Not invented examples. Six real, named people describing the specific, unglamorous task they pointed Copilot at, and what actually changed because of it.
Six real, named people used Microsoft Copilot for something genuinely unusual, and every example here traces back to a Microsoft customer story, a Microsoft WorkLab interview, or an independent Windows Central interview rather than an invented composite. A disability charity employee spends less effort fighting inaccessible technology, a creative agency compresses weeks of visual feedback into a single working session, a sports marketing agency builds an agent that matches clubs with brand sponsors, a writer troubleshoots technical problems outside his expertise, a video editor generates religious imagery that actually reflects his congregation, and a CIO built Copilot into the first five minutes of his morning. None of them started by asking how to use more AI. Each one pointed Copilot at a specific, repeated friction point in their own work, which is the pattern worth stealing rather than any single workflow.
There is a certain kind of AI article I have stopped trusting. You know the one: “17 WILD ways people are using Copilot,” and then you start reading and the wild examples are write an email, summarize a meeting, make a PowerPoint, brainstorm some ideas. I teach Microsoft Copilot, so I spend a lot of time looking for examples that make people think differently about what the tool can do, and one thing I have learned is that finding genuinely interesting, traceable use cases is harder than you would think. A lot of examples online are plausible. That does not mean someone did them.
So for this article, I wanted a higher bar. I went looking for named people doing specific things with Copilot that I could trace back to a published interview or customer case study, and I found some good ones: a disability charity using Copilot so employees can spend less effort fighting inaccessible technology and more time doing the analytical work they were hired to do, a creative agency that compressed the gap between an idea and something visual enough for a client to react to, a sports marketing company building an agent that matches brands with football clubs, a writer who used Copilot to troubleshoot Docker, WordPress and reverse-proxy problems far outside his day job, and a video editor who used AI image generation to create biblical artwork that reflected the racial identity of his congregation in a way the existing artwork he could access did not.
Those are much more interesting to me, not because they involve secret features. They don’t. What makes them interesting is that each person noticed a specific friction point in their own work and used AI against that. That is the pattern I think is worth stealing, and it is also why one example I originally had lined up for this piece is not here. I could not independently verify it against any published source, so rather than include a compelling story I could not trace, I cut it. More on that in the About section at the end.
Microsoft has made the Copilot naming more confusing than I would like. Today, “Copilot” can refer to consumer Microsoft Copilot, Copilot Chat, Microsoft 365 work experiences, agents built with Copilot Studio, and Copilot capabilities inside products such as Teams, Outlook, Excel and Word. What you can access also depends on your Microsoft 365 subscription and your organization’s licence.
So I am going to label the examples, because seeing someone do something with “Copilot” online does not mean you automatically have the same capability in the version your company gave you. This is one of the first things I check when a client says, “I saw someone doing this with Copilot, why can’t I?” Sometimes it is a prompt problem. Sometimes it is simply a different product.
This one is less flashy and, to me, more consequential. Scope is a UK disability equality charity that supports more than 4.6 million disabled people a year. One of its employees, evaluation officer Nana Kabagwira, has cerebral palsy, which can make typing and interacting with technology harder. Microsoft’s 2026 customer case study describes Scope using Microsoft 365 Copilot for writing, meeting insights, analysis and information retrieval, alongside more than 640 internally created agents.[1]
For Kabagwira, the value is not “AI writes faster.” The value is that she can focus on a meeting rather than splitting her attention between participating and taking notes. Scope also built an internal tool, sometimes called the Hargreaves engine internally, that gives employees real-time feedback on whether a document meets accessibility, tone and clarity standards. There is a line from Kabagwira in Microsoft’s case study that I think gets to the heart of this: “My thinking and analysis can be judged for the work it produces rather than the effort it takes to produce.”[1]
That is such a different way to think about workplace AI. The standard productivity framing asks how we make people faster. This use case asks where the way we designed work is creating unnecessary friction for a specific person. Those are different questions, and when I am mapping AI opportunities with a client now, I try not to only look for tasks that take a long time. I look for tasks that take more effort for some employees than the value of the task itself warrants: meeting notes, formatting, searching through poorly organized documents, re-entering information, turning one file format into another, keeping track of where something lives. These look like small annoyances until you see how unevenly the burden falls across a workforce, and that unevenness is the actual AI use case.
If you have ever worked in marketing or an agency, you know this problem. A group sits around a table, someone has an idea, everyone gets excited, then the meeting ends. A creative team goes away, and days or weeks later the idea comes back as something visual. Only then does everyone realize that is not what they pictured at all.
James Thomas, Global Head of Technology at Dentsu Creative, described using Copilot differently in Microsoft’s WorkLab feature on how business leaders are using Copilot. “We can be ideating, we can have someone in the corner prompting, and by the end of that session, we’ve already got visuals,” he said, adding that the process of turning ideas into images “has been reduced from weeks and months to minutes.”[2] The client now has something to react to while the idea is still being shaped.
That does not mean AI is doing the creative thinking, and that would be the wrong takeaway. The useful part is that the distance between thought and artifact collapses, which changes the conversation. Instead of “we’ll come back and show you what we mean,” you get “something like this,” and the client says “yes, except more like this.” Now the team is iterating against something concrete. I would look for the places in your own work where the next stage cannot begin until someone produces an artifact, a sketch, a prototype, a chart, a draft, a mock-up, a storyboard, a first version of a deck. The value of AI may not be that the first artifact is good enough to use. The value may be that you can react to something sooner, and in creative work that can be far more useful than “write this faster.”
This one made me smile because it is exactly the kind of narrow, slightly strange use case I think organizations should look for. SPORTFIVE is a global sports marketing agency based in Hamburg with more than 1,900 employees across 20 countries. In its 2026 Microsoft case study, the company describes using Copilot across research, outreach, translation, meetings and project work, but one example stood out: colleagues are building agents through Copilot’s Agent Builder, including, in the company’s own words, “a chatbot that matches one of our clubs with a potential brand.”[3]
Think about the work underneath that. Who is this club’s audience, what categories fit, which brands are expanding into this market, who already sponsors competitors, does the brand’s customer profile overlap with the club’s fan base, what campaigns has the brand run before, what would make the partnership strategically interesting. That research can eat hours before the first conversation even starts. The relationship is still human, the pitch still needs judgment, and someone still has to know whether the pairing makes sense, but AI can help narrow a huge possibility space, which is where this gets interesting.
I would look for jobs where a person spends a large amount of time finding the right pair: candidate to role, supplier to requirement, customer to product, expert to problem, investor to opportunity, sponsor to property, mentor to mentee. If the logic behind a good match can be expressed using information you already have, an agent may help with the search and filtering layer, while the person still owns the relationship.
Liam Shotwell is a SaaS content writer, which is what makes his Copilot use interesting. He told Windows Central that he used Copilot while launching a website to help troubleshoot WordPress issues, Docker setups, reverse proxies and related technical problems, saying, “I needed to ask a lot of questions with steps to fix problems in WordPress, get things working in Docker, and set up various other things like reverse proxy configurations.”[4] Those are not copywriting tasks. He was using AI to help him operate outside his core expertise.
I think this is one of the most under-discussed ways AI changes work. The conversation often focuses on whether AI can do part of your existing job. There is another question: what can you now attempt that you previously would not have attempted yourself. A marketer can interrogate a dataset without becoming a data scientist, a finance professional can prototype a small internal tool without becoming a developer, a trainer can build a simple interactive experience without hiring a development team for the first version, and a writer can troubleshoot his own website.
There is an important caveat, and it matters as much as the use case itself. Shotwell also warned against blindly trusting Copilot’s output, saying, “I always say that I put Copilot and other generative AI tools under a microscope because they’ll often be confidently incorrect.”[4] He described using AI for research and learning rather than treating generated material as automatically correct. I would make a list of the tasks where you regularly say “I don’t know how to do that,” then separate them into two categories: tasks where lack of expertise makes AI use genuinely unsafe, and tasks where AI can help you learn enough to move the work forward as long as you verify the answer. That second category is getting much larger.
Kenneth E. Moody, a video and sound editor interviewed by Windows Central, described using Microsoft’s image generation capabilities to produce biblical artwork for his church network. His frustration was specific: much of the religious imagery available to him represented biblical figures with light skin, and he wanted imagery where members of his congregation could see people who looked more like them. “I used Bing Image Creator to depict ideas in the bible and illustrate them,” he said, describing how “I was able to depict Joseph with brown skin, whereas before he is always depicted with very light skin.”[4]
This is a completely different use case from anything else in this list, and that is why I wanted to keep it. AI image generation is usually sold as “create images faster,” which misses the interesting part here. The friction was not production speed. The friction was that the asset he wanted did not exist in the stock library he had access to. Generative AI changed the economics of creating something for an audience too specific for a traditional stock-image catalogue, and that has implications far beyond religious art.
I would look for the long-tail assets your own organization repeatedly wishes existed: a particular customer scenario, a niche industry setting, an unusual product configuration, a training scenario reflecting a specific workforce, a visual for an audience traditional stock libraries poorly represent, a mock-up nobody would have commissioned because it was only needed once. Generative AI makes that long tail cheaper to serve. The review still matters, especially when history, identity, religion or representation are involved, but the use case is bigger than “make me a picture.”
After five examples like that, I want to include one that sounds almost boring. Anup Purohit, Group CIO at Wipro, a technology services firm with roughly 245,000 employees, told Microsoft’s WorkLab that he built Copilot into the first few minutes of his day. “I use it when I wake up to check my calendar, it’s totally reinventing the way I look at my daily schedule,” he said, and he also uses it to summarize long email threads and turn call notes into follow-on actions for his team “within 10 minutes” of a call finishing.[2] Nothing dramatic. He simply inserted AI into the place where his morning already had friction.
I think this is worth including because unusual examples can accidentally teach the wrong lesson. People read about sponsor-matching agents and culturally specific artwork generation and think they need to invent some brilliant AI workflow. You don’t. Sometimes the best use is: every morning, tell me what needs my attention before I start reacting to everything. The fact that the workflow repeats matters more than how impressive it sounds.
Laid out next to each other, the range across these six is a useful reminder that a “Copilot use case” is not one shape.
| Person or organization | What Copilot is doing | The friction underneath it | Source type |
|---|---|---|---|
| Nana Kabagwira / Scope | Reduces note-taking and technology friction for accessibility needs | Accessibility barriers get in the way of the person’s core work | Microsoft 2026 customer case study |
| James Thomas / Dentsu Creative | Turns live ideation into early visuals | Teams wait too long before an idea becomes something people can react to | Microsoft WorkLab interview |
| SPORTFIVE | Matches sports clubs with potential brand partners | A huge research space before a human relationship even starts | Microsoft 2026 customer case study |
| Liam Shotwell | Helps troubleshoot technical systems outside his expertise | Work sits outside the person’s existing skill set | Windows Central interview |
| Kenneth E. Moody | Generates culturally representative religious imagery | The needed niche visual assets were difficult to source | Windows Central interview |
| Anup Purohit / Wipro | Reviews calendar, email and follow-ups every morning | Attention gets consumed before the day properly begins | Microsoft WorkLab interview |
A sourcing note matters here. Microsoft customer stories are first-party vendor case studies: useful because the organizations and people are named and the workflows are described specifically, but I would not treat vendor-reported results as the same thing as an independent study. Shotwell and Moody’s examples come from an independent Windows Central interview.
When I first started collecting these examples, I thought the interesting part would be the range. A disability charity to a sports marketing agency to religious art is quite a range. But laying them beside each other, something else became more interesting: none of these people started with “how can I use more AI.” They started, explicitly or implicitly, with friction. Something takes too much attention, something arrives repeatedly and has to be checked, something takes days before anyone can react to it, something is difficult to find, something falls outside the person’s expertise, something needed does not exist. Then AI gets pointed at that piece.
The best Copilot use cases are usually hiding inside repeated friction you have stopped noticing, not inside a prompt library.
The newest examples also point to a second pattern worth watching: people moving from using Copilot to building with Copilot. Scope reports more than 640 internally created agents. SPORTFIVE is exploring agents that match sports properties with brands. AGCO, the agricultural equipment maker, reports more than 900 employees volunteering to build agents through Copilot Studio, a maker community that has since grown to roughly 2,000 people across the company, with some quality-review workflows shrinking from weeks to about an hour.[5] The first wave of workplace AI was “ask Copilot something.” The next wave looks more like “this process keeps happening, can we give Copilot a defined job inside it.” That requires understanding the workflow, defining the inputs, deciding what a good result looks like, and knowing where human judgment remains, and noticing the weird little friction points everyone else has accepted as normal may be the most important skill of the four.
This is where list articles can get people into trouble. You read the Scope example and think you need an accessibility tool. Maybe you don’t, if nobody on your team is fighting the technology to get their work done. You read the SPORTFIVE example and think you should build a matching agent, but for what? The point is not to collect somebody else’s workflow. The point is to understand why their workflow works: Kabagwira had recurring accessibility friction sitting between her and the work she was capable of doing, Dentsu had an expensive delay between ideas and visual feedback, SPORTFIVE had a matching problem with a huge research space, Shotwell had a knowledge boundary he wanted help crossing, Moody had a niche asset gap, Purohit had attention fragmentation at the start of every day. Those are patterns you can look for in your own week.
I would not ask a team to brainstorm “creative Copilot use cases.” That wording puts pressure on people to invent something impressive. Instead, take one normal week and ask four questions: what keeps happening, since recurring work has a chance to become a workflow rather than a one-off trick; where do I lose time or attention, since the friction is often more useful than the task name; what information already exists that would help, since Copilot becomes much more useful when it can work from real emails, files, meetings or data; and what still needs me, which gives you the human review point before the workflow starts.
Then get specific. “Meetings take too much time” is not specific. “I spend 25 minutes after every client call finding the commitments, writing them into our tracker and drafting the follow-up” is. “Research takes too long” is not specific. “I spend two hours before every prospect meeting finding what changed at the company since our last conversation” is. That is the level where I want people looking, and it is also the level where the review question gets easier to answer. I use a simple sentence to force that last part: Copilot does ________, a person still checks ________ before ________. For example, Copilot flags unusual campaign performance, a performance marketer checks the underlying data before recommending a budget change. Or, Copilot drafts responses to routine member questions, a human handles anything involving a complaint, money or policy exception before a message goes out. If you cannot complete that sentence, the workflow probably needs more design, and the more unusual your workflow becomes, the more important it is to name the check. For Kabagwira, the employee still owns the analysis and judgment. For Dentsu, an early generated visual is something to react to, not the final creative simply because AI produced it quickly. For Shotwell’s technical troubleshooting, verification against reliable documentation matters. For Moody’s imagery, human review matters because representation and historical interpretation carry meaning.
If I were running this as a workshop exercise, I would give everyone ten minutes and ask them to write down the recurring things they did last week that felt unnecessarily manual, then make them get painfully specific in the same way: not “inbox” but “every morning I scan 60 emails to work out which five contain a commitment I need to act on,” not “reporting” but “every Friday I copy numbers from three places into the same slide deck and then write three sentences explaining what changed.” Then I would ask three questions: does Copilot have access to the information, can we describe what it should look for, and can we check the result quickly. If the answer to all three is yes, you have a much stronger candidate than “maybe I should use Copilot to write my emails.”
I would also pick the recurring task that annoys you enough that you already know every step, not the flashiest example on this page, because you will know when Copilot gets it wrong, and that matters when you are learning. A good first workflow is frequent enough to matter, narrow enough to explain, easy enough to verify, and low enough risk that a mistake is recoverable. Run it repeatedly. If it still saves effort after three weeks, keep it. If you find yourself constantly correcting, checking and rebuilding the output, change the workflow rather than forcing AI into the task.
The most interesting Copilot users in this article are not sitting around coming up with clever prompts. They are noticing work: that meeting is harder to process than it needs to be, those reports arrive every morning and somebody has to inspect them, we waste two weeks turning an idea into something the client can see, finding the right sponsor takes hours of research, I know how to write but I do not know Docker, the images I need do not exist, my calendar owns the first 20 minutes of every morning. That is the starting point. Then comes the AI question: can Copilot take some of that friction away without taking away the judgment I still need. That is a far better question than “what is the wildest thing Copilot can do,” and, ironically, it is where the wildest use cases seem to come from.
Yes. Every example in this article is tied to a named person or organization and a traceable published source. The Scope, SPORTFIVE and Dentsu examples come from Microsoft-published interviews or customer case studies. Liam Shotwell and Kenneth E. Moody were interviewed by Windows Central about their own Copilot use. Microsoft customer stories are vendor-published evidence rather than independent experimental research, so I treat them as documented case studies rather than neutral proof of business impact.
Kenneth Moody’s is probably the most unexpected. A video and sound editor used Microsoft’s image generation to produce biblical artwork with brown-skinned figures for his church network, because the religious imagery he had access to before mostly depicted biblical figures with light skin. It has nothing in common with a typical business rollout deck, which is exactly why I wanted to keep it.
No. Copy the pattern behind the workflow, not the workflow itself. Find a recurring friction point in your own work, identify the information Copilot would need to help with it, define what a good result looks like, and decide what still needs your review before anything goes out.
Not exactly. Microsoft uses Copilot branding across several consumer and business experiences, and what you can do depends on whether you’re using free Microsoft Copilot, Copilot Chat, Microsoft 365 work experiences, app-specific Copilot features or agents, and on your organization’s licence. Check what your account actually includes before assuming a use case shown online will be available to you.
Start with repeated friction rather than features. Ask what happens every week, which step wastes time or attention, what source information already exists, and what still requires your judgment. The more specific the problem, the easier it becomes to design a workflow worth keeping, and the easier it becomes to know when Copilot got it wrong.
I wanted this article to meet a higher sourcing bar than the usual “creative AI use cases” roundup. Every use case here is tied to a named person or organization and a source I checked directly: Microsoft’s 2026 customer case studies for Scope and SPORTFIVE, Microsoft’s WorkLab feature “How Business Leaders Are Using Copilot Right Now” for James Thomas and Anup Purohit, and an independent Windows Central interview for Liam Shotwell and Kenneth E. Moody. An earlier draft of this piece included a seventh example, a natural gas company technician using Copilot for turbine anomaly detection, that I could not verify against any published Microsoft case study or news source after extensive searching. Rather than publish a compelling story I couldn’t trace, I cut it. Microsoft customer stories are first-party vendor case studies rather than independent research, and I’ve noted that distinction throughout rather than treating vendor-reported results as neutral proof.
Weekly tips. Real examples. Practical help for busy professionals.
We care about your data, check out our privacy policy.