AI for small business: how one IT person planned a Microsoft 365 migration himself

This is what AI for small business looks like when it works, and a Southwest Washington business with one IT person, or none, can copy it.

The project that lands on a one-person IT team

If you run a business with ten to seventy-five people, you probably know this person. He keeps the lights on. Staff call him directly; nobody files a ticket. He resets passwords and fixes what breaks. Whatever project the owners drop on his desk lands with him too.

This time the project was a big one: move the entire company from Google Workspace, where it had lived since the early days, onto Microsoft 365. Email, calendars, files, sign-ins, licenses, the copiers that scan to email. All of it has to move without dropping an order or interrupting a register.

He didn't have a spare week to become a migration specialist on top of everything else. So our AI arm, Arthur Igloo, sat down with him for about an hour and set up a desktop AI to carry that load.

Why does this job need a desktop AI instead of a chatbot?

A desktop AI remembers the whole project, and that is what this job needed most. As we wrote in what we've learned about AI in small businesses, there is no single "use AI" answer; different shapes of AI fit different shapes of problem.

An autonomous agent was the wrong shape. Nothing here should run on autopilot, least of all the DNS cutover (the moment your email switches from Google to Microsoft). A chat window was wrong too. A chatbot is fine when you have one question. This project had hundreds, and they build on each other: question forty depends on the answer to question seven.

Claude Cowork for business is the right shape. It lives in Anthropic's Claude desktop app (we will just say Cowork from here), works alongside you, holds the whole project in its head, can be taught whatever you point it at, and leaves every real decision with you. The subscription is a small monthly one, and it is what we set up for him.

How do you set up desktop AI for a project like this?

The setup takes about ten minutes and goes like this:

1. Install Cowork on the machine you work at every day.
2. Make a folder called something like Cowork, with one subfolder per project. His was called M365 Migration.
3. Point Cowork at that folder. Everything it writes for the project lives there, where you can open it in Word.
4. Turn on the skill-creator skill. It lets Cowork build itself real expertise on Microsoft 365 migrations and save it to the folder, so it survives future sessions.
5. Allow it to read the web so it can crawl reference material.

Then came the first real instruction, three sentences long: "I'm an IT manager. I need to migrate from Google Workspace to Microsoft 365. Become an expert on this whole process; I'm going to give you reference material, then we're going to plan the migration together."

We handed it a few Microsoft Learn pages and told it to crawl out from there.

What the migration playbook covered

Fifteen minutes later there was a Word document in the project folder: Google Workspace to Microsoft 365 Migration Playbook. It was laid out in sections like a runbook, and it was specific to his company. You could hand it to a junior tech and say "follow this."

The second pass is where it got good. We asked Cowork to research the topic again and write a companion document covering everything the playbook did not. It came back with fifteen sections we would hand to a paying consulting client. Some of them:

• A DNS cutover timeline. Shorten your domain's refresh time a week ahead so you can roll back fast. Build the new email settings (for your IT person: the MX, SPF, DKIM, DMARC and autodiscover records) but don't publish them yet. Pilot with a handful of users, cut over, then test from an outside Gmail account and a real vendor within the first half hour. Never leave both Google and Microsoft mail settings live at equal priority, or your mail will land in the two systems at random.
• Licensing. Which staff need the higher tier for device management, which are fine on the standard tier, and which mailboxes (shared inboxes like info@, conference rooms) usually need no license at all. It also flagged the seat ceiling on the small-business tiers, so a growing company can plan for it now.
• The copier problem. Older printers and scanners send email with a plain username and password, a method Microsoft has been retiring. Every scan-to-email device stops working on cutover day unless someone plans for it.
• File storage. Do not move Google Drive over as-is. Decide what belongs in SharePoint (shared) and what belongs in OneDrive (personal), or you end up with individuals holding enormous personal drives full of company files.
• Calendars and mail rules. Only the primary calendar comes across; secondary calendars and room bookings do not. Mail rules arrive switched off, and every user has to turn theirs back on. That one was number one on the pitfalls list.

The document also covered a third-party app inventory (nobody does one, everybody regrets skipping it), a customer communications timeline, a pilot-wave-cutover plan, and a day-one survival kit for end users.

What AI for small business looks like from the IT chair

While Cowork worked, he was at his desk with the phone ringing every few minutes. The receptionist needed a vendor-portal password.

Every time a real-world wrinkle came up while he was doing the rest of his job, he just told Cowork: "We have copiers that send mail." Cowork folded it into the plan. The open questions, the earlier answers, the correction you made in passing forty minutes ago: a chatbot drops all of that, and a coworker keeps it in the room.

Most consulting projects lose momentum because the context goes stale between meetings and the runbook turns into "ask Dave." So before we wrapped up, we asked Cowork to write itself a project memo, a short file in the project folder that it reads every time the project opens. Think of it as the assistant briefing its future self: what the project is, what has been decided, and what is still open (final license mix, backup product, who owns customer communications). Three weeks from now, he opens the folder and Cowork picks up exactly where it left off. None of it has to live in his head.

What did it replace? A week of reading, a couple of paid consulting hours, a runbook written alone at night, and an app inventory discovered half-finished at 11pm on cutover day. What it cost was a small monthly subscription and one hour with us.

He will also come out sharper. By cutover day he will know this material like he wrote it, because he did. The story of that grocer and its IT manager is in a family grocer built to last; this post is the how-to.

What this means in SW Washington

Most businesses along the I-5 corridor south of Tacoma have one IT person, or an office manager who became the IT person by default, and no department behind them. Take a Centralia warehouse, a Longview clinic, or an Olympia law office near the Capitol Campus. Sooner or later each one gets handed a project bigger than the person holding it: a migration, a compliance review, a new practice-management system, anything you would hire a consultant for if you had the budget.

This is where our AI productivity training fits. With a Wallaby plan, the first sit-down to set up a project like this is included. We come to you in Centralia, Olympia or Longview, install the desktop AI, and point it at a real project you have been putting off. Then we spend the hour walking your person through the rhythm. After that, the AI is in the room every day and we are a phone call away.

The rest of your infrastructure stays our job. We run the phones, the network and the firewall, and we watch them around the clock from The Igloo in Tenino. Your IT person stops being the network guy and gets to be the person who finishes projects.

To see how this works on a project of your own, book a free on-site assessment. We will look at your network while we are there and show you the first AI project before we leave. Or call (360) 515-2500 any time; our AI answers.