How to understand prompting with Copilot for Microsoft 365

When I was a kid watching Star Wars, do you know what fascinated me the most? It wasn’t the laser guns or hyperdrive, but the robots like C-3PO who seemed to know everything and could be conversed with like a regular, albeit slightly nerdy, human. (I guess my future as a computer geek was foretold early on.)

So it’s exciting to see that kind of science fiction somewhat turn into reality with the new wave of generative AI.

As someone who spends day in and day out in Microsoft Office, Microsoft 365 Copilot is such a cool assistant that I can communicate with at my level to help with tasks ranging from explaining Excel formulas to summarizing meetings for me.

If you remember Star Wars, C-3PO and other robots often needed to be spoken to in a slightly different way. They lacked some of the human nuances, intuition, and emotion that guide human-to-human interaction. This mindfulness in giving the robot instructions in a way that it can easily understand and provide reliable results back is called prompt engineering.

In this post, I’d like to explore a few ways to optimize your prompts and enhance your prompt engineering skills with Microsoft 365 Copilot.

The four elements of Copilot prompting

Microsoft, on its official Copilot Lab website, offers insights on how to get the most out of prompts in Copilot. While there are numerous great resources out there for prompt engineering at different skill levels, it seems like a good idea to start here at the proverbial “horse’s mouth.”

Let’s take a look at how Microsoft advises us to craft the perfect prompt.

Goal

First, what do you want from Copilot? This could be anything from an outline of a PowerPoint deck to a 3-5 bullet point list to an Excel formula. The more specific, the better. AI tools can tend to ramble if you don’t provide solid constraints, leading to a kind of “wet noodle” output—not exactly what you were looking for.

Context

Next, explain your audience. Who are they? What do they need it for? This could be for yourself, to understand a VBA macro, for a client looking to grasp a financial plan, or for a manager wanting to know what happened in a meeting.

Expectations

Here, you can add the human touch to your prompt results. Tell Copilot what kind of voice and tone you want. Should it be humorous, professional, diplomatic, or something else?

Sources

Finally, let Copilot know what source files it should use. Remember, Copilot has been pretrained on a vast amount of data, but if you want it to also consider specific sources—like your emails over the last week or a particular Excel workbook—let it know. This ability to work with your own data in a secure format is one of the best parts of Copilot, so make sure you take advantage of it!

Example prompts

• “I need a 3-5 bullet point summary of our recent marketing campaign’s performance. The audience is the marketing team, and they need this for a meeting presentation. Please use a friendly and encouraging tone. Refer to Marketing Campaign – July 2024.docx report and my emails from the past two weeks for details.”
• “Develop an Excel formula to calculate the weighted average cost of capital (WACC) for a client’s portfolio. This formula will assist the financial analysis team in evaluating investment opportunities and advising on capital costs. Produce a clear and accurate formula with a professional tone, including comments within the formula to clarify each component for those unfamiliar with finance. Utilize data from the client’s financial statements in the Client_Data_FY2024.xlsx workbook.”
• “Create a detailed outline for a training session on Power Query basics. The audience is a group of new hires in the data analytics department. The tone should be instructional and supportive. Use Power Query Training Manual.docx as the primary source of information.”

Additional pointers

In addition to using this basic four-part mental model for prompting Copilot, here are some things to keep in mind:

1. Don’t expect to get the perfect result on the first try. Experiment with different phrasing and levels of detail to see how Copilot responds. Refine your prompts based on the feedback and outputs you receive.
2. Avoid jargon and ambiguous terms. Use straightforward, professional language that is easy to understand. This helps ensure that Copilot interprets your request accurately.
3. If your task is complex, break it down into smaller, manageable parts. For example, instead of asking for a comprehensive report in one go, start with an outline, then ask for details on each section.
4. Offer examples of what you’re looking for to guide Copilot in understanding your preferences better. For instance, include a sample summary or a template to follow.
5. After receiving an output, evaluate it and provide feedback. Adjust your prompts based on what works and what doesn’t. This iterative process can help you fine-tune your interaction with Copilot.
6. Develop a set of template prompts for frequently performed tasks. This can save time and ensure consistency in the results you get from Copilot. Microsoft even has some starter prompt templates available at Copilot Lab, where you can even save your own prompts.

As you can see, while Microsoft provides a structured starter prompt structure, much of the work will come down to trial and error. Interact and iterate on your prompts using feedback and examples, just like problem-solving in most other domains. Be proactive and experiment with Copilot—you won’t break anything.

Conclusion

This post served as a starting point for crafting and building effective prompts for Microsoft 365 Copilot. After everything you’ve learned here, as interesting, powerful, and enjoyable as it is to learn prompt engineering techniques, perhaps nothing surpasses the collaboration between a highly trained, well-communicated subject matter expert and an AI.

After all, at their best, these tools serve as complements to human intelligence, not substitutes. It’s unwise to believe that crafting the perfect prompt alone can consistently yield the correct answers. Knowledge seldom operates on a simple plug-and-play basis; it often requires human intuition and ingenuity.

So, don’t become overly fixated on mastering prompt engineering. While it can definitely provide useful keywords, patterns, and templates to enhance your AI interactions, if you are a trained subject matter expert, you’re likely already adept at articulating your needs, your goals, and your previous attempts—often to those who may not grasp the nuances of your deep expertise.

Learn a bit about the tools of the trade in prompt engineering, but also recognize that this won’t suddenly make you an expert in Excel, data analysis, or any other field for that matter.

Perhaps the best resources for working with Microsoft 365 Copilot come from Microsoft! In particular, I’d point you once again to their Copilot Lab, which has excellent resources, such as the following cheat sheet on the “Art and Science of Prompting.”

A couple of my other favorite resources on prompt engineering include Gilbert Mizrahi’s Unlocking the Secrets of Prompt Engineering (Packt, 2024) and the Awesome ChatGPT Prompts repository on GitHub. Keep in mind that these resources are not specific to Copilot and are a bit more technical, but they will definitely help you understand how to use prompt engineering and LLMs to get high-quality outputs. Remember, you’ll get the best results when you’re an expert in that area yourself!

What questions do you have about prompting in Microsoft 365 Copilot? Please let me know in the comments.