How I Set Up Claude Teams for a Small Business (Safely)

A construction company in the DMV came to me earlier this year with a simple ask. The owner had been using a paid personal AI account for a few months, and it had changed how he worked. He was getting through bids, emails, and writeups faster than he ever had. He wanted to give that same edge to his whole team.

So he called me with a clear goal: roll this out across the company, and do it right.

I respected that, because it is the opposite of what I usually walk into. Most owners do not call me about AI until they find out, by accident, that their people have already been using it on their own. This owner saw the value himself, then made a deliberate decision to bring it to his team instead of letting it spread on its own.

But there is a real gap between one person using AI on their own account and a whole team using it across a business. One person knows what he is and is not pasting into the tool. A team of ten does not share that judgment automatically. The moment you scale AI from one set of hands to many, you need rules, or you are just hoping everyone makes the same good calls you would.

That gap is exactly what I set out to close.

Why I Did Not Just Hand Them a Login

The easy version of this engagement would have been to buy a handful of Claude Teams seats, send out the invites, and call it done. The owner already believed in the tool. I could have just turned it on for everyone.

I did not, because handing a whole team an AI tool with no rules is how you end up with client data sitting in a place you cannot account for. For a construction company, that means bid pricing, subcontractor terms, and signed contracts. For other businesses I work with, it is patient information, financial records, or controlled federal data. The stakes change. The principle does not.

What works fine for one careful owner does not automatically work for ten people moving fast. So before anyone else logged in, I built the governance first. The tool came second.

This is the part that catches people off guard. They think AI adoption is a software decision. It is actually a policy decision that happens to involve software.

Step One: A Real Conversation About What They Actually Do

I do not write a policy from a template and walk away. I sat down with the owner and his project managers and asked them to walk me through their week.

What takes too long? What do you hate doing? Where does the same task come up over and over?

The answers were specific. Turning a messy site visit into a clean client update took someone an hour every time. Writing up change orders ate an afternoon a week. Pulling a week of job notes into a summary for the Monday meeting was a recurring slog. Drafting subcontractor emails was constant.

None of those tasks require sending sensitive data anywhere risky. That mattered, because it told me where AI could safely save them real time, and where it could not.

Step Two: The AI Acceptable Use Policy

Every business I onboard to AI gets a written AI Acceptable Use Policy. It is part of the SNL-Tech AI Governance Kit, and it is not optional.

The policy does a few plain things. It names which AI tools are approved, so there is one answer instead of a free-for-all. It spells out what AI can be used for. And it spells out, in clear language, what must never go into an AI tool.

For this client, the "never" list was the important part. No customer Social Security numbers or financial account numbers. No full signed contracts. No employee records or payroll data. No passwords or login credentials. No anything marked confidential or under NDA.

I keep one rule at the center of every AI policy I write, because it is the one people remember: if you would not email it to an outside vendor, do not paste it into an AI tool.

That single sentence does more work than three pages of legalese. A project manager does not need to memorize a data classification chart. He needs one clear test he can apply in the moment.

Every employee signed it. Not because signatures are magic, but because a signed policy is the difference between "we told people to be careful" and "we have a documented standard our people agreed to." If a client, an auditor, or an insurance carrier ever asks how this business governs AI, there is an answer on paper.

Step Three: The Guardrails

A policy on paper is a start. Guardrails are what keep it real.

Claude Teams matters here for a reason that is easy to miss. Business AI plans like Claude Teams do not train their models on your company's data the way free consumer tools may. That alone is a meaningful difference for any business handling client information. It is the difference between a tool built for a company and a tool built for a curious individual.

On top of that, I set up a shared team workspace so the owner had visibility into how AI was being used across the business, instead of it living in scattered personal accounts he could not see. I standardized on one approved tool so "which AI are you using" stopped being a question. And I tied access into the same identity I had already secured: people sign in through their Microsoft account, which means the same MFA and access controls protecting their email and files now protect their AI tool too. One login, one set of rules, all documented.

For regulated businesses, this is where the industry-specific work comes in. A law firm, a healthcare practice, and a government contractor each carry different obligations, and the guardrails have to map to them. This construction client did not carry a heavy regulatory load, which kept things simpler. But I built the governance the same way I would for a client who did, because doing it right once is cheaper than redoing it later.

Why Claude, Not Copilot

This is the question I get most from owners already on Microsoft 365: if you have Copilot sitting right there, why bring in something else?

Fair question, and worth being straight about. On price they are close. A Claude Team Standard seat runs $20 per user per month on annual billing, with a five-seat minimum. Microsoft 365 Copilot for business runs in the same range per user, though it is an add-on that requires a qualifying Microsoft 365 license underneath it, so you are paying for the base license plus the Copilot seat. Cost was not the deciding factor either way.

What it came down to was fit, and the client worked that out for himself. He already had a Copilot license and had tried ChatGPT too. When he tried Claude, he kept coming back to it on his own, before I steered him anywhere. We both felt it simply performed better for the kind of work his team does. He also liked that he could work out of the desktop app on his computer, not just a browser tab. And the piece that sealed it was the skills and projects model. The ability to build a project around a specific job, give it the right context, and control who has access to it was exactly what their operation needed. That is the structure this whole rollout was built on.

I will be honest that part of this is preference. I use skills and projects myself, every day. I have projects built for running my own company and separate ones built for client work. So when a client tells me the skills-and-projects approach is what fits how they think, I know exactly what they mean, because it is how I work too. Copilot is a capable tool and it is deeply tied into the Microsoft apps, which suits some businesses well. For this client, and for how I work, Claude was the better fit. Your business might land differently, and that is a conversation worth having before you commit either way.

Why the Microsoft 365 Foundation Made This Possible

Here is the part most people miss. This rollout went smoothly because of work I had already done long before AI was on the table.

A while back, I moved this client into Microsoft 365 and retired their on-premise servers. Their data no longer lived on a box in a closet. It lived in SharePoint, with access controlled by security groups, so each employee could only reach the files their role actually needed. That was a real project at the time, and it was the hard part.

When AI came along, that foundation paid off in a way the owner never planned for. Because their data was already organized and permissioned correctly, walking through how the Microsoft 365 connector would work for them was straightforward. The connector lets the team ask Claude to find things across their own email and SharePoint files without downloading and re-uploading anything. And it respects the permissions that were already in place. Claude can only see what each person can already see. It searches their mailbox and the sites their login already opens, and nothing else. It is not a way around access. No new exposure. No reshuffling.

The configuration that usually trips up an AI rollout was already done. I was building on a clean, secure base instead of trying to bolt AI onto a mess.

I also set up the Microsoft 365 add-ins so Claude shows up right inside Excel, Word, and Outlook, in a side panel next to whatever the person is already working on. That was a deliberate choice. The fastest way to kill adoption is to make people stop what they are doing and go somewhere else. Bringing the tool into the apps they already live in, instead of forcing them into a new one, is why people actually used it.

This is the argument for working with someone who knows your whole environment, not just the latest tool. The AI rollout looked easy because the foundation underneath it was solid. That is not luck. That is the earlier work showing up.

Moving Claude From One Person to a Whole Team

The owner had not just been using AI. He had been building with it. On his personal account he had created around twenty custom skills, little tools that handled specific tasks the way he wanted them done. That work was good. The problem was that it all lived in one person's account, where nobody else could use it.

So the first job was getting that work out of his hands and into the team's. I downloaded all of his skills into a folder, then uploaded them into the Teams account at the organizational level. That one change matters more than it sounds. At the org level, those skills are shared across everyone, instead of trapped in a single login. The work he had already done became the whole team's starting point on day one.

Then I built the structure. He had skills but no projects, and a pile of skills with no organization does not scale. I walked him through setting up projects based on the type of work each one was for and who would need access to it. That is the part that keeps a Claude workspace manageable as it grows, instead of turning into a junk drawer six months later.

This is also where naming conventions earned their keep. It sounds like a small thing. It is not. How you name projects is how you keep them findable, how you decide who gets into which one, and how you keep the whole thing from becoming a mess as you add more. We agreed on a naming approach up front, before there were twenty projects to rename later. If you take one practical lesson from this post, let it be that one.

I built the project structure and set the access. Then I went to the owner for the part only he could provide: what each project actually needs to know to do its job. He pulled that together and organized it into SharePoint folders that mirrored the projects, because he is the one who knows what each is supposed to do for his team and how it should work with their data. I took that raw knowledge and turned it into the project descriptions, tweaked the instructions until each one behaved the way it should, and built it into every project.

That split is deliberate. I own the architecture, the access model, and the work of making the projects actually function. The client owns the knowledge that goes inside them. He knows his business; I know how to make Claude use that knowledge well. Neither of us is the right person to do the other's half.

Why I Stay Involved After the AI Setup

The rollout went to seven people, and one of those seats is mine. That was a deliberate conversation with the owner, not an afterthought.

I keep a seat because building this was not a one-time job. I use it to test that the skills and projects actually work the way they are supposed to, to fine-tune the project descriptions and instructions as the team grows, and to build new skills when a new need comes up. Just as important, I use it to make sure the hard stops actually hold. A rule in a policy is only real if someone has tried to get around it and could not. I am the one who tests that, so the owner does not find out the hard way.

This is the difference between setting up AI and managing it. The first is a project. The second is a relationship. A team's needs change, the work changes, and the tool has to keep up. Holding a seat means I am there to keep it tuned instead of leaving the owner to figure it out alone six months later.

How I Trained the Team to Actually Use Claude

Here is the part I am most proud of. The fastest way to waste an AI rollout is to hand people a powerful tool and a generic training video, then wonder why nobody uses it.

So instead of giving them a one-time overview of Claude, I built them a guide that lives inside the tool and never goes away. It is a project called "AI at [Client Name], Your Use Guide," shared with the whole team. The moment someone types "hi" into it, it offers them twelve prompts to choose from, depending on what they want to do or learn.

Some of those prompts teach the tool: what Claude can and cannot do, what each project they have access to is for, how to get a good result instead of a vague one. Others get real work done, walking them straight into the tasks they actually have in front of them. It is built for their environment and their projects, not a generic explainer that could apply to any company.

The reason I built it this way is simple. A training session is a moment. A guide built into the tool is always there. When someone forgets how to use a project three weeks later, they do not have to call me or dig through an email. They type "hi" and the guide walks them through it again. That is what gets a team past the awkward first weeks and into actually using AI every day.

I also gave the team one simple rule to keep it all straight: one tab is where you learn, the other is where you work. The guide is where you go to figure out how to do something. The projects assigned to you are where you actually do it. When people are new to a tool, a clear mental model like that does more for adoption than any feature list.

The Results: Hours Back Every Week

Here is what that looked like a month in.

Because this is a construction company, the work AI took off their plate is construction work. Reviewing an estimate and flagging what is missing. Pulling a week of field notes into a clean update. Turning rough job notes into a two-week look-ahead. Drafting change orders. Reviewing time-and-material tickets and weekly field reports. Summarizing a cost report from their accounting system into something the owner can actually read. These are the repetitive, time-eating tasks that sit between the real work and getting paid for it.

The hour-long site-visit writeups became about ten minutes. The project manager pastes his rough notes into the right project, and a clean, professional client update comes back that he reviews and sends.

The Monday meeting summary that used to eat half a morning now takes a few minutes. The change orders that swallowed an afternoon a week are faster and more consistent, because the format does not get reinvented every time.

Here is the one that made the owner sit up. Because Claude is connected across their tools, someone can ask it to read a cost report sitting in Excel and draft the summary email in Outlook, in one move, without copying anything between the two. It reads from one app and produces in another. That is the kind of thing that turns a twenty-minute shuffle into a thirty-second ask.

None of this replaced anyone's judgment. A human still reads everything before it goes out. That was a guardrail too: AI drafts, people decide. On the numbers that matter in their world, quantities, unit prices, markup, it is set up to cite where a figure came from or flag it for the person to verify, so nobody treats an unsupported number as final. The tool does the first 80 percent so the people can spend their time on the part that actually needs them.

The owner got something he did not expect. He stopped being the only one with the edge. The thing that had made him faster was now making his whole team faster, on an approved tool, under rules everyone agreed to. He could see how it was being used instead of wondering. The worry he might have had about turning ten people loose on AI never materialized, because the guardrails were in place before the seats were.

Claude Teams for a Small Business

Most small business owners I talk to in Columbia, MD, and across the DMV are somewhere on the same path this construction company was. The same is true for the owners I work with a little farther out, across Pennsylvania, West Virginia, and Delaware. Some are already using AI themselves and want to bring it to their team. Others have people quietly using it without any rules. Either way, the question is the same. It is not whether AI is in the building. It is whether anyone has decided how.

You do not get to opt out of that decision. You only get to decide whether you make it on purpose, the way this owner did, or let it get made for you one paste at a time.

The good news is that doing it right is not complicated or expensive. It is a conversation, a written policy, the right tool, and a few guardrails that match how your business actually works. For most small businesses, it is a short project with a long payoff.

A Quick Note on the Next Step

Everything above is about Claude Teams, the shared, cloud-based workspace I set up for this client's whole team. It is the right tool for collaborative, day-to-day work with proper oversight built in.

But there is another side of Claude built for a different job. It can run as an agent right on a machine, working through real files and multi-step tasks on its own, not just answering questions in a chat. That is a different tool with a different purpose, and frankly a different risk profile, which is why I keep it separate from a team's everyday workspace.

I have been putting that side to work in my own business in a way I am genuinely excited about. I built a dedicated machine that runs Claude to handle some of the most tedious parts of running an IT company: tracking nightly backups across my clients, and pulling my receipts, mileage, and expenses into clean, categorized reports without me touching a spreadsheet. I am writing that up as its own post, because how I built it, and why it lives on its own machine, is a story worth telling properly. Keep an eye out for it.

My Services Include

At SNL-Tech Services, I help small businesses adopt AI safely and put real time back into their week. My AI-related services include:

• AI Acceptable Use Policy creation and rollout.

• Setup and governance for the major AI tools, including Claude Teams, Microsoft Copilot, ChatGPT, and Perplexity, plus honest guidance on which fits your business.

• Microsoft 365 connector and SharePoint integration for AI tools.

• AI governance and risk assessment for small businesses.

• Shadow AI discovery and cleanup.

• AI guardrails that keep regulated data, including CUI under CMMC, out of unauthorized AI tools, and that map to HIPAA and FTC Safeguards obligations.

• Employee training on safe AI use, including in-tool use guides built for your environment.

• Ongoing AI management: new skills, instruction tuning, and rule testing as you grow.

• Ongoing managed IT and security for businesses across the DMV.

Frequently Asked Questions

Is Claude Teams safe for a small business to use?

Claude Teams is built for business use, which means it does not train its models on your company's data the way free consumer AI tools may. That makes it a much better fit for a business handling client information. The bigger safety question is governance: you still need a written policy and clear rules for what data goes into it. The tool is safe when it is paired with the right guardrails.

Should I use Claude Teams or Microsoft Copilot for my small business?

Both are solid, and on price they are close. A Claude Team Standard seat is $20 per user per month billed annually, with a five-seat minimum. Microsoft 365 Copilot for business is in a similar per-user range but is an add-on that requires a qualifying Microsoft 365 license underneath it. The real difference is fit. Copilot is built deep into the Microsoft apps and grounded in your Microsoft data, which suits some businesses well. Claude's strength is its skills and projects model, building a workspace around a specific job with its own context and access rules. For the construction client I worked with, he tried Copilot and ChatGPT and chose Claude himself because it fit his work better. Claude and Copilot are not the only options either. ChatGPT and Perplexity both have business tiers worth considering depending on what you need, and I help clients set up and govern any of them. The right answer depends on your business, which is worth talking through before you commit.

What does it cost to set up AI for a small business?

There are two parts. First, the software: a Claude Team Standard seat is $20 per user per month on annual billing, with a five-seat minimum, so a small team starts around $100 per month for the licenses. Second, the setup work to get the policy, structure, guardrails, and rollout done right. I do not quote a flat fee for that, because it genuinely depends on what you already have in place. Are you fully set up in a Microsoft 365 tenant? How are your security groups organized? Do you want a simple team deployment or something closer to automated workflows? A small-team rollout on a clean Microsoft 365 foundation is a short project. A more involved or automated build takes longer. I scope it to what you actually need.

How long does it take to roll out AI to a team?

For this client it took about two weeks start to finish. That included meeting with the owner, working through a few questions back and forth, setting up the structure and access, moving his existing work into the team account, and building the use guide. A clean Microsoft 365 foundation made it faster. A more complex environment or a more automated build would take longer.

What is an AI Acceptable Use Policy and do I really need one?

An AI Acceptable Use Policy is a short written document that names which AI tools are approved, what they can be used for, and what data must never be entered into them. If your employees are using AI at all, and they almost certainly are, you need one. It is the difference between "we hoped people were careful" and "we have a documented standard our people signed."

What is shadow AI?

Shadow AI is when employees use AI tools the business never approved, secured, or wrote rules about. It usually happens with good intentions, people are just trying to get work done faster. The risk is that sensitive company or client data ends up in tools the owner cannot see or account for. Even when an owner adopts AI on purpose, like the construction client I worked with, putting an approved tool and clear rules in place is what keeps shadow AI from creeping in later.

How much time can AI actually save my business?

It depends on the work, but the gains come from repetitive writing and summarizing tasks. For the construction client I worked with, hour-long site-visit writeups dropped to about ten minutes, and a half-morning Monday summary task shrank to a few minutes. The key is targeting tasks that repeat often and do not require sensitive data.

If I connect AI to my company files, will it see data my employees should not?

Only if your permissions are set up wrong to begin with. The Microsoft 365 connector respects the access controls you already have, so an employee using AI against company data sees exactly what they were already allowed to see, and nothing more. This is why getting your data organized in SharePoint with proper security groups matters before you add AI. If that foundation is solid, the AI inherits it. If it is a mess, AI will surface the mess faster.

Do I need to be on Microsoft 365 to use Claude Teams effectively?

No, but it helps. Claude Teams works on its own, and the time savings on writing and summarizing tasks do not depend on Microsoft 365. Where it gets powerful is connecting it to your company data and workflow, and if you are already on Microsoft 365 with your files organized in SharePoint, that integration is much cleaner. A lot of the value comes from work like that being done right beforehand.

I have been using Claude on my personal account. Can I move my work to a Teams account for my staff?

Yes, though it takes a few steps. Custom skills can be downloaded from a personal account and uploaded into a Teams account at the organizational level, which shares them across your whole team instead of leaving them stuck in one login. Projects are worth setting up fresh, organized by the type of work and who needs access, so the workspace stays manageable as you add more. I handle the structure and access, and I turn your knowledge into the project descriptions and instructions, so you supply what each project needs to know and I make it actually work.

How do you train a team to actually use Claude?

A one-time training session does not stick, so I do not rely on one. For this client I built a use guide that lives inside the tool as its own project, shared with the whole team. When someone types "hi," it offers a set of prompts that either teach them how to use Claude in their environment or walk them straight into real work. Because it is always there, people can relearn something the moment they need it instead of calling for help. That is what gets a team past the first few weeks and into daily use.

Do you stay involved after the AI is set up, or is it a one-time project?

I stay involved, and I hold a seat in the client's workspace to do it. AI setup is not a finish line. A team's needs change, new uses come up, and instructions need tuning as people lean on the tool more. My seat lets me build new skills as they are needed, fine-tune project instructions as the team grows, and keep testing that the safety rules still hold. Setting AI up is a project. Keeping it working well is a relationship.

Can AI help a construction company, or is it just for office work?

It helps with the office work that construction runs on. For the construction client I worked with, AI took on estimate review, turning field notes into two-week look-aheads, drafting change orders, reviewing time-and-material tickets and weekly field reports, and summarizing cost reports from their accounting system. It does not pour concrete, but it clears the paperwork that sits between the real work and getting paid, and that is where a contractor loses hours every week.

Will AI replace my employees?

No, and that is not how I set it up. The rule I use is that AI drafts and people decide. A human reviews everything before it goes out the door. AI handles the first rough draft so your people can spend their time on judgment, relationships, and the work that actually needs a person.

Can I use AI if my business has to follow HIPAA, CMMC, or other compliance rules?

Yes, but the guardrails have to match your obligations, and that is where most businesses get it wrong. The honest answer is that some regulated data should never go near a general-purpose AI tool at all. Controlled Unclassified Information under CMMC is the clearest example, the requirements around where and how it can be processed are strict enough that the right move is keeping it out of these tools entirely, not finding a way to feed it in. Other frameworks like HIPAA leave more room, as long as the right protections and agreements are in place. The real work is knowing which is which and building guardrails that draw those lines clearly, so your team gets the benefit of AI without putting regulated data somewhere it does not belong.

Do you only work with businesses near Columbia, MD?

I am based in Columbia, MD, and I work with small businesses across the DMV, including Maryland, Northern Virginia, Washington DC, Pennsylvania, West Virginia, and Delaware. AI governance and setup can be handled remotely, so location is rarely a barrier.

I want to bring AI to my whole team. Where do I start?

Start with the rules, not the logins. Before you hand seats to everyone, decide which tool is approved, write a short policy on what data can and cannot go into it, and put basic guardrails in place like signing in through a secured Microsoft account and a shared workspace you can actually see. That is the order the construction client and I followed, and it is why the rollout went smoothly instead of creating new problems. If your people are already using AI on their own, the same steps apply, and right now is the best time to put them in place.

Ready to talk?

If your team is already using AI and you have not put any rules around it, that is worth a conversation before something goes sideways. I will give you an honest read on where you stand and what it would take to do it right. No pressure.

Based in Columbia, MD, and serving small businesses across the DMV, I help owners adopt AI safely and get hours back in their week. Contact me today.