In our latest Ask the Expert video, Steve Drake, director of technology at TrueNorth IT, shares insights into how organisations can get the most from Microsoft Copilot. While the Artificial Intelligence (AI) assistant offers huge potential, its performance depends heavily on data quality, structure and strategy. Here are four key takeaways from the conversation.

1. Copilot is only as good as your data

A common misconception is that Copilot “learns” from your business data. In fact, it doesn’t. Copilot is a large language model that understands and interacts with your data, but only if that data is in the right place and format.

Steve explains that Microsoft built tools like Dynamics 365 with specific data structures in mind. If users enter information in the wrong fields or create too many custom fields, Copilot simply can’t find what it needs. The result is inaccurate summaries and missed information.

To make Copilot effective, organisations will ideally use product-standard fields and maintain consistent data input.

2. Getting Copilot-ready means tidying up your data

Before you start expecting value from Copilot, it’s crucial to prepare your data environment. That means identifying where you’re not getting results and finding out why. Are key fields missing? Is the form structured incorrectly? Could a hidden field be affecting accuracy?

Steve recommends sandbox testing to align with Microsoft’s default expectations. For quick wins, he suggests tools like Xrm ToolBox to refine form design and field visibility.

3. Be smart about security and sensitivity

Security concerns are often mentioned when AI is discussed, but Copilot respects existing security settings if they’re correctly set up. If users can’t access a file, neither can Copilot. However, its language model means it may search more creatively, surfacing obscure or oddly-labelled content. To manage this, Steve advocates for a clear, simplified security model, using Teams for group-based permissions, and more advanced tools like Microsoft Purview for enhanced data governance. This keeps Copilot helpful without compromising privacy.

For organisations dealing with sensitive data, like healthcare, Steve cautions that predictive text in forms can sometimes feel intrusive or inappropriate. In such cases, disabling Copilot, or refining its scope, is a valid approach

4. Look out for advances in custom agents and autonomy

Looking ahead, Microsoft is investing heavily in customisable agents within Copilot. These agents can be trained to perform specific tasks, access resources and even respond to natural language queries. Steve cites the example of employee onboarding or training: imagine a new starter asking, “can I wear trainers in the office?” and instantly getting a precise answer from the staff handbook.

In addition, agents will soon be able to create and train other agents, unlocking a new level of automation. Steve predicts that Copilot will increasingly work in the background – surfacing leads, creating tickets or flagging key updates – without the user needing to prompt it. Much like how spell check quietly improved productivity, Copilot will become a silent partner in business operations.

In summary: Don’t wait, just start

Steve’s closing message is simple: just have a go. Set up a sandbox, test some data and start exploring what Copilot can do. AI isn’t going away – and the sooner you engage with it, the faster you’ll unlock value from the tools you already pay for.

For help getting started or evaluating your data readiness, reach out to the TrueNorth team.