Standardizing on Copilot keeps your data safe and your best people stuck.
Recently, I spent a full day with a distributor’s leadership team building their AI roadmap. Mid-morning we reached the governance section, and we hit the question that comes up in every one of these rooms: which data are your people allowed to put into these tools? Sales numbers, pricing, customer records.
One of the leaders answered with complete candor. “We found out somebody in accounting was using it, using some customer information. So, we just reacted and shut down ChatGPT and went to Copilot. We figured since it was through Microsoft and we licensed it, it was safer. But that was just a guess.”
A colleague confirmed that their Copilot was chosen at the company because it was the “safe” choice. I agreed it was inherently more secure than a free tool someone is using on their own. The problem, as I see it, is it doesn’t help users adopt generative AI tools and unlock the power of the organization on what can be solved.
By late afternoon, after a day of live demos and building out tools that are actually useful in the business, one of the company’s senior leaders said what everybody in the room was thinking: “I think the thing that limits it is our endorsement of Copilot only.” To me, it’s like giving your best warehouse team a hand truck and telling them the forklifts are off-limits because they’re too risky. The work still gets done. Just a lot slower, and your best people notice.
I’ll put my position on the table now and defend it for the rest of this piece. Dating back to the origins of ChatGPT, I’ve put a lot of my own time into these tools, using them every day, building with them and teaching them to distribution teams. Based on that, Copilot isn’t the right choice right now for a company that wants generative AI to really work in its business. The workflows I build in Claude are clunky at best in Copilot. They aren’t intuitive, and people walk away without feeling like they’ve got a handle on the tool. For the people doing real work with AI, Copilot is the weakest of the five major platforms. I’ve been accused of being a “fanboy” with Claude but as an operator I use what works.
I’ve heard some version of that story in most of the AI workshops we’ve run for distributors. The pattern is consistent, and IT has good reason to worry. Our research shows that 90% of employees use AI tools at work, but only 40% stick to the tools their company provides. So, IT moves fast, Copilot becomes the sanctioned tool, and the people who were getting real work done with AI either stop or go quiet about it.
I Understand the First Move
IT got the motive right and the tool wrong. Protecting customer data is their job, and they did it fast. Getting real value out of AI is a different job, and Copilot is the wrong tool for it.
Put yourself in that chair. Your people are asking for AI, leadership wants an answer, and you need something safe you can roll out today. Copilot is already sitting inside your Microsoft environment. It follows the permissions you’ve already set; it’s covered by a contract procurement already signed, and you manage it from the same administration console you’ve used for ten years. That’s an easy call, and for that moment it’s the right one. The problem is the business never got a seat at the table. Most of the people who would use these tools every day don’t know them well enough yet to make the case for something better, so IT makes the call by default.
Here’s where it goes sideways. The company lets the quick fix become its AI strategy, and nobody ever sits down to make that decision on purpose. I’ve spent more than 30 years running customer experience and operations at Grainger, HD Supply and Hisco, and I’ve watched that happen with more technology than I can count. You sink money into a tool because it’s easy to approve, and before long everybody’s expected to run the business on it, whether it solves the real problem or not.
Claude Is Just as Safe
Here’s the part most leadership teams haven’t caught up to. Claude Enterprise is just as secure as any Microsoft system you’ve already put in place. It carries the same security certifications your IT team checks for with every vendor. IT controls who gets an account and can see how it’s being used.
Anthropic doesn’t train its models on your company’s data by default. When Claude connects to your Outlook, SharePoint or Teams, each person only sees what they could already open on their own.
Copilot and Claude play by the same file permissions you already have. If your pricing file is open to the whole company in SharePoint, either tool will hand it to anyone who asks. Your real risk is personal accounts and files shared too widely. Both tools sit on the safe side of that line.
Then comes the pushback I hear in every room: “We have Copilot. We don’t need Claude.” Usually that’s because Microsoft now runs Claude inside Copilot. Since January, Claude has been switched on by default in several Copilot features for most U.S. business customers, and Microsoft’s newest agent product, Copilot Cowork, runs on Claude too.
That misses the point. Claude inside Copilot gets you Claude’s intelligence. It doesn’t get you the tools Anthropic built around it, the ones that let a sales manager or pricing analyst build something once and have the whole team run it every day. It’s like bolting a forklift engine onto a hand truck. If your people only ever meet Claude through somebody else’s tool, they’re judging it on a fraction of what it can do.
The Quiet Acknowledgment in IT
One on one, away from the steering committee, IT professionals tell me something the official tool policy rarely reflects. There’s a silent acknowledgment that Claude is simply better at helping people get work done.
An IT lead at a heavy-duty parts distributor told me Claude is what he likes to use, but the company is on Copilot “for right now, just because of all the licensing.” His team had already built an internal warehouse badge-tracking tool with Claude. He’d picked Copilot because it connected to Outlook and SharePoint, then tested Claude’s Microsoft 365 connector on the company’s enterprise account and called it phenomenal.
An IT manager at another company, whose first concern is data loss prevention, had already turned on the Claude and OpenAI models inside Copilot for two testers. He’d also run into permission friction getting Claude to pull data through his own tightly governed Microsoft environment, and he was still working to get it right.
My read on why this stays quiet: the safe choice is easy to defend in a meeting, and reopening it means admitting the first decision was incomplete. Nobody gets fired for picking Microsoft. Plenty of good builders quietly stop building because of it.
Copilot Earns Its Place in Office Work
I’ve worked extensively with all five of the major AI platforms. Copilot has a real job, and I’d say so in any room.
It drafts an email in Outlook without anyone opening another window. It turns a Teams meeting into a recap. It helps someone rough out a PowerPoint deck or format a spreadsheet. IT governs it with tools they already know, and the licensing rides on an agreement that’s already in place. Microsoft also says that in its own testing, Copilot Cowork ran 30% to 40% cheaper per prompt than Claude Cowork with the Microsoft 365 connector, based on 125 runs across 12 prompts with both products on the same Anthropic model.
For broad, everyday office productivity, Copilot is a sensible default. For the people doing the actual work, Copilot ranks fifth of the five, and it isn’t close.
Copilot Breaks the Moment Someone Builds
The ranking comes from what happens when an operator tries to build something.
Before a workshop with a distributor that had standardized on Copilot, my team checked every Claude example in the deck against Copilot, slide by slide. Of 16 examples, nine worked as shown, six worked only partly and one didn’t translate at all.
Here’s what that looks like in real terms. A sales team wants a pre-call brief for each rep: account history from SharePoint, invoices and open orders from the enterprise resource planning (ERP) system, recent activity from the customer relationship management (CRM) system, all in one document before the rep pulls into the customer’s lot. In Claude, a capable operator builds that as a skill, connects it through certified connectors, and has a working version the same week. The rep can schedule it to land in the inbox before 7 a.m. In Copilot, the same request usually becomes a licensing conversation and a ticket.
Adoption lives or dies on that gap more than on any feature list. Your people get a taste of AI, bring Copilot a real problem, and get a mediocre answer. Most of them decide AI isn’t any good, or they go back to the personal account you banned and stop telling anyone.
I’ve seen this movie with other software. When I was at Hisco, I tore out Salesforce because we were paying $300,000 a year for it and it wasn’t getting us what we needed. It was approved, licensed and secure. It just wasn’t useful to the people who were supposed to live in it, and an unused system is the most expensive software you own. A Copilot seat your best people have stopped opening is the same line item at $30 a month.
The broader market has felt the same drag. Gartner reported in August 2024 that only 6%of organizations in its survey had finished a Microsoft 365 Copilot pilot and were moving to a large-scale deployment.
What This Means for Your Operation
You can keep Microsoft and still avoid buying two licenses for everyone. The IT lead I mentioned earlier raised exactly that worry: paying for Copilot and Claude Enterprise on the same seat. That’s the wrong math.
Split the roles. Keep Copilot as the broad default for drafting and summarizing inside Office. Give Claude Enterprise to the people who build: inside sales leaders, pricing analysts, operations managers, the branch manager who already has three spreadsheets that should be one tool. That’s usually a small group, and it’s the group whose output the rest of the company ends up using.
Govern both with the same discipline. Limit connectors to certified providers. Put a human review step between any skill someone builds and its release to the team. Set spend limits on usage-billed agents, because Copilot Cowork bills by the task in Copilot Credits on top of a Microsoft 365 Copilot license ($30 per user per month at list), and a report somebody leaves running can burn a budget either way. I run hard caps for my own team for the same reason.
Think back to that warehouse team with the hand truck. Nobody’s asking you to rip out the safety rules. They’re asking for the forklift, with the training and guardrails that come with it.
Here’s my bottom line. If all you need is a safe place for people to draft emails and summarize meetings, Copilot does the job. If you want your team using these tools to get more productive and more effective, Copilot isn’t it, at least for now. Make that call on purpose because your best people are already making it for you.
Don’t take my word for it. Come to the Applied AI for Distributors Conference in Chicago and see what distributors are building with AI right now. Details at appliedaifordistributors.com.
Agentic AI is here. Are you ready?
Share this article:

