInSagnik Bhattacharya

What the page decides, and what it hands to other documents

A one-page strategy sets direction and limits. It deliberately leaves three things to other documents, which keeps it short enough to be read:

Subscribe on YouTube@codingliquids

• Sequencing month by month belongs in a roadmap. If you want to see what that looks like, there's an example AI roadmap for a 12-person business. Your strategy page only needs the order of the first two or three jobs.
• Day-to-day rules for staff belong in a usage policy: approved tools, what may be pasted in, when a person must check the output. The strategy's data box should point to it. If you haven't got one, writing an AI usage policy is a sensible second document.
• The detail of each job (the prompt, the steps, who checks what) belongs in a short sheet per job, written when you start piloting it.

Keeping these apart is what lets the strategy stay on one side. The moment you start pasting prompts or tool settings into it, it turns into a manual and stops being read.

The seven boxes, with word limits

Each box answers one question. The word limits are there to force choices; a strategy that lists eleven priorities has none.

Notice that box 3 is where most drafts are weakest. Owners find it easy to list what they hope AI will do and awkward to write what it won't. Yet the "no" list is the part staff use most, because it answers the question they're too polite to ask: am I allowed to put this into ChatGPT?

Drafting it in five short sittings

You can do this in one afternoon, but splitting it over a week gives each box a night's thought. The timings below assume you already know the business well.

1. Start from the business, not the technology (30 minutes). Write box 1 first. Pull it from what already hurts: slow replies, the owner doing admin every evening, quotes that never get followed up, stock descriptions that are months behind. If you can't name a business problem, stop here. You don't need an AI strategy yet; you need a business priority.
2. List a week's repetitive work and pick two or three jobs (45 minutes). Ask each person to jot down the tasks they repeat, roughly how long each takes, and whether it's mostly reading or writing. Choose jobs that are frequent, text-heavy and easy for a person to check. Our tutorial on what a small business should automate first goes through the traits that make a job a good opener. Stop at three: each one needs a pilot, and a small team can't run five at once.
3. Write the "no" list and the data rules (20 minutes). Think about where a confident wrong answer would hurt: refunds, safety advice, prices, anything legal or medical. Then sort your information into what can go into an AI tool, what can go only into a business account, and what never goes in. If you're unsure how to sort it, classifying business data before using AI tools gives a simple three-tier scheme.
4. Name the owners and set the budget (15 minutes). One person decides and signs off spend; one person keeps the tools running. For money, write a monthly cap, not a wish list. For reference, business chat plans such as ChatGPT Business and Claude Team list at $25 a seat a month billed monthly (minimum two seats), and many firms already have some AI inside the Google Workspace or Microsoft 365 plan they pay for. Staff hours matter as much as subscriptions; budgeting for AI covers both.
5. Choose three measures and a review date (20 minutes). One measure of speed, one of quality, one of customer effect is a good mix. Put a real date in the diary about 90 days out. Our list of AI KPIs worth tracking has more options if the obvious ones don't fit.

Step 2 is where most of the thinking happens, so it helps to see the raw material. A four-person picture-framing workshop's week of jottings might come back like this:

The quote emails win clearly: frequent, text-heavy and checkable against a price sheet. Order write-ups come second. Mount choices are the biggest block of hours, but they fail every trait because the value is the framer's eye at the counter, so they stay off the page however tempting six hours looks.

Then leave the draft alone for a day and run the tests further down before anyone else reads it.

A filled-in example: a garden centre with a café and a delivery van

Here is how the page might look for an illustrative garden centre with 22 staff, a plant area, a gift shop, a café and a delivery service. The owner does most of the admin in the evenings, and in spring the enquiry inbox runs three or four days behind.

AI STRATEGY - version 1 - written 3 March - review 30 June

1. WHAT MUST IMPROVE THIS YEAR
   Reply to website and email enquiries within one working day
   (spring backlog is currently 3-4 days).
   Give the owner back 4 hours a week of evening admin by June.

2. WHERE WE TRY AI FIRST (in this order)
   a) First drafts of replies to plant-care and stock emails,
      checked by the plant-area supervisor before sending
   b) Summarising supplier price-change emails into one weekly
      list for the buyer
   c) First drafts of web descriptions for new product lines

3. WHAT AI WILL NOT DO HERE
   - Decide refunds, complaints or dead-plant claims
   - Tell a customer a plant is safe for pets or children
     without a trained person checking
   - Write allergen information for the cafe menu
   - Publish anything without a named person approving it

4. DATA RULES (full detail in the AI usage policy)
   - Customer names, emails and order details only in our
     business AI account, never in personal accounts
   - No card details, staff records or supplier contracts
     in any AI tool
   - Photos from customers: delete from the AI chat once answered

5. OWNERS
   Decides and approves spend: owner
   Runs tools, accounts and prompts: operations manager
   Each job: the supervisor of that area

6. BUDGET
   Tools: up to $150 a month for the first six months
   Time: 3 hours a week of the operations manager until June

7. HOW WE JUDGE IT ON 30 JUNE
   - Average time to first reply (March baseline: 3.4 days)
   - Share of AI drafts sent with minor or no edits (aim: 70%)
   - Complaints about wrong advice (aim: none)
   For each job: keep, change or stop.

A few choices in this example are worth copying. The pet-safety line exists because plant-care advice is where a fluent wrong answer does real harm; some common plants are toxic to cats and dogs, and a chat assistant can state the opposite with total confidence. The café allergen line is there for the same reason. The budget is a cap with an end date, so the June review is a genuine decision rather than a formality. And the baseline figure (3.4 days) was measured before anything changed, which is the only way the June comparison means anything.

Before signing off box 6, turn it into a sum. If the operations manager's time costs the business about $22 an hour, three hours a week over the 17 weeks from March to June is roughly $1,120. The tool cap adds at most $600 over four months. So the June review is really asking whether about $1,700 of combined spend bought back the owner's four evening hours a week and cleared the spring backlog. Written that way, nobody can mistake the $150 subscription line for the whole cost, and the staff time, which is the larger figure, gets the scrutiny it deserves.

Five tests to run on your draft

Read the page against each of these. If it fails one, rewrite that box before sharing it.

• The vendor test. Take the last three AI sales emails you received. Does the page tell you yes or no for each one? If you'd still have to think hard, box 2 or box 3 is too vague.
• The new-starter test. Could someone who joined yesterday read the page and know which tool to use for customer emails and what they must never paste in? If not, the data box needs plainer words.
• The Monday test. Does each priority job have an obvious first action someone could take next week? "Improve customer service with AI" fails. "Draft replies to stock enquiries" passes.
• The money test. In month three, could you tell from the page alone whether you're over budget? A cap in dollars and hours passes; "reasonable spend" fails.
• The stop test. Is there any result that would make you stop a job? If every outcome counts as success, the review date is decoration.

Here is the vendor test run against the garden centre page, using three pitches of the kind that land every week:

• "Add an AI chatbot to your website that answers customers around the clock." Box 2 lists email drafts that a supervisor checks, not live answers, and box 3 bars unsupervised pet-safety advice. Answer: no, not this year.
• "Your till system's new AI stock forecasting is included free from next month." It isn't a priority job, but it costs nothing and touches only sales data the supplier already holds. Answer: leave it switched off and add it to the June review agenda.
• "Turn product photos into web descriptions for $39 a month." This matches job (c) and fits the $150 cap, but job (c) is third in the order. Answer: not until job (a) is running routinely.

Three pitches, three answers, none needing a meeting. If your own page produces "it depends" for any of them, the gap is usually in box 2's order or box 3's wording.

A chat assistant makes a useful sceptical reader for the other four tests, provided you stop it doing what it likes best, which is adding more. Remove customer and staff names first, then try something like:

Here is my one-page AI strategy. Read it as a sceptical new
manager who has to act on it. For each of the seven boxes,
name one decision you still couldn't make from what's written.
Do not suggest new uses of AI and do not make the page longer.

[paste the page]

For the garden centre draft, the reply might read (illustrative):

Box 2: "checked by the plant-area supervisor" doesn't say who checks on the supervisor's days off. Box 6: it's unclear whether the $150 cap covers AI features added to software you already use, such as the till system. Box 7: "minor edits" isn't defined, so two people could score the same draft differently. You may also want to add social media captions and a website chatbot to box 2.

The first three points are real gaps. Fix them by naming a deputy checker, stating that the cap covers new AI subscriptions only, and defining a minor edit as "under two minutes of changes". Ignore the last sentence: it's the assistant padding box 2, which is exactly what the word limits are there to stop.

Phrases to strike from any first draft

Certain lines appear in almost every first draft and add nothing. They sound strategic but can't guide a decision. Delete or replace them:

• "Explore AI opportunities across the business." Replace with the two or three named jobs.
• "Use AI responsibly and ethically." Replace with the specific "no" list and data rules.
• "Stay ahead of competitors." Replace with the business result you want, with a number.
• "Upskill the team in AI." Replace with who will learn which tool, by when, for which job.
• "Measure ROI." Replace with the three measures and the baseline you'll compare against.

The data box suffers most from this. A three-physio clinic's first draft said: "Staff should take care with personal data and follow data-protection rules when using AI." Every physio agreed with it, and it told none of them whether they could paste a patient's exercise notes into a chat assistant. The rewrite sorted the three kinds of information they actually handle:

• Patient names, clinical notes and appointment details: only in the practice-management software, never in a chat assistant.
• Exercise sheets and general clinic emails with no patient identifiers: the clinic's business AI account only.
• Website and social drafts: any tool on the approved list.

The before version asked each person to make a judgement call on every paste. The after version made the call once, for everyone.

If a draft still reads like a press release after this pass, try reading it aloud to one member of staff and asking what they would do differently tomorrow. Silence means rewrite.

Keeping the page alive after the first month

Put a version number and date at the top, and keep old versions in the same folder. The history is useful: in a year you'll see which jobs stuck and which quietly disappeared.

Store it where staff already look, such as the shared drive's top folder or the staff noticeboard, not in the owner's email. Book a 30-minute review every quarter. Beyond the scheduled review, rewrite the page early if any of these happen:

• You start paying for a new AI tool, or an existing tool changes its price or plan structure.
• A data mistake or a customer complaint involves AI output.
• A pilot job becomes routine, which frees a slot in box 2 for the next one.
• The person named in box 5 leaves or changes role.

Record each change in two or three lines at the top of the new version, so the reasoning survives. After the June review, the garden centre's note might read:

• Job (a), enquiry replies: kept. Average first reply is now 0.9 days; 74% of drafts went out with minor or no edits; no advice complaints.
• Job (b), supplier price summaries: stopped. Most price changes arrived as PDF attachments, and the weekly summary missed two increases that the buyer only spotted on invoices.
• Job (c) moves up to second; new third job: first drafts of replies to café rota swap requests. Budget cap unchanged.

That stopped job is not a failure of the strategy. It's the stop test doing its work, after four months rather than a year.

Once three or more jobs are running routinely, one page may no longer be enough to hold the order of work. That's the point to add a simple roadmap on a second side, while keeping this page as the filter every new idea has to pass.

Questions owners ask about a one-page AI strategy

Should my AI strategy name specific tools?

Name the tools the business already subscribes to or has chosen, because that stops people buying duplicates. Avoid committing to a product for a job you haven't tested yet. A line such as 'trial one drafting assistant for enquiry replies' ages better than naming a tool before you know it handles your enquiries well. Revisit tool names at each quarterly review.

How is an AI strategy different from an AI usage policy?

The strategy says where the business will use AI, why, and what it will spend. The usage policy tells staff how to use AI safely day to day: which tools are approved, what they may paste in, and when output needs checking. The strategy's data box points to the policy. Even a five-person firm benefits from both, kept short.

Is a one-page strategy worth writing if I'm a sole trader?

Yes, in a smaller form: one goal, one or two jobs, a data rule, a monthly spending cap and a review date. It takes about 30 minutes. Its main value for a sole trader is stopping the drift of trying every new tool and giving you a fixed date to ask whether each subscription is earning its keep.

Who should see the finished page?

Everyone who might use AI at work, plus anyone who buys software for the business. Staff don't need to agree with every line, but they should know which jobs are being tried, what is off limits and who to ask. Sharing it also surfaces tools people are already using quietly, which you'll want to know about.

Further reads

• AI Use Case Template: Score Every Idea on One Page — Once a job is on your page, score its detail on a single sheet.
• Who Should Own AI in a Small Business? Roles and Responsibilities — Helps you fill the owner box with the right names.
• How to Run Your First AI Pilot Project in a Small Business — Turns your first priority job into a four-to-six-week trial.
• AI Strategy vs AI Implementation Consultant: Which Do You Need? — Explains when outside help with strategy is worth paying for.
• How to Review an AI Tool After 90 Days: Keep, Fix or Cancel — The keep, fix or cancel review your page's date should trigger.
• AI Adoption Stages: Where Is Your Business Now, and What's Next? — Check how ambitious your first page should be.
• How to Run a SWOT Analysis of Your Business With AI — Six stages and copyable prompts for a SWOT built on your own figures and reviews, with a podiatry clinic example and a TOWS step that turns it into actions.
• What an AI Implementation Plan Looks Like for a Small Restaurant — One illustrative 48-cover restaurant's AI plan, start to finish: a week of tracking, three jobs chosen, four ruled out, and the day-90 numbers.
• Can AI Write a Marketing Plan for a Small Business? — The input pack AI needs, a three-prompt sequence, how to red-pen the first draft, and a veterinary practice's 90-day plan with budget and owners.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: ChatGPT Business and Claude Team pricing pages (per-seat prices checked September 2026).