ChatGPT, Copilot or Claude: which one for your business

Three tools that are more alike than you think

All three write, summarize, analyze an Excel file and answer your team’s questions. For most office work, the quality gap is small. What really sets them apart is how much of your business each one can see, where your data goes, and what it costs once everything is counted.

• Microsoft 365 Copilot. It lives in Outlook, Teams, Word and Excel, and works with whatever each employee is already allowed to see.
• ChatGPT Business. The one everybody knows. It connects to Microsoft 365, Google Workspace and several other tools.
• Claude Team. Very strong at reading and writing long documents. It also connects to Microsoft 365, and its coding tool is included.

What it really costs

On paper it is almost the same price: about US$20 per person per month, billed yearly. Copilot also needs a Microsoft 365 business plan, which most companies already have.

If you are on Microsoft 365, check one thing before buying anything. Copilot Chat is often already included in your plan at no extra cost. It is a good place to start: it is already paid for, and your IT people have already approved it.

The license is rarely the big expense. Ten people at US$20 is US$200 a month. What costs money is the time to connect the tool to your files, set it up for your tasks and bring the team along. We cover this in our article on what an AI project costs.

Your data: the question to ask before you sign

Good news first: none of the three business plans uses your data to train its models. Your employees’ free accounts do not give that guarantee, and that is often where quotes and client names end up.

For a Quebec company, the real question is geography. Microsoft keeps a Canadian tenant’s data in Canada and has announced in-country processing for 2027. With ChatGPT, Canadian hosting only comes with the Enterprise plan. With Claude, it is possible through the API, via Amazon or Google.

Quebec’s Law 25 adds a step. Before sending personal information outside Quebec, you need a privacy impact assessment, and a server in Ontario counts as outside Quebec. We do it with our clients before connecting anything. Our guide to Law 25 and AI explains how.

One last point about Copilot: it sees everything each employee is allowed to see. If payroll folders or price offers have been shared too widely for years, it will find them. Cleaning up permissions before launch saves you some unpleasant surprises.

How to choose: look at where your work happens

If your day runs through Outlook, Teams and SharePoint, Copilot has a head start. If you are on Google, or your tools are mixed, ChatGPT or Claude fit better. And if your work revolves around long documents, such as quotes, contracts or tenders, Claude is often our pick, especially in French.

If what you want is a whole process that runs by itself, with your ERP and email connected to it, the subscription matters little. We then choose the right model for each step. Sometimes there is more than one in the same system.

What we have seen with our clients

We have never picked the same tool twice for the same reason.

In the health network, the organization already had an approved assistant in its office suite. We built the solution inside it. Switching tools would have added months of approvals for no gain.

At 3D Coupe, estimating a part went from 58 to 15 minutes with agents built on the OpenAI API. The choice of model mattered far less than the calculation formula we wrote down with the estimators. See how.

At Group NDL, we connected Claude to the team’s Microsoft 365 mailboxes. Every morning, replies to clients are already drafted, with the right signature and attachments. It has become a service we offer our clients.

The mistakes we see most often

• Licenses for everyone on day one. A month later, a handful of people use it. Start with three to five people and one specific task.
• Letting everyone use their free account. If your employees already use AI, and they probably do, give them a business tool and a one-page rule.
• Choosing from the model leaderboard. It changes every quarter. Your files, your ERP and your habits stay.
• Thinking the subscription solves everything. An assistant saves a few minutes a day. The big gains come when AI is connected to a specific process, like quoting or answering clients.

Where to start

List what you already pay for and what your people already use, often without saying so. Pick a task that comes back every week, time it, then try one tool for 30 days with a few people. The number before and the number after will decide what comes next.

If you want us to look at it with you, with no brand preference, we start by listening. And if AI is not the right answer for that task, we will tell you.