Most AI advice tells you what's possible, not what to do on Monday. Here's a concrete 90-day plan for a small business — sequenced so each stage builds on the last, and deliberately unambitious at the start.
Why 90 days, and why slowly
The businesses that get lasting value from AI aren't the ones that adopted fastest. They're the ones that built a habit. Habits take weeks, and they form around one thing at a time.
The failure pattern is predictable: enthusiasm, five tools in a fortnight, nobody uses any of them by March. This plan is designed to avoid that by doing much less than you'll want to.
The single principle behind the whole plan: one task, done repeatedly, until it's automatic. Then the next. Breadth comes later; if you chase it in week two you'll end up with nothing.
Days 1–14: One task, one person
Week 1 — choose and start
Pick one task. It should be something you personally do at least three times a week, that involves writing, and that you find tedious. Common good choices: replying to routine enquiries, drafting quotes, writing social posts.
Then just use AI for that task, every time it comes up, for two weeks. Nothing else. Don't explore, don't try other tools, don't read about agents.
Prompt — finding your first task
I run a [type of business] with [number] staff. Here's roughly how my week goes and what eats my time: [describe]. Which single task should I use AI for first? Pick the one with the best combination of frequency, tedium and low risk — and tell me why you chose it over the alternatives.
Week 2 — get good at that one thing
Your first attempts will be mediocre. That's the point of two weeks. Refine your prompt, save the version that works, and notice what you keep having to correct. See how to write AI prompts for business.
End of fortnight check: is this task genuinely faster than before? If not, either the prompt needs work or you picked the wrong task. Fix that before continuing.
Days 15–30: Add two more, and set the rules
Week 3 — expand carefully
Add two more tasks in the same category. If you started with enquiry replies, add quotes and follow-ups. Related tasks reuse what you've already learned about prompting your business context.
Week 4 — write down the rules before anyone else joins
This is the week people skip and regret. Before staff start using AI, you need a simple, one-page position on what's allowed.
- No customer personal data, financial details or anything confidential goes into AI
- Which tools are approved
- Everything customer-facing gets read and edited before sending
- Who to ask when unsure
See a small business AI policy example and is it safe to put business data into AI. Half a page is enough — but it must exist before week five.
Days 31–60: Bring the team in
Week 5 — show, don't announce
Don't send an email saying "we're adopting AI". Show one person a task you've made faster and let them try it. Enthusiasm spreads better sideways than downwards.
Pick the person most likely to take to it — not the most senior. See introducing AI without resistance.
Weeks 6–7 — train properly, share prompts
Sit the team down for an hour. Cover the rules first, then hands-on with their real work — not a demo. Everyone leaves having done one actual job with AI.
Then start a shared prompt library. A document where anyone can paste a prompt that worked. This is the single most valuable artefact your business will build in these 90 days, because it compounds — every person's discovery becomes everyone's.
Week 8 — deal with the sceptics honestly
Someone will hate this. Take their concerns seriously; they're often the sharpest ones — about accuracy, about jobs, about quality. Answering them properly wins the room far better than enthusiasm. See how to stop staff using AI badly.
Days 61–90: Measure, then automate
Week 9 — actually measure something
You should now have real usage to assess. Don't guess.
Prompt — reviewing progress
I've been using AI in my [type] business for two months. Here's what we're using it for and roughly how often: [describe]. Help me work out what this is actually saving in hours per week, what I should measure to know if it's working, and what I might be missing. Be sceptical rather than encouraging.
Asking for scepticism is deliberate — you want an honest read, not validation. See measuring ROI on AI.
Week 10 — cut what isn't working
Some things won't have stuck. Drop them without ceremony. A shorter list of things people actually use beats a long list of good intentions.
Weeks 11–12 — consider your first automation
Only now, with habits established and rules in place, is it worth looking at automating something end to end. You'll pick a much better candidate than you would have in week one, because you now know from experience which tasks AI handles reliably. See building your first AI agent.
What to deliberately ignore for 90 days
- New tools — whatever you started with is almost certainly fine
- AI news and hype — it will not change what you should do this week
- Big transformation plans — you don't yet know enough to make one
- Anyone selling you an AI strategy before you've used it yourself
What good looks like at day 90
Not a transformed business. Realistically: three or four tasks that are genuinely faster, a team that uses AI without asking permission, a shared prompt library that's growing, a one-page policy everyone knows, and a clear sense of which task to automate next.
That's a strong position — far stronger than a business that bought five tools in January and abandoned them by March. See where to start with AI and how to use AI in your small business.
If you fall behind
You will, at some point, because a busy week will swallow it. That's fine — pick up where you left off rather than starting again. The plan is a sequence, not a schedule.
Want to do this properly rather than alone?
Monthly support means someone to check in with at each stage — and to ask when you're stuck. AskColin gives small businesses low-cost monthly support with one-to-one help whenever you need it — so when you're stuck on a real job, there's someone to ask. Practical, jargon-free, built around how you actually work.
Get one-to-one AI support for your businessFrequently asked questions
How should a small business start using AI?
Pick one task you do at least three times a week that involves writing and is tedious, and use AI for only that task for two weeks. Don't explore other tools. Breadth comes later — the businesses that get lasting value built a habit around one thing at a time.
How long does it take to get value from AI?
You should see a single task get genuinely faster within two weeks. A realistic 90-day outcome is three or four faster tasks, a team using AI without asking permission, a shared prompt library, a one-page policy, and a clear sense of what to automate next.
When should staff be brought into AI use?
Around week five, after you've proven it yourself and written down the rules. Show one person a task you've made faster rather than announcing an initiative — enthusiasm spreads better sideways than downwards. Pick whoever's most likely to take to it, not the most senior.
What's the most valuable thing to build in the first 90 days?
A shared prompt library — a document where anyone can paste a prompt that worked. It compounds, because every person's discovery becomes everyone's, and it's the artefact that survives when individual enthusiasm fades.
What should businesses ignore when starting with AI?
New tools (whatever you started with is fine), AI news and hype (it won't change what you should do this week), big transformation plans (you don't know enough yet), and anyone selling you an AI strategy before you've used it yourself.