Copilot is a super powerful tool but if you are not getting the results you expect, you might be “using it wrong”. This is based on my experience working with organisations large and small over the past year or so.

Momentum continues to increase and we are seeing more and more public case studies show casing the value of Copilot for Microsoft 365. These focus on the “huge” time savings and efficiency gains organisations such as Barnsley Council, Clifford Chance, and Hargreves Lansdown experienced and how it’s now a tool their employees would not be able to give up.

Getting to this stage is not as simple as just allocating a license and turning it on. The “view” of Copilot I see in many organisations, is very different and as such is so important organisations get off on the right foot with Copilot to ensure that their early experiences with Copilot go well, they understand and evaluate where and how it fits and also where it doesn’t (or doesnt yet).

It all starts with the right expections

It’s hardly suprising that expectations of what Copilot can achieve are high. It’s not that they shouldn’t be, but we need to remember that most of what we see online and in demos are staged, based on the “perfect use case”, data in the “right place” and with Copilot embedded across all your apps and services.

You need to make the cool aid before you can drink it.

There are two scenarios with Copilot which I commonly see.

First there are organisations that grab a handful of licenses, allocate them to a bunch of people and then expect Copilot to know everything about everything and do anything you can think of – setting themselves up for a fail or “less successful” trial.

Secondly, there are those that follow the general guidance shared by Microsoft and their Copilot partners, who build a pilot team, annonce the trial (and what they expect of the pilot users), train and educate their users and share feedback amongst the teams to ensure everyone learns from each other before they expand it to more people.

Whilst the second approach, almost always succeeds (assuming the pilot team are engaged and are open to sharing), I still find (in both scenarios) that part of the reason Copilot trials/pilots can fail quickly is that, in my experience, “people” try to go from zero to hero and rather than looking for many small gains, are trying to get Copilot do that one big thing that will totally change their workload and save them hours of manual effort per day or week.

I find this problem goes away in the main, where organisations are working either directly with Microsoft or via their Copilot Adoption Partner, since the specialists they work with have the time to work with the employees to not only coach them through how and what Copilot can do, but also where it (on its own) may not be the answer they are looking for.

So in short, I find that, many people are simply using Copilot wrong, misunderstand what it can and can’t do and then simoly give up on it and go back to what they did before.

It’s not Copilot it’s you!

In this blog, I am going to share the most common Copilot mistaken use cases that make people the most frustrated….and how to fix them….

1. Trying to use Copilot as an automation tool

This is very common. I often see people who have watched the Microsoft marketing videos (often sequence shortened) contour up an image in their mind that Copilot will solve all their business challenges. I see organisations map out end to end business processes they “would like” Copilot to solve for them that are, to be frank, sometimes outside of what Copilot is really designed to (or is able) to do.

Copilot is not an RPA (Robotic Process Automation) tool. Yes, it can do many many things for you – such as reviewing documents, summarising meetings and actions and creating other “things” from them such as summaries, presentations or new documents.

One of things I hear a lot (and experience) is often that Copilot doesn’t (or wont) do things in exactly the way I ask it to and it can be inconsistent if I ask the same thing twice.

This is true and it is one of the “behaviours” of Copilot (and other Generative AI tools like ChatGPT or Gemini). I do find the more specific I am, the more “similar” my responses from Copilot become, but it’s serves back to the point that Copilot is not designed to carry out multi-step activities with a fixed predetermined output. It is also not currently possible to schedule or trigger Copilot based on an event or other trigger. See Prompting.

Tip:
Understanding the value and use cases of where Copilot fits into the business process and workflow is key to getting the best results. There are many areas where Copilot will provide real business value (either as a first eye, second opinion, note taker, author or creator). Working with a Copilot specialist partner or building a team of them internally will help the rest of the business get used to working “with” Copilot as the “assistant” it is expecting to be.

2. Not understanding Copilot’s Grounding Capabilities

One of the most common pitfalls in adopting Copilot is overlooking its grounding capabilities and limitations. Grounding is the term that refers to the context of the data Copilot can access/will access to perform the request you pass it.

One of the unique values of Copilot for Microsoft 365 over “other” Generative AI experiences, is its deep-rooted access, context awareness and understanding of your business data [through the Microsoft Graph] to enhance its functionality and response. This means that, in short, you can ask Copilot about a particular file that someone shared or a meeting you had last tuesday and it will understand the context and content of that scenario and base it’s response around that. This is hugely value and one of the reasons it has so much value in the workplace. You can also point Copilot at specific content that is within your organisation (though these connected services) or via a URL – for example web content.

The limitation of this grounding is also it’s benefit, in that for Copilot to be able to interface with your data (or application services), that data needs to be connected to the Microsoft Graph through either a connector or plug-in. These are sometimes provided (at cost) by the third party provider or can be built by your devops team or your partner.

This is important because if your employees work off data that is stored primarily in a Document Management System (DMS) that is not connected to Microsoft 365, then Copilot will not have visibility over it or it’s data meaning that employees would need to manually copy and paste data from it, to Copilot in order for Copilot to do anything with it.

Tip:
When running your initial or later stage Copilot rollout, it’s important to ensure you understand where and how people work in different teams so that you know what data people work on and where it is stored.

This grounding process is key as it allows Copilot to build upon the foundation of previous work, such as project histories or meetings. To maximise Copilot’s potential, it’s important to ensure that relevant data is accessible (through connectors or plug-ins) within Microsoft 365 or stored / migrated to Microsoft 365 (OneDrive, SharePoint etc.). For Copilot to work on meetings, you also need to be using Teams and must ensure they are recorded and/or transcribed.

3. Over simplifying or over complicating Prompts

Another key to Copilot’s effective use is the quality of prompts provided. Unlike an AI powered smart speaker, Copilot can handle really complex and specific requests , allowing you to write up to 2,000 (soon to be 8,000) characters in its input boxes.

This means that prompts can well thought out, detailed and quote refined, leading to more specific and sophisticated responses. I have written other blogs on prompt perfection if you wish to read them.

That said, the key when adopting and getting to know Copilot is to encourage your people / teams to experiment with prompts, iterating them and and refining them to achieve the desired outcomes whilst rememering that the same prompt will rarely generate exactly the same answer.

Whilst simple prompts are often to vague, not specific enough and therefore may not yield the desired/expected results, prompts that are too complex can also be ambiguous, conflict or go “off topic” so getting the right balance is needed.

Tip:
Another way to get what you wanted it to make your prompt a conversation: Copilot support interative responses meaning that you can have a conversation with it. Writing the perfct prompt is not always easy to do and can be frustrating so feel comfortable in having a conversation with Copilot. You may find the iterative process becomes fundamental to how you work with Copilot and perfecting the response.

You see unlike human assistants, Copilot thrives on repeated, slightly altered instructions to refine its output. This approach requires a shift in mindset from one-shot solutions to ongoing dialogues with Copilot. By embracing this iterative process, businesses can tailor Copilot’s responses to their specific needs more effectively.

4. Not feeding Copilot properly.

One of things I often find when working with Copilot on more “complex” prompts and tasks is that you can tell it to base its response or output based on a specific file that contains the things you want Copilot to do / check / review for me. Whilst Copilot can leverage the Microsoft Search (via Graph API), if you know what file(s), meeting, person etc, you want Copilot to leverage, you can include that in your prompt.

In the example below (a real one that I was working with a customer on), you can see that we are using to Copilot to review some CVs against a Job Description and Criteria document.

In this example, I am “feeding” Copilot the CVs I want it to review along with a Job Requirements document which contains a table of criteria about the role and the scores available based on this criteria. This means I can focus my prompt on what I need Copilot to do for me using existing data which is used to “ground” it’s response.

Tip:
One of the bug bears I hear alot around Copilot is why it can’t fill out a form for me. It cant. However…. you can ask Copilot to create an output for you based on an existing document. The video below shows an example of doing just this.

Conclusion

The integration of Copilot for Microsoft 365 presents a transformative opportunity for businesses.

By educating employees, understanding its grounding capabilities, mastering effective prompting, practising an iterative approach, and implementing a strategic adoption plan, businesses can position their organisations to really see the benefits of Copilot.

I hope the tips above help – welcome your questions and views.