Most small businesses sit at stage 1 or 2 of five: a few people use AI on their own, often on personal free accounts, with no shared rules and nothing measured. What comes next depends on your stage. From private experiments, move to paid business seats and a written policy; from shared habits, automate one measured workflow.
The stages below are described by what you would notice in the office on an ordinary Tuesday, not by which software you own. A ten-question check places you in about five minutes (the answer is often a stage lower than owners expect), and each stage then has one next move, with its rough cost and time, that gets you to the stage above.
The five stages, as they look from inside the business
Maturity models from consultancies tend to run to five or six levels with names like "transformational". For a firm of three to thirty people, that language hides the practical question: what is actually happening at desks today? This scale is deliberately plain.
| Stage | What you'd notice | Typical tools | What's missing |
|---|---|---|---|
| 0. Watching | AI comes up in conversation; nobody admits to using it for work, or it has been informally banned | None officially | Any hands-on experience of what it does well or badly |
| 1. Private experiments | A few people use chat assistants on their own accounts; what works stays in their heads | Free or personally paid ChatGPT, Claude or Gemini | Rules on client data, a business account, any sharing |
| 2. Shared habits | Staff who need it have business seats, there's a short written policy, and people swap prompts that work | ChatGPT Business, Claude Team, Copilot Chat, Gemini in Workspace | Anything that runs without someone typing into a chat box |
| 3. One embedded workflow | One recurring job runs partly on its own, was measured before the change, and has an owner who checks it | An automation tool with an AI step, or AI features inside your existing software | A repeatable way to add the second and third workflow |
| 4. Managed portfolio | Several workflows, each with an owner; a list of every tool and what data it touches; spend reviewed each quarter | As stage 3, plus a tool approval route and a risk register | Nothing essential; the work is upkeep and pruning |
Two things surprise owners when they read this table. First, stage 1 carries the most risk for the least return: client details are going into accounts the business doesn't control, and none of the learning is captured. Second, having an AI feature switched on doesn't put you at stage 3. A website chatbot a supplier installed two years ago, which nobody owns or measures, is a stage 1 artefact with a monthly invoice attached.
A ten-question check to place your business
Answer each question yes or no for how things are today, not how they will be once the plan you've been meaning to write is written. Partly counts as no.
- (Stage 1) Has at least one person used an AI assistant for real work in the last month?
- (Stage 1) Do you know who those people are and roughly what they use it for?
- (Stage 2) Is everyone who uses AI for client or company work on a business plan the company pays for, rather than a personal account?
- (Stage 2) Is there a written rule, even half a page, on what may and may not be pasted into AI tools?
- (Stage 2) Do people share prompts or worked examples somewhere the rest of the team can find them?
- (Stage 3) Is there at least one recurring task where AI runs as part of the process, without someone copying text in and out by hand?
- (Stage 3) Did you measure that task (time, errors or turnaround) before AI was added?
- (Stage 3) Does one named person own that workflow and check its output at least monthly?
- (Stage 4) Are there three or more such workflows, each with its own owner?
- (Stage 4) Do you review AI spend and tools at least quarterly, against a list of every tool and the data it can reach?
The scoring rule: your stage is the highest one where every question in that group, and in every group below it, is a yes. If you answer no to anything in the stage 1 group, you are at stage 0.
This "weakest link" rule matters. Take a firm that has two slick automations built by a freelancer but no policy and half the team on personal accounts. It scores yes on questions 6 and 8, no on 3 and 4. By the rule, it is at stage 1. That's the honest answer: the automations sit on top of an unmanaged base, and the first thing that goes wrong will be a data or ownership problem, not a technical one.
What moving up from each stage involves
From stage 0 to 1: one person, two weeks
The quickest route is for the owner to use a chat assistant on their own admin for a fortnight: drafting replies, summarising long emails, turning meeting notes into actions. A free tier is enough to start; a paid individual plan is about $20 a month (ChatGPT Plus, Claude Pro, Google AI Pro at $19.99). Spend 20 minutes a day on it. You're ready to move on when you can name three tasks where it saved real time and one where it confidently got something wrong. The approach is laid out in using AI on your own admin before rolling it out.
What "ready to move on" looks like on paper: the owner of a café with a second small site might end the fortnight with a note like this (illustrative):
Saved time:
- Three suppliers' price-change emails turned into one table:
about 20 min down to 5
- Weekly rota message to staff from my scribbled grid: 15 min to 4
- Reply to a catering enquiry using my own price list: 25 min to 8
Got it wrong, confidently:
- Asked it to total a week of card-machine receipts. Out by $38,
with no warning. Totals stay in the spreadsheet from now on.The last line is worth as much as the three wins. It becomes the first rule in the stage 2 policy.
From stage 1 to 2: bring the experiments into the open
Start with an amnesty: ask everyone what they use and for what, with a clear promise that nobody is in trouble. Then do three things.
- Buy business seats for regular users. ChatGPT Business is $25 per user a month billed monthly or $20 billed annually, minimum two seats; Claude Team Standard is priced the same. If the business runs on Microsoft 365 or Google Workspace, check what's included first: Copilot Chat comes with Microsoft 365 business plans, and Gemini is built into Workspace business plans. Business plans don't train on your content by default.
- Write the one-page policy. What can go in, what can't, who checks output before it reaches a client. Writing an AI usage policy walks through the wording.
- Start a shared prompt collection. A single document with the ten prompts people already rely on is enough. See building a shared prompt library.
Expect two to four weeks and six to ten hours of the owner's time.
The amnesty message itself can be three sentences. An illustrative version:
Quick one: I'd like to know who's using ChatGPT, Claude, Gemini or any other AI tool for work, on any account, including free ones. Nobody is in trouble and I'm not taking anything away; I want to pay for proper business accounts for the people who find it useful. Reply to me by Friday with the tool, roughly what you use it for, and whether any client material has gone into it.
The last question is the one that matters, and the "nobody is in trouble" line is what gets it answered honestly.
From stage 2 to 3: pick one workflow and measure it first
Choose a task that happens at least weekly, follows a recognisable pattern, and does limited harm if a draft is wrong because a person reviews it. Record how long it takes now for two weeks before changing anything. Then connect the pieces: Zapier's Professional plan starts at $29.99 a month billed monthly for 750 tasks, and its built-in AI step (AI by Zapier) needs no separate API key; Make starts from about $9 a month on a credits basis. Run the new version alongside the old one for two to four weeks. Name the owner before you build, not after. For choosing the task, see what to automate first.
From stage 3 to 4: make the next workflow routine
The jump to stage 4 is about process, not tools. Write down how the first workflow was chosen, tested and handed over, so the second one follows the same route. Set up a simple way for staff to propose new uses, keep one list of tools and the data each touches, and put a quarterly review in the diary. The mechanics are in scaling AI from one workflow to the whole business.
The tool list doesn't need special software. For the video company in the walk-through below, a first version could be as short as this (illustrative):
| Tool | Owner | Data it can reach | Monthly cost | Next review |
|---|---|---|---|---|
| ChatGPT Business, 5 seats | A director | Whatever staff paste or upload; excluded client projects named in the policy | $100 | January |
| Zapier Professional | Producer | The shot-log folder and the producer's drafts | $29.99 | January |
| Transcription built into the editing software | Lead editor | Raw interview audio | Included in plan | April |
Three rows, five columns, and question 10 becomes answerable in a minute.
The stage you can skip, and the one you can't
Stage 0 is optional. Plenty of owners go straight to stage 1 by trying a tool on a quiet afternoon, and nothing is lost.
Stage 2 is the one businesses most often try to jump, and it's the jump that causes most of the mess I'd expect to find in a small firm. The pattern: someone on a personal account gets excited, a freelancer builds an automation using that person's login, and client data flows through a consumer account the business doesn't control. It works until the person leaves or the card on their personal subscription expires. Then nobody knows how the workflow runs, where the prompts live, or what data went where.
The other skip to avoid is stage 3 to "stage 4 in one go": buying five tools in a month because a competitor seems to have them. Each needs an owner and a baseline, and a small team rarely has the attention for more than one new workflow at a time.
An illustrative walk-through: a seven-person video production company
Say a video production company has two directors who own the business, two editors, a producer, a production coordinator and a junior. The directors believe two people use AI. The amnesty reveals five: three on personal ChatGPT Plus accounts they pay for themselves, two on free tiers. Nobody knew that interview transcripts from a client's unreleased campaign had been pasted into a free account to draft captions.
Score on day one: yes to questions 1 and 2 (after the amnesty), no to 3, 4 and 5. Stage 1.
Days 1 to 30, reaching stage 2. The company buys five ChatGPT Business seats on annual billing: 5 × $20 = $100 a month. The directors check client contracts and find two clients whose terms forbid unreleased material being processed by third-party tools without approval; those projects are excluded in the policy by name. The coordinator gathers twelve prompts people already use (call sheets from a shooting script, shot lists, client update emails, quote cover letters) into one shared document.
Days 31 to 90, reaching stage 3. They pick the post-shoot wrap email: the producer's summary to the client of what was filmed, what's outstanding and the next dates. A two-week baseline shows it takes about 40 minutes per shoot day, and there are around 14 shoot days a month, so roughly 9 hours a month. The new workflow: when the coordinator saves the day's shot log and transcript notes to a set folder, a Zap sends them to an AI step that drafts the wrap email for the producer to edit and send. Zapier Professional at $29.99 a month billed monthly covers it comfortably, since each run uses only a handful of tasks.
After a month of running both ways, the producer is spending about 12 minutes editing each draft instead of 40 minutes writing from scratch: roughly 6.5 hours a month back, for about $30 of extra spend. The producer owns the workflow and checks a sample each month against the shot log.
Score on day 90: yes to questions 1 to 8. Stage 3. Questions 9 and 10 are the plan for the next two quarters, starting with transcript-to-paper-edit notes for the editors.
How a business slips back down a stage
Stages aren't permanent. These are the usual ways a firm drops a level without noticing:
- The champion leaves. If the stage 3 workflow ran on their login, you are back at stage 1 the day their account is closed. Put every automation under a company-owned account from the start.
- Error alerts go to nobody. Automations fail quietly when their error emails land in an inbox no one reads. The workflow is still "live" on paper while staff have gone back to doing it by hand. An automation audit finds these.
- New starters never hear the rules. The policy was read once by the people there when it was written. Two hires later, a third of the team hasn't seen it, and question 4 is effectively a no.
- Seats renew for people who stopped using them. Spend creeps while use shrinks, and question 10 turns into a formality.
A nastier version of the second one, at the video company: Six months in, the coordinator sets up a new folder structure for the year and starts saving shot logs somewhere new. The Zap still watches the old folder, so it never fires. Because nothing technically fails, no error alert is sent at all. The producer assumes the drafts are "slow this week" and quietly goes back to writing wrap emails by hand, and nobody notices for five weeks, until the monthly sample check finds nothing to sample. A rule that folders feeding a workflow never move without telling its owner would have prevented it; a weekly count of drafts produced would have caught it within days.
Re-run the ten questions every six months and whenever someone who owns an AI workflow leaves. It takes five minutes and catches all four.
Questions owners ask about the stages
How long does it take a small business to move through the stages?
As a rough planning guide, getting from stage 0 or 1 to stage 2 takes a few weeks, because it is mostly buying seats and writing a short policy. The first stage 3 workflow usually needs one to three months, including a baseline and a parallel run. Stage 4 takes longer, often most of a year, because you need several workflows running long enough to review them properly.
Can different teams in one business be at different stages?
Yes, and it is common. The editors might be at stage 3 with an automated transcript workflow while the accounts side is still at stage 1. Score each team separately if the answers differ, but keep one policy and one tool list for the whole business, because data leaks and unused subscriptions do not respect team boundaries.
Is stage 4 the right goal for every business?
No. A three-person firm can sit happily at stage 2 or 3 for years if the work does not have many repetitive, high-volume jobs. The aim is to reach the stage where the time saved clearly outweighs the admin of running it. Moving up only makes sense when you can name the next workflow and the hours it would give back.
Further reads
- Example AI Roadmap for a 12-Person Business, Month by Month — A month-by-month plan once you know your starting stage.
- Who Should Own AI in a Small Business? Roles and Responsibilities — Stage 3 and 4 need a named owner; this sets out the roles.
- Shadow AI: Is Your Team Using AI Without Telling You? — Find out how much stage 1 use is already happening quietly.
- How to Review an AI Tool After 90 Days: Keep, Fix or Cancel — The review that stops a stage 3 workflow drifting.
- How to Write a One-Page AI Strategy for Your Business — Turn your stage and next move into a one-page plan.
- AI Governance for a Small Business: Who Decides, Approves, Checks — What stage 4 decision-making looks like in a small firm.
- How to Keep Up With AI in 30 Minutes a Week — A timed weekly routine for owners: scan your own tools' release notes, filter hard, test one change on real work, and log what you decide.
- Small Business AI Statistics 2026: Adoption, Spend, and Results — 2026 figures on small-business AI adoption, spending and results from eight named reports, with what each means for a five-person business.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: ChatGPT Business, Claude Team, Zapier and Make pricing pages; Zapier help page on AI by Zapier (checked September 2026).