Copilot is landing: here’s why you need to prepare before you procure

I’m a sucker for software demo videos, particularly in the field of AI.

Marvel as your workload vanishes like magic! Behold the technology that will pull new ideas out of a hat!

And the star of the AI demo video genre is Microsoft.

After ploughing $14 billion into Open AI early this year, Microsoft teased the demo of its Copilot integration to the tools most of us use multiple times a day — Outlook, Word, PowerPoint, Excel and Teams.

This was no AI Clippy.

As later demos have revealed, Copilot is set to transform the way we work, helping to write new proposals and documents in seconds based on past projects, answering accountant-level questions about your data in Excel, and creating beautiful, on-brand presentations at a single click.

Even better, Copilot would attend your meetings for you and read your emails, providing you a neat summary of each so you didn’t have to waste time doing either.

The first demo was unveiled in March.

Now, seven months later, we are about to get the first taste of what Copilot looks like with the release of Windows 11, which has a sort of Copilot lite, that will do some basic tasks like darken your screen on request or answer the kind of questions you could Google. (The example provided by Microsoft is ‘when is the best time to visit Hawaii?’)

Within two months, though, Copilot should be rolled out as a fully-fledged offering for those businesses who already have a licence for Microsoft 365 E3 or Microsoft 365 E5.

To put that in context, Microsoft’s published price for these tiers is between $56 and $85/user per month, although many companies and not-for-profits will pay less. There is also a suggestion by Microsoft that they might accept Business Premium as an acceptable base tier.

In any case, the organisation will need to add on a licence for Microsoft 365 Copilot, which could be another $44 for each user each month.

There is no doubt that if Copilot delivers everything the demos promise, this should be a worthwhile investment — but possibly not just yet.

Microsoft 365 Copilot flowchart

Before we look at why that’s the case, here’s a quick primer on how Copilot works.

Copilot works by connecting Microsoft apps (like Word, OneDrive, SharePoint, Teams and Outlook) with what is known as Microsoft Graph, what it describes as “the connective tissue that binds all your Microsoft 365 services and data together”.

A user who asks a question or uses Copilot to undertake a task in an app will have this input routed through Microsoft Graph for what is known as pre-processing — effectively a step that puts the question into context for your organisation.

The prompt is then routed through a Large Language Model built on OpenAI (but importantly, not used to train the open-source model), before the answer returns to the user.

Microsoft represents this almost instant process as shown below.

Diagram showing a visual representation of how Microsoft 365 Copilot works.

So why might you hit pause on procurement?

Copilot represents a significant shift in the way you will work

Any software change can take longer to roll out behind the scenes than your users will see at the front end. That’s because few system changeovers are simple — and Copilot is no different.

In fact, because it sits at the heart of the system that underpins millions of businesses, integrating the change is likely to be more complex and require greater consideration than just adding in another third-party piece of software.

While some large organisations are being nursed through the beta testing, most businesses have not yet had a chance to review the various changes that Copilot will require, but Microsoft has released an early training program that provides some insights.

Businesses that don’t already have OneDrive will need to add this functionality in to be able to get the best out of Copilot’s file sharing and reference system, and they will also need to upgrade to the New Outlook experience on Windows.

In addition, they will need to have clear permissions set up in Azure to reduce the risk of material being surfaced by Copilot to people who might not normally have access to that information.

In a perfect world, everyone’s document permissions would be right all the time, but the reality is that often privacy is maintained because people don’t know what is out there or how to find it.

Copilot will make it a lot easier for anyone to track down unprotected but sensitive material.

1. Your data is probably dirty

Been in business a long time? Chances are you are carrying around a lifetime of legacy data, outdated guides and standard operating procedures, and superseded information.

When you have to wade through this material manually, it doesn’t really matter. You simply ignore the folder marked “old documents” and carry on your way.

As Copilot will rely on your material to work, though, the pressure is on to clean up the ROT — redundant, obsolete or trivial files that will get in the way of appropriate responses.

Since as much as 30% of an organisation’s data can be ROT, according to Gartner, allowing this material to be stored with your good and useful documents risks reducing the value of Copilot while also costing you more, training a model on something less that useful.

A good early step for organisations, therefore, is to review untouched or superseded folders and either store outside their training environment or delete altogether.

Take this opportunity to also review any customer, client, supplier or personnel data that should no longer be stored and ensure it is not sitting somewhere it could inadvertently be exposed.

This process could take some time.

2. AI is a change management challenge, not just a software one.

A third reason to pause relates to the humans you will be pairing with AI. While most of the organisations we have spoken to have experimented with AI through ChatGPT, suddenly having access to one of the most powerful systems in the world on every platform is an abrupt change.

You will need to ensure your teams are comfortable with the incorporation of AI — and equally that they are not over-relying on it to perform the critical thinking that humans do best.

You will want to ensure they are skilled in the best way to prompt the tools for different responses, and also that they understand the differences between Large Language Model-based generative AI and other kinds of tools.

Generative AI is incredibly useful for many things but it isn’t a calculator, search engine or Wikipedia. It won’t necessarily return the same response even to the same question, so helping your people understand why it answers in a particular way is as important as preparing them to use what it answers.

Our recommendation is to start this process well before you start to add in Copilot, building baseline AI literacy (your AI-Q) before you introduce a robot co-worker.

3. Don’t wait for the demo to come to life. Get started now.

There’s little doubt that Copilot will be a gamechanger for organisations big and small. For some it will level up their teams; for others it will reduce time wasting and allow them to be more competitive and efficient.

Making this magical tool work for your organisation will take human labour, though, and the time to start preparation is now.