Getting Value from AI: Getting Started with Copilot

With the arrival of Microsoft 365 Copilot, even more people have the opportunity to make their work easier with Artificial Intelligence (AI). Copilot is a powerful tool, but as with any innovation, a well-thought-out approach, realistic expectations, and a solid foundation are crucial. How can you go beyond the trend with Copilot and truly create value with it? Our expert William Vroegindewey explains more in this blog.

Expectations for Copilot Are High

For organizations, IT departments, and users, Microsoft’s Copilot seems to open up new dimensions for more efficient work and innovation. “The value that Copilot can deliver is significant. Expectations are high, but the risk is that this technology will overwhelm us if we don’t approach it step-by-step and with a solid foundation,” says William Vroegindewey. As a business consultant at Macaw, he regularly visits clients who are starting to work with Copilot. “Think of this technology as a golden faucet from which you want pure water to flow. Don’t turn that faucet all the way open at once. Do it step by step. So start by experimenting with Copilot with a small group of users and, at the same time, take the first steps to improve the quality of the output.”

Building the foundation as the first step

And before you do that: always start with the foundation, is William’s advice. “To work on your foundation and truly derive value from AI, you must first gain insight into all the documents, data, and information you have in your systems. Then clean it up: what’s outdated and can be discarded, what’s duplicated, and which version is the correct one? In short: move from unstructured data to structured data, and improve both quality and quantity before you start using Copilot,” he explains.

A Critical Approach to AI

That’s also where Macaw likes to guide customers who are getting started with Copilot. “It’s the role of a Microsoft partner to work patiently alongside the customer and start with that foundation. We advise our customers: don’t enable all of Copilot’s features right away, expecting that everything will just work out,” says William. “For IT, it’s important not to let yourself be rushed by the eager enthusiasm of users, who’d prefer to do everything with Copilot right away. Realize that users need a critical mindset to take advantage of the opportunities AI offers them. And involve users as early as possible in building that foundation: they know better than anyone which information or work processes are still current or accurate.”

Getting Value from Copilot

In practice, William observes that the reality of working with Copilot doesn’t always live up to expectations. “Everyone is wildly enthusiastic about the potential of Copilot and other AI applications. But: the main pitfall with Copilot is that you quickly get the feeling you’re collaborating with a human. That’s where disappointment lurks; you ultimately feel misled and don’t get any real value from the technology,” says William. “Realize that Copilot operates based on the data you provide: poor-quality content and data will therefore skew the answers you receive from Copilot. That’s why the foundation is so important—it really has to be solid first.”

The quality of the content determines the quality of Copilot’s response

Low-quality content is a risk for every company—and even more so for large companies with massive amounts of data. For example, Macaw conducted an analysis for a well-known energy company. Three million documents were reviewed, ranging from invoices to contracts and practical guidelines for the workplace. William: “As many as 800,000 documents turned out to be duplicate files, and in 2 percent of the cases, no metadata was available. So, without even realizing it, you’re undermining Copilot’s potential. It’s a shame—and, on top of that, risky for your organization.” Another factor to consider, according to William, is the permissions structure. “Not everyone is allowed to see everything. Make sure that’s properly set up. Otherwise, Copilot becomes a sort of snitch.”

Asking Copilot Exactly the Right Question

Once the foundation is in place, the next step is to roll out Copilot across the entire organization . For example, by training users on prompts—the way to ask questions of Copilot. “Teach people that reasoning is still an area of focus with AI. You have to ask Copilot exactly the right question: all the answers are there, but which one you get depends on how you phrase your question. As a simple test, a client had two different versions of the same AI model solve the same little riddle. Two different answers were generated,” says William. “Human oversight always remains very important when using AI. Only by approaching it thoughtfully can you achieve a great deal with it.”

From Foundation to Agentic Enterprise

A solid data foundation is the starting point, but not the end goal. Organizations go through six distinct phases on their journey toward mature AI adoption, from individual tasks in Copilot Chat to an Agentic Enterprise where AI agents actively contribute to business processes. Each phase has its own opportunities, challenges, and prerequisites, and knowing which phase your organization is in is the foundation for every next step. Discover where you stand and what the smartest next step is on our page about Copilot adoption, which includes a guidebook that walks you through all six phases.