Published 5th August 2026
Many organisations we work with are approaching Microsoft 365 Copilot with the same question: “How quickly can we roll this out?”
I can understand the enthusiasm – AI has quickly moved from experimentation to a proven priority for adding value and improving productivity.
But after working across intranet, governance, migration and Copilot readiness engagements, I have seen that many organisations are often missing a more fundamental challenge. The more important question is: “How confident are we in the information Copilot will use?”
Because Copilot does not create organisational knowledge. It works with the content, structure, permissions and governance practices that already exist throughout Microsoft 365. If information is difficult to find, poorly maintained, duplicated across multiple locations or owned by nobody, AI does not eliminate those problems. In many cases, it exposes and amplifies them.
In my experience, information architecture is rapidly becoming one of the most important predictors of Copilot success.
Our Principal Consultant, Thomas Lalor, shares key lessons from Copilot readiness projects, highlighting why content governance, search, permissions and information architecture are critical to unlocking value from AI
Why information foundations matter
When organisations begin preparing for Copilot, the conversation often centres on technology. Our clients want to discuss topics like licensing, security controls, agent development, user training and change management. All these things absolutely matter.
However, the organisations that seem to gain the most value from Copilot are rarely the ones that just focus on the technology. They are typically the ones that have spent time improving their information foundations.
They know who owns important content, which content is authoritative or trusted, how their information is organised, who has access to what, and what governance processes are in place. In other words, they have confidence in the strength of their information ecosystem.
That confidence becomes increasingly important when users begin relying on AI-generated answers to make decisions, prepare reports or find organisational knowledge.
The patterns we are seeing
Across projects that Engage Squared works on, a handful of information architecture challenges appear time and again. The platforms, technologies and organisational structures might vary, but the patterns are remarkably consistent.
Nobody owns the content
We see so much content that has been published out to an organisation but over time has been effectively abandoned.
Policies remain online years after they should have been reviewed. Project sites outlive the projects they supported. Teams create new versions of information because they can’t find or no longer trust existing versions.
When ownership is unclear, content quality gradually degrades.
This could often be papered over before AI. Employees would ask a colleague, send an email or spend extra time validating information. However, the challenge is different when people begin interacting with the underlying information through a conversational interface. There’s an implicit assumption that the responses can be trusted. If they’re wrong, it can have real consequences for the business. And if users feel they cannot trust the underlying information, they will quickly lose confidence in the answers being generated.
Search isn’t trusted
One question that I have found to be very revealing to ask when assessing Copilot readiness for an organisation is this: “Can people find what they are looking for and trust the results they see when using search?”
If the answer is no, there is usually a deeper issue. A poor search experience is often a symptom of broader information architecture problems such as inconsistent content placement, poor lifecycle management, poor metadata and incorrect permissions.
Permissions have evolved rather than been designed
Another common challenge is permission sprawl.
Many environments have accumulated years of changes, broken inheritance, exceptions and one-off decisions. Short-term convenience is chosen over good governance. Site owners leave or move, teams restructure, content owners change, but access remains.
By the time organisations want to turn Copilot on, they often discover they are less certain about information access than they expected. This creates understandable concern. After all, one of the most common fears that executives have about AI is: “Will it let people see things they shouldn’t?”
The reality is that Copilot respects existing permissions, however the true problem that we frequently encounter is that it exposes permission management issues that have unknowingly existed for years.
Metadata isn’t achieving what it’s intended for
I have also seen organisations invest significant effort designing metadata models that are rarely used in practice.
Good metadata is valuable because it provides context, improves discoverability and supports governance objectives. This is true for your employees as well as for Copilot. Good metadata can improve both the accuracy and completeness of Copilot responses. When content is tagged with rich, consistent metadata, Copilot can more precisely retrieve, rank, filter and ground answers in the most relevant and up-to-date information.
But successful metadata strategies tend to be the ones that are more pragmatic than overly ambitious. The goal is not to capture every possible property for every item – it is to help people find, understand and manage information more effectively. In our experience, increasing the number of metadata properties that need to be captured only makes it increasingly likely that content owners will start to skip that step altogether.
We recommend identifying and prioritising a core minimum set of metadata to capture for all content – things like content type, content owner, audience and review date – and capturing other properties for specific content types only when use case and benefits are clear.
Why this matters more in the AI era
Poor information architecture is not a new problem. The difference today is in the impact that it has. Historically, poor information architecture has created friction – employees spending longer searching for information, some duplication of effort occurring, teams recreating content that already existed elsewhere.
Annoying and inefficient, certainly, but not often mission critical. AI changes that. When users begin asking questions in natural language and receiving synthesised responses, the quality of the underlying information environment directly determines if the information those users receive and act on is accurate and reliable.
Where you should focus first
The good news is that you do not need perfect information architecture before adopting Copilot. In fact, I have never seen a large organisation with perfect information architecture. The goal doesn’t need to be perfection – we just need to be able to provide confidence.
A practical starting point includes five areas.
- Clarify content ownership
Identify who owns critical information assets and who is responsible for reviewing them. Ownership drives accountability, and accountability drives content quality. Use this information to establish regular ongoing content review processes.
- Address obvious content debt
Look for outdated content, duplicate repositories, abandoned sites and unused workspaces. You don’t need to solve every issue immediately, but reducing known problem areas can significantly improve trust. Usage analytics can help to prioritise key focus areas.
- Simplify metadata
Metadata should help people make decisions and find information. If it feels complicated to maintain, adoption will suffer. Focus on business value rather than technical completeness.
- Review permissions
Understand who has access to critical content and whether those permissions remain appropriate. AI readiness often provides an excellent opportunity to address long-standing permission issues.
- Measure search confidence
Before introducing AI, understand how users currently engage with information. If your employees already struggle to locate trusted information, improving the information experience should be part of the Copilot strategy. Search analytics can help you understand what people are searching for, and what they are not finding.
Final thoughts
There is a lot of excitement around Copilot, and rightly so. The technology has the potential to fundamentally change how people discover information, create content and complete work.
But the organisations seeing the greatest value are not necessarily the ones moving fastest. We see that they are often the ones building confidence in the information environment that sits beneath Copilot.
Because while AI may change how people interact with information, it does not remove the need to manage that information well. If anything, it makes it more important than ever.
About the author:
Thomas Lalor is a Principal Consultant in Engage Squared’s Modern Work & Security practice, specialising in digital workplace transformation, Microsoft 365, governance, and employee experience. Having led large-scale projects across a range of industries, Thomas combines strategic thinking with hands-on delivery expertise to help organisations build workplaces that are more connected, efficient, and ready for the future. He is passionate about creating solutions that balance technology, governance, and user experience, ensuring organisations can confidently embrace modern ways of working and the opportunities of AI.
