Key takeaways
- “Prompt engineering” sounds technical, but for business users it is simply learning to ask Copilot clearly, a skill anyone can master.
- A strong prompt has five parts: goal, context, source, format and tone. Microsoft’s own framework is goal, context, expectations and source.
- Naming the source (the specific file, thread or dataset) is the single biggest lever for accurate, grounded results.
- Expect a conversation, not a one-shot answer. Iterating on Copilot’s first response is where the best results come from.
- A shared prompt library turns one person’s good prompt into everyone’s productivity, and Copilot Prompt Gallery makes it easy.
Prompt engineering is not as technical as it sounds
The phrase “prompt engineering” can be off-putting. It sounds like something for developers, not for the people in finance, fundraising, operations or reception who simply want Copilot to help them get through the day. But for business users, prompt engineering means nothing more than this: learning to ask Copilot clearly enough to get a genuinely useful answer.
That is a skill, and like any skill it can be learned quickly. The difference between a vague request and a well-formed one is the difference between a generic, disappointing response and a draft you can actually use. The encouraging part is that you do not need any technical background. You need a simple structure, a handful of examples, and a willingness to refine. This article gives you all three.
For SMEs, corporates, and non-profits, this matters enormously. You have likely invested in Copilot licences. The return on that investment depends almost entirely on whether your staff can prompt well. Two organisations with identical licences can get wildly different value, and the difference is rarely the technology.
It is the prompting.
The anatomy of a good prompt
Microsoft’s own guidance describes a strong Copilot prompt as having up to four parts: a goal, context, expectations and a source. In practice, we find it helpful to talk about five plain-language ingredients, because separating “format” and “tone” out of “expectations” makes prompts easier to write. Think of it as a simple recipe:
- Goal: what you want Copilot to actually do. This is the only truly required ingredient. “Draft an email…”, “Summarise…”, “Create a table…”, “Analyse…”
- Context: the background that helps Copilot get it right: who the audience is, why you need it, any constraints. “…for our board, who are time-poor and non-technical…”
- Source: the specific information Copilot should use: a named file, an email thread, a dataset, a meeting. “…based on the attached Q3 finance report…”
- Format: how you want the output structured: an email, three bullet points, a one-page summary, a table with named columns. “…as five short bullet points…”
- Tone: the voice you want: formal, friendly, concise, encouraging. “…in a warm but professional tone.”
You do not need all five every time. A clear goal alone will get you something. But the more of these you add, the closer the first response lands to what you wanted, and the less time you spend rewriting.
Source is the ingredient that matters most
Of the five, one deserves special attention: the source. Because Copilot for Microsoft 365 is connected to your organisation’s own data through Microsoft Graph, telling it precisely which information to use is the single biggest lever you have over accuracy.
Compare these two prompts:
- “Summarise the latest project update.” Copilot has to guess which project, which update, and where to look. The result is often generic or simply wrong.
- “Summarise the email thread from Priya about the Westfield project from this week, focusing on the decisions we made.” Now Copilot knows exactly where to look and what you care about.
Naming the source does two things. It dramatically improves accuracy, because Copilot is working from real, specified material rather than its best guess. And it improves traceability, because you can check the answer against the document you pointed it at. When people complain that Copilot gives vague or made-up answers, a missing or unclear source is the most common cause.
Before and after: Small changes, big difference
The fastest way to improve is to see weak and strong prompts side by side. The strong versions are not longer for the sake of it; they simply add the missing ingredients.
| Weak prompt | Stronger prompt | Why it is better |
| “Write a summary.” | “Summarise the attached board paper in one paragraph for non-technical trustees, in a formal tone.” | Adds source, audience, format and tone. |
| “Help me with my email.” | “Draft a friendly reply to Tom’s email thread declining the meeting and proposing two alternative times next week.” | Names the source and the exact outcome. |
| “Analyse this data.” | “Using the ‘Q1 Sales’ sheet, list the top five products by revenue in a ranked table.” | Specifies the sheet, the question and the output. |
| “Make a presentation.” | “Create a five-slide overview of our annual results for staff, focusing on three achievements and our priorities for next year.” | Sets length, audience and content. |
Notice that every “stronger” prompt is still plain English. No jargon, no special syntax: just clarity about what you want, where to find it, and how it should look.
Iterate: expect a conversation, not a one-shot answer
One of the most common mistakes business users make is treating Copilot like a vending machine (one prompt in, one perfect answer out) and giving up when the first response is not quite right. The best results almost always come from a short back-and-forth.
Copilot remembers the context of your conversation, so you can refine without starting over. After the first response, simply steer it:
- “Make it shorter and more direct.”
- “That’s too formal, warm it up a little.”
- “Add a sentence about the deadline.”
- “Turn the third point into a table.”
Think of the first response as a draft to react to, not a final answer to accept or reject. That change in mindset, treating the first answer as a starting point rather than a verdict, is what separates people who find Copilot frustrating from those who find it indispensable. It is also more forgiving: you do not have to craft the perfect prompt up front, because you can always nudge the result into shape.
Examples across the app you use every day
Copilot appears in the Microsoft 365 apps your staff already use. The prompting principles are the same everywhere, but the practical applications differ. Here are grounded examples for each.
Word
Copilot appears in the Microsoft 365 apps your staff already use. The prompting principles are the same everywhere, but the practical applications differ. Here are grounded examples for each.
Excel
Excel’s Copilot helps you ask questions of your data in plain language. – “Using this table, which three regions had the lowest margin last quarter?” – “Add a column that calculates the percentage change between the two years.” – “Create a short summary of the main trends in this data.”
Outlook
Outlook’s Copilot is excellent for summarising threads and drafting replies. – “Summarise this email thread in three bullet points and tell me what’s outstanding.” – “Draft a polite reply confirming I’ll attend and asking for the agenda in advance.” – “Catch me up on emails from the finance team this morning.”
Teams
Teams’ Copilot shines at recapping meetings and conversations. – “Summarise the decisions and action items from this meeting.” – “What questions were raised that we didn’t answer?” – “Give me the follow-up tasks assigned to my team from the last two project meetings.”
In every case, the same pattern applies: state the goal, point to the source, and say how you want the output to look.
Common mistakes, and how to avoid them
A handful of habits account for most disappointing Copilot results. Once your staff recognise them, results improve immediately.
- Being vague about the source. “The report” could be any of a dozen files. Name it, or describe it precisely.
- Asking for too much at once. Multi-part, sprawling requests confuse the response. Break a big ask into a sequence of focused prompts.
- Leaving expectations implicit. If you do not say “three bullet points” or “formal tone”, you will get Copilot’s default, which may not match your need.
- Giving up after one try. The first answer is a starting point. Refine it.
- Accepting answers without checking. Copilot can occasionally get things wrong or fill gaps with assumptions. Always review important outputs against the source, especially numbers, names and commitments.
That last point matters most for trust. Treating Copilot as a fast, capable assistant whose work you still sense-check is exactly the right posture: productive, but never blindly reliant.
Build a prompt library so good prompts spread
When one person discovers a prompt that works brilliantly (the perfect way to draft a client update, summarise a board paper or tidy a spreadsheet) that knowledge is far too valuable to stay with one person. A prompt library captures your organisation’s best prompts so everyone can reuse them.
Microsoft makes this straightforward. Copilot Prompt Gallery lets users discover, save, edit and share prompts directly within Copilot, so a great prompt becomes a reusable asset rather than a one-off. Beyond the built-in gallery, many organisations keep a simple shared document or SharePoint page of “prompts that work for us”, grouped by task or department, written in plain language, ready to copy, adapt and run.
A good prompt library does three things. It flattens the learning curve for new and less confident staff. It encodes your organisation’s tone and standards into the prompts themselves. And it turns prompting from an individual knack into a shared organisational capability, which is exactly where the return on your Copilot investment comes from.
Getting your team prompting well
Prompt engineering for business users is a communication skill, not a technical discipline: be clear about your goal, point to the right source, say how you want the output to look, and refine from there. Teach your staff that simple structure, give them examples from the apps they use every day, and capture what works in a shared library. Do that, and the same Copilot licences that feel underwhelming in one organisation become genuinely transformative in yours.
At 365 Architechs, we help Australian SMEs, SMBs, corporates, and non-profits get real value from Copilot, through practical adoption support, hands-on prompting training for everyday staff, and prompt libraries tailored to how your organisation actually works.
Want your team getting more from Copilot? Contact 365 Architechs about practical prompting training and Copilot adoption support.
