← All posts

AI Adoption Plan for a Small Business: A 90-Day Roadmap

August 25, 2026 · Mycel Team · Updated October 1, 2026

You run a business of 2 to 25 people. There is no IT department, no steering committee, and nobody whose job is "AI strategy," though most roadmaps online assume all three. You have a busy front desk, a few people wearing four hats, and a sense that you should be using AI for something.

Here is the short answer. Spend the first two weeks writing down what your business already knows: prices, policies, common questions, and how you do the core jobs. Then pick one task and let AI produce drafts that a person approves before anything goes out. After about six weeks of that, check whether it saved time without causing mistakes. If it did, add a second task. If not, fix it or stop.

Key Takeaways

Days 1-14: What should you write down first?

Enterprise roadmaps open with weeks of workshops and budget approvals. You can skip that, but you have a different problem: almost nothing is written down. The after-hours fee and the rescheduling rule live in two or three people's heads, and an AI tool cannot follow rules it has never seen. That is why it invents a policy or quotes last year's price. Write down these five things.

  1. Prices. Your current rate sheet, including the awkward ones: the trip charge, the after-hours fee for a plumbing call, the rush fee, what you charge when a client changes scope.
  2. Policies. Cancellation and no-show rules (a salon's 24-hour policy and what happens on the second miss), deposits, refunds, warranties, payment terms.
  3. The questions customers ask every week. Open your inbox and texts from the last month and list the ten most common questions with the answer you actually gave.
  4. How we do it. One page per core job. For an agency, the new-client onboarding checklist. For a trades shop, what the tech checks before leaving a job. For a clinic, what front desk confirms at booking.
  5. Things we never say or promise. Exact arrival times, medical advice, guarantees you do not offer. These guardrails matter as much as the answers.

Put it all in one shared place your team can edit, written the way you would explain it to a new hire. If you want a fuller list of what is worth documenting, AI automation for small business: where to start goes deeper on this step.

Two weeks is a target, not a rule. Do not start phase two with half your prices missing.

Days 15-45: How do you run your first AI task safely?

Pick one task that is repetitive, written, and low risk because a person reads every draft. Common ones:

If you are not sure which to choose, the first 5 processes to automate with AI walks through how to compare them.

Run it as drafts, always. The AI writes, a person reads, edits if needed, and sends. In this phase nothing goes out automatically. That is how you learn where the AI gets things wrong while a mistake costs a few seconds of editing.

Keep a simple log: the date, whether each draft was usable as-is, lightly edited, or rewritten, and why. When a draft is wrong because of a missing rule ("it didn't know we charge extra on Sundays"), add that rule to your written basics. The fix goes into the documents, not into the approver's memory.

Before you start, get a baseline. For one normal week, note roughly how long the task takes now. Say your office manager spends about 3 hours a week answering the same scheduling questions. Write down your own number, even a rough one.

Who on a small team owns what?

Three roles cover it, and one person can hold more than one.

Write those names down. "Everyone" owning something on a five-person team means nobody does.

What should you tell your staff?

Tell them early and specifically. People fill silence with the worst version of the story. A short team meeting covers it:

Also tell them what not to paste into AI tools. Customer card numbers, health details, and passwords stay out.

Days 46-90: How do you measure results and pick the next task?

After a month of drafts, sit down with the log and your baseline:

  1. Time. Is the task faster now, counting the approval time? If the 3 hours a week is now about 1 hour of reviewing drafts, you got 2 hours back.
  2. Quality. How many drafts went out with only light edits? Are the rewrites getting rarer as you add missing rules?
  3. Mistakes. Did anything wrong reach a customer? What caused it?
  4. The approver's view. Does the person approving think it helps?

If the answers are good, keep task one running as drafts and start a second task the same way. Do not switch the first task to fully automatic at the same moment you start a new one; change one thing at a time. If you later let some drafts go out unreviewed, start with the lowest-risk type (confirming business hours, say) and keep checking a sample.

For a more detailed way to put a dollar value on the hours saved, see how to measure ROI on AI automation.

When should you stop or roll back?

Decide this on day 15, not in a bad week. Reasonable stop signs:

Rolling back is easy. Because everything ran as drafts, stopping means the person writes the replies themselves again. Keep your written basics either way; they help new hires too.

Your one-page roadmap checklist

Copy this into a shared document and fill in the blanks.

Owner: (name). Basics keeper: (name). Draft approver: (name).

Days 1-14: write down the basics

Days 15-45: one task as drafts

Days 46-90: measure and add one

Where Mycel fits

The hardest part of this plan is days 1-14 and keeping those documents current. Mycel is built for that: your team keeps prices, policies, and how-we-do-it notes in one shared vault of plain folders, and a chat assistant answers from them. Connected Martello apps such as Relay or TradesBackbone read only the folders you share, and anything that would change something in another app waits for your approval.

Mycel is not a CRM or an automation builder, and it does not run tasks itself. There is a free tier. See how Mycel works and pricing.

Frequently Asked Questions

How do I start using AI in my small business?

Start by writing down your prices, policies, common customer questions, and how you do your core jobs in one shared place. Then pick one repetitive written task, such as replying to common questions, and have AI produce drafts that a person approves before sending.

What is the best first task to give AI in a small business?

A good first task is repetitive, written, and low risk if a draft is wrong, because a person reviews it first. Replies to frequent customer questions, follow-up messages after appointments, and cleaning up job notes into customer summaries are common choices. Avoid payments, legal commitments, and medical or safety advice at first.

Do I need an IT person to adopt AI in a small business?

No. The work in the first 90 days is mostly writing down what your business already knows and having someone review AI drafts. What you do need is named ownership: one person who decides, one who keeps the written information current, and one who approves drafts.

How long does it take to see results from AI in a small business?

It depends on the task and how complete your written information is. The plan here checks results after about a month of running one task as drafts, which is usually enough to see whether the approver is saving time. Thin documents mean more edits early on.

Will AI replace my staff?

In a small business, AI used this way drafts work that your staff review and send, so the person stays responsible for the customer. Be honest with your team about what is changing in your business specifically. If you have not decided something, say so.

Your next step this week

Block one hour this week and write down your current prices and your three most important policies in a single shared document. Add the five questions customers asked most this month, with the answers you gave. That hour is day one, and it is useful even if you never turn on an AI tool.