AI implementation for small business. A 30-day plan that sticks.
Most small businesses that buy AI never install it. The tools are not the problem, the install is. This is the 30-day plan I use to take a small team from "we bought AI" to "the team runs it without me." No developer. No agency. One bottleneck, one workflow at a time, read-only before anything writes.
Most small businesses that buy AI never install it.
They buy ChatGPT for the team. Maybe Claude too. They send a Slack message that says "start using this." Two people try it twice. Three weeks later everyone is back to email and spreadsheets, and the subscription quietly renews every month for work nobody does with it.
I spent 7 years in corporate finance at Big4 and then in bigtech before I started installing AI inside small businesses. The pattern is the same in almost every one. The tools are not the problem. The install is.
This is the 30-day plan I use to take a small team from "we bought AI" to "the team runs it without me." No developer. No agency. No platform to migrate to.
Why AI does not stick in a small business
The default way teams adopt AI is to open a chat window and ask it to do something. They get a generic answer. They compare it to what they would have written. They decide it is not better, and they stop.
That is not an AI problem. That is a briefing problem. The model had no context about the business, the customer, or what a good result looks like. A new hire with zero brief produces the same generic work.
Small businesses also try to install AI everywhere at once. Every process, the same week. Each new workflow needs its own calibration, so spreading thin means nothing gets to "trusted." Two workflows that run reliably beat five that run shaky.
So the plan does the opposite. One bottleneck. One workflow at a time. Read-only before anything writes.
The 30-day plan, week by week
Pick the one workflow that costs the most hours this month. Not the most impressive AI demo. The most expensive repeating task.
Connect one source the AI can read. Run the workflow once, read-only, with the owner watching. The team logs every place the AI got it wrong. By the end of week 1 the team can say what the AI does well and badly for their specific business. That is the goal, and it is why week 1 looks like nothing is happening.
Switch the workflow to approval-gated. The AI drafts the real output. A person reviews, edits, approves. The owner approves the first ten actions personally, then hands the workflow to the one team member who already opens it daily.
The first real saved hour usually shows up here. Often it is a side effect, like a Monday status email that used to take 90 minutes and now takes 12. That is the moment the install becomes hard to reverse.
Add a second source if it opens up a new workflow. Onboard a second person the same way: shadow first, then run with approvals. Tighten the prompts using the list of edits the team had to make before approving. The owner stops attending every session. That is the point.
Grade each workflow. Trusted, needs work, or kill. Most installs end with two or three trusted and one killed. Write a plain runbook for the trusted ones. Train the team, with them operating and the owner observing. Hand over admin access.
By day 30 the team owns two or three workflows that run every week, the documentation lives in their tools, and there is no dependency on me.
AI without a developer or an agency
Two things make this work for a small business specifically.
The first is that none of it requires code. The team talks to the workflow through chat. The setup happens once, by someone who has done it before, and then it is theirs.
The second is that it is finite. An agency builds for you and bills every month. The 30-day method installs the workflow, trains one person, and steps out. You are not renting a black box. You own a system you understand.
This is the whole reason I do not call myself a developer or an agency. The work is not writing software. The work is making a team actually use AI, and then leaving.
What to automate first
If you do not know where to start, look for a task that is true on all four counts:
- It repeats every week.
- It eats at least an hour, usually more.
- It needs almost no judgment.
- You can name the person who does it today.
That task pays back fastest, and its result is visible to the whole team, which is what makes the next workflow easier to adopt.
FAQ
The first saved hour usually shows up in week 2, once the workflow moves from read-only to approval-gated. The full install takes 30 days, and by then the team owns two or three workflows.
No. The workflows run inside the tools the team already uses, and the team operates them through chat. The technical setup happens once and does not require anyone on your side to code.
The single weekly task that costs the most hours and needs the least judgment. Drafting reports, triaging the inbox, cleaning records, and writing follow-ups are common first picks.
An agency builds for you and bills monthly. This is a finite 30-day install. One person on your team is trained to run it, you own the system, and there is no ongoing platform fee from me.
Want the first workflow mapped for your business?
Take the free AI audit. Answer a short set of questions and get a one-page map of the workflow to automate first, written for your business. No call. No card. Just the map.
Start my Audit →