Treat it as an eight-week plan with six moves: explain why and what won't change, pick one workflow, let the people who do that work shape it, agree simple rules, train on real tasks, then make the new way the default and retire the old one. In a team under 20, your own visible use outweighs any document.
Change management sounds like a big-company discipline, but small teams arguably need it more. In a five-person business, one sceptical colleague is a fifth of the workforce, and there's no spare capacity to absorb a messy rollout. The plan below gives you a week-by-week schedule, a script for the first team meeting, a way to read resistance, and clear signs that the change has actually stuck.
Why change lands differently in a team of five to twenty
Large organisations have change teams, communications staff and training budgets. A small team has you, and everyone knows everyone. That changes the job in five ways:
- You are the sponsor, the approver and often the first user. If you ask the team to use AI but never do yourself, they'll notice within a week.
- Relationships carry more weight than announcements. A quiet word from a respected colleague will do more than any email you send.
- One reaction spreads. A single "I tried it and it was rubbish" in the staff room can stall adoption for months.
- There's no slack. Training time comes out of real working hours, so it has to be short and tied to real tasks.
- The old way lives in people's heads. Processes are rarely written down, so "the new process" has to be agreed and recorded, or it drifts back.
If you've come across formal models such as ADKAR (awareness, desire, knowledge, ability, reinforcement), the plan below covers the same ground at a scale a small team can manage without a project manager.
The eight-week plan at a glance
| Week | Move | What happens | Output |
|---|---|---|---|
| 0 (before telling anyone) | Prepare | Choose the first workflow, decide what won't change (jobs, hours, who signs off), and record how long the task takes now | A one-paragraph reason and a baseline |
| 1 | Announce | Team meeting using the script below; short one-to-ones with anyone likely to worry | A list of questions and concerns |
| 2 | Shape | 45-minute session with the people who do the work to agree where AI fits and where a person checks | The new process on one page |
| 3 | Set up and agree rules | Business accounts, data rules, starter prompts | One page of rules |
| 4 | Train | 60 to 90 minutes, doing the actual task with AI | Everyone has done it once, with help |
| 5 to 6 | Trial | New way in live use; 15-minute check-in each week; issues written down | An issues log |
| 7 | Decide | Compare with the baseline; fix the top two issues | Go, adjust or stop |
| 8 | Make it the default | Switch off the old way, update induction notes | One way of working, written down |
Eight weeks feels slow for one workflow. It isn't: the second workflow usually takes half the time, because the rules, accounts and habits already exist. If you want a read on the team's mood before week one, a short anonymous survey helps; how to survey your staff before an AI rollout has the questions.
The first team meeting: what to say
Keep it to 20 minutes, in person if possible, and say the hard parts plainly. People fill silence with the worst interpretation. A script you can adapt:
1. WHY (2 minutes)
"We spend about [X hours] a week on [task]. I want to get some
of that time back for [the work that matters most here]."
2. WHAT (3 minutes)
"We're going to try AI for one thing first: [workflow]. Not
everything, and not all at once."
3. WHAT WON'T CHANGE (2 minutes)
"Nobody's hours or pay are changing because of this. [Only say
this if it's true.] A person still checks everything before it
goes to [customers / parents / clients]. [Name] still signs off."
4. WHAT I NEED FROM YOU (3 minutes)
"The people who do this work will design how it runs. I'd like
[names] in a 45-minute session next week."
5. TIMELINE (2 minutes)
"We'll try it for two weeks from [date], then decide together
whether to keep it, change it or drop it."
6. QUESTIONS (8 minutes)
"What worries you about this? Nothing's a silly question.
If you'd rather ask me privately, grab me afterwards."Point 3 is the one people remember. If you can't honestly promise that jobs and hours won't change, don't say it; the guidance on talking to staff who fear AI will take their job covers what to say instead. Follow the meeting with brief one-to-ones for anyone who went quiet.
Filled in, the script gets much more specific. Here's an illustrative version for the reception team of a nine-person veterinary practice trying AI on aftercare emails:
1. WHY
"Between the three of us we spend about six hours a week writing
aftercare emails after neutering and dental procedures. I'd like
that time back for the phones on Monday mornings."
2. WHAT
"We're trying AI for one thing: the first draft of the aftercare
email. Not booking, not billing, and nothing clinical."
3. WHAT WON'T CHANGE
"Nobody's hours change. A vet or nurse still reads every email
that mentions medication or wound care before it goes out. The
head nurse still signs off the templates."
4. WHAT I NEED FROM YOU
"You three send most of these, so you'll design it. Tuesday
after evening surgery, 45 minutes."
5. TIMELINE
"We try it from the 3rd for two weeks, then decide together."The filled-in version names the exact task, rules out the frightening possibilities ("nothing clinical") and puts a named clinical check in the middle of the process. Vague reassurance such as "AI will just help with admin" leaves people to imagine the rest, and they rarely imagine the kind version.
Letting the people who do the work shape the new process
This is the step that turns "the owner's AI project" into "how we do things now". In week two, sit down for 45 minutes with the two or three people who do the task most often:
- List the current steps on a whiteboard or a shared document, in order. Include the unofficial ones ("I check last week's version so I don't repeat myself").
- Mark each step: AI drafts it, a person does it, or a person checks it.
- Agree what "good" looks like for the finished output, in a sentence or two.
- Agree who has the final say before anything goes out.
Here's what the one page might look like after that session in a six-person kitchen-fitting firm, where the workflow is turning a surveyor's site notes into a written quote (illustrative):
| Step | Marked as | Who, and the rule |
|---|---|---|
| 1. Upload site notes and measurements after the visit | Person does | Surveyor, same day; photos stay in the job folder |
| 2. Check phone notes for extras the customer mentioned (tile removal, skip hire) | Person does | Office manager; this was the unofficial step nobody had written down |
| 3. Turn notes into a scope of works | AI drafts | Starter prompt, from the notes only |
| 4. Price the job from the price list | Person does | Office manager; the AI never sets prices |
| 5. Check the scope against the measurements | Person checks | The surveyor who measured |
| 6. Covering email to the customer | AI drafts, person checks | Office manager |
Their definition of "good" fitted in one line: a customer with no building knowledge can read the scope and tell what's included and what isn't. Step 2 is the interesting one. It only surfaced because the office manager mentioned she always rings back about old-tile disposal, since it's never in the survey notes. Had the owner designed the process alone, that step would have vanished and the first AI-drafted quotes would have left it out.
Let them make the calls where you can. They know where the task goes wrong, and people defend a process they designed. Write the result on one page. That page becomes the training material in week four, which is where role-based training on real tasks picks up.
Week three's rules can be short: which accounts to use (a business plan that doesn't train on your content, not personal free accounts), what information never goes into the tool, and who checks what. If you don't have an AI policy yet, writing a simple AI usage policy gives you wording to start from.
Reading resistance in a small team
Resistance is information. Most of it is about something specific, and each kind needs a different response:
| What you see | What's often behind it | What helps |
|---|---|---|
| Quiet non-use: nods in meetings, never opens the tool | Unsure what to use it for, or embarrassed to get it wrong | Sit with them for 15 minutes on one of their own tasks |
| "I tried it and it was rubbish" | A poor first attempt with a vague request | Give them a tested starter prompt and ask them to try it on the same job again |
| "I haven't got time to learn this" | Genuinely no slack in their week | Protect training time on the rota; take something else off their list for that week |
| "Is this about replacing us?" | Job security | A private, honest conversation with specifics |
| "It'll make mistakes and we'll get blamed" | Quality and accountability | Show them where the human check sits and who signs off |
| Over-enthusiasm: uses it for everything, skips checks | Excitement, or relief at saving time | Welcome the energy, then restate the checking rule |
The last row deserves attention because it doesn't look like resistance. The colleague who sends AI output unchecked creates the mistakes that give the sceptics their evidence. The tutorial on handling staff who over-rely on AI covers that conversation. If one person emerges as a natural helper during the trial, consider making it official by choosing and supporting an AI champion.
The "rubbish" row is worth seeing close up. In an illustrative six-person lettings office, a typical first attempt is two words pasted above a tenant's complaint about a leaking boiler: "Reply to this." What comes back reads something like: "Dear Tenant, thank you for reaching out. We understand your concern and will look into this matter promptly." It's polite and useless, because it doesn't say when anyone is coming. The starter prompt that replaced it gave the tool something true to say:
You write replies for a lettings office. Use the tenant's first
name, keep it under 120 words, plain rather than formal.
Include these facts and no others:
- Contractor: [trade], visiting [day and time window]
- Access: [office holds key / tenant to be home]
- Meanwhile: [what the tenant should do, e.g. turn off stopcock]
If a fact is missing, write [CHECK] rather than guessing.
Tenant's email:
[paste]Same colleague, same job, and this time a draft they could send after a 30-second read. The facts slots and the [CHECK] rule changed their mind far faster than any argument about what AI can do.
Worked example: a 14-person nursery moving parent updates onto AI
Here's how that might play out, as an illustration. A children's nursery has 14 staff: a manager, a deputy, four room leaders, practitioners, a cook and a part-time administrator.
Week 0. The manager picks one workflow: the weekly parent newsletter and the four rooms' weekly updates. She sets a firm boundary first: no child's name, observation, photo or developmental note goes into a general AI tool. Learning journals stay in the nursery's own system. Baseline: the newsletter takes her about 2.5 hours a week, and each room leader spends about 40 minutes on their room's update.
Week 1. At the staff meeting after closing, she follows the script. One practitioner asks whether parents will think "we can't be bothered to write to them any more". That goes on the concerns list rather than being brushed aside.
Weeks 2 and 3. The room leaders and administrator design the process: room leaders jot five bullet points about the week's activities (no names), the AI turns them into a warm paragraph in the nursery's style, and the room leader adds one personal line and checks it. Dates and times are pasted in from the nursery calendar, never left to the AI. The deputy sets up business accounts and a shared starter prompt.
Week 4. A 60-minute session after closing, where each room leader writes that week's real update with the AI.
Weeks 5 and 6. Two issues appear in the log. The AI once added a "reminder" about a trip that wasn't happening, caught by the room leader. And the cook's menu descriptions came out sounding like a restaurant, which parents found odd. Fixes: a line in the prompt ("Only include events and dates from the list I give you"), and the cook going back to writing her own menu notes, which was the right call.
Week 7. The newsletter now takes about 1 hour including checks, and room updates about 20 minutes each. That's roughly 1.5 hours plus 4 × 20 minutes, close to 2 hours 50 minutes back each week. The personal line answered the practitioner's worry: two parents mentioned that the updates had become more regular.
Week 8. The old newsletter template is removed from the shared drive, the new one-page process goes into the induction folder, and the manager stops accepting updates in the old format.
Retiring the old way: the step small teams skip
Most small-team AI changes don't fail in the trial. They fade in month three, because the old way was never switched off. The old template is still in the shared folder, the old habit is quicker on a busy Friday, and nobody minds. Within weeks, half the team is back where they started.
Retiring the old way is a deliberate act with four parts:
- Set a date and say it out loud: "From Monday the 14th, updates go through the new process."
- Remove or archive the old templates, forms and documents so they can't be picked up by accident.
- Update induction notes and checklists, so new starters only ever learn the new way.
- Stop accepting work done the old way, politely and consistently. If you accept it once, it becomes an option.
Leave one exception route for when the tool is down or a job is unusual, and write it down. An exception is fine; a parallel system isn't.
How skipping this shows up, in an illustrative twelve-person printing firm that moved job quotes onto an AI-drafted template: the trial went well, everyone agreed to switch, and nobody deleted the old Word quote file from the shared drive. At the 60-day check the owner pulled the last 20 quotes. Seven had been written the old way, all by the two people on the Saturday shift, who had missed the training session and simply opened the file they'd always used. None of the seven carried the new standard line about artwork proof deadlines, and two had turned into arguments about who pays for a reprint. The fix took ten minutes: archive the old file, put a shortcut to the new template where it used to sit, and walk the Saturday pair through one quote each.
Signs the change has stuck, and signs it hasn't
Check at 30, 60 and 90 days after week eight. Good signs:
- People use the new process without being reminded, including on busy days.
- New starters learn it as simply "how we do it".
- Someone suggests the next task that could work the same way.
- The time saving is still there when you re-measure against your baseline.
Warning signs:
- The old template reappears in a shared folder or someone's downloads.
- One person quietly does everyone else's AI drafting.
- Checks get skipped because "it's always fine".
- The weekly check-in stopped and nobody noticed.
If you see two or more warning signs, go back to the shaping step with the people involved and ask what's getting in the way. The answer is usually practical: a login problem, a prompt that stopped working, a task that changed. Fix that, restate the date, and pick the second workflow only when the first is solid.
Further reads
- How to Get Your Staff to Actually Use AI Tools — If adoption is patchy after week eight, start here.
- How to Run Your First AI Pilot Project in a Small Business — How to size and judge the trial in weeks five to seven.
- How to Roll Out an AI Policy So Staff Actually Follow It — Make the rules from week three stick in practice.
- How to Redesign Job Roles Once AI Handles Routine Tasks — What to do with the time the change frees up.
- How to Get Staff Buy-In When You Introduce AI — More on winning over sceptics before you start.
- AI Rollout Plan for a Ten-Person Marketing Agency — The same idea applied to a ten-person agency.
- Example AI Roadmap for a 12-Person Business, Month by Month — A 12-person driving school's first AI year, month by month: five workflows, about $30 a month in new tools, one idea postponed and why.
- How to Scale AI From One Workflow to the Whole Business — How to copy your first working AI workflow across the business: a readiness gate, a playbook template, adjacency rules and a pet shop's nine months.
- How to Run a Pre-Mortem Before You Launch an AI Project — A facilitator script, a 75-minute agenda and AI-specific failure prompts for finding what will sink your AI project before it goes live.
- Will AI Replace My Employees? An Honest Answer for Small Firms — An honest answer for owners: what the evidence shows, a task-by-task method to score each role, a bookkeeping firm example and what to tell staff.
- Barriers to AI Adoption in Small Businesses and How to Clear Them — Eight practical barriers that keep small firms from adopting AI, the cheapest way past each, and the few barriers you should respect rather than clear.
- Why AI Projects Fail in Small Businesses (It's Rarely the Tech) — The seven organisational reasons small-business AI projects fail, what each looks like by week three, and an eight-question check to run before you start.
- Why AI Adoption Stalls in Accounting Firms, and How to Restart It — The five reasons AI stalls in accounting practices, a twenty-minute diagnosis, and a 30-day restart timed around the deadline calendar.
- Custom GPT vs Claude Project vs Gemini Gem: Which Should You Use? — Two of the three are being retired. Here's what each vendor offers now for shared business context, and how a bakery re-homed its three custom GPTs.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: the ADKAR change model (awareness, desire, knowledge, ability, reinforcement) is referenced by name only; plan, timings and example are illustrative.