The numbers are in: no one is paying for Microsoft 365 Copilot.
No, not the official numbers designed by Microsoft to support their narrative. I’m talking about something that MS would prefer not to address as these numbers tell a whole different story. As a result, they are from a source that we cannot verify and therefore something that every reader needs to evaluate the trustworthiness of.
❝ “A source that has seen materials related to sales has confirmed that, as of August 2025, Microsoft has around eight million active licensed users of Microsoft 365 Copilot, amounting to a 1.81% conversion rate across the 440 million Microsoft 365 subscribers.” Ed Zitron: The Case Against Generative AI
I know, Ed Zitron is hardly a neutral observer of the tech industry. He is doing something that almost no one else out there bothers to do, though. Which is checking whether the numbers released by tech vendors (especially OpenAI) make any sense. This is the foundation upon which his case against the financial validity of GenAI is built on, and being a regular reader of his premium newsletter issues, I find it fairly convincing.
So, when someone on the inside wants to find a channel to bring out the inconvenient truth about AI adoption, Ed is where they would likely reach out to. These numbers that have been disclosed about Microsoft’s commercial success (failure) of selling Copilot licenses are in line with those reported about the adoption rate of other paid AI plans from other vendors.
The story can be summarized as follows:
M365 Copilot became available for enterprise customers to purchase on November 1st, 2023. That was almost 2 years ago. Now, if we assume that the adoption rate is constant, the conversion rate mentioned by Ed could grow to around 2% by the time we reach November 2025.
The 2% adoption rate in 2 years is diabolically bad. This is not just any lil’ Power Apps product that was promoted as a new tool for citizen developers to improve personal productivity. It has been the centerpiece of everything Microsoft has done and talked about for over 2 years now. I have never seen a bigger push for any MS product.
After all this — what do we even have here? 8 million active licensed users. For some tech products that might be a sizeable user base. But let’s be real here: Microsoft has at least 400,000 channel partners. If each partner org would have bought on average 20 seats of Microsoft 365 Copilot for their employees, that would already make up the 8M figure.
Indeed, most partners have to pay for the M365 Copilot seats. There are hardly any freebies available in the partner benefits packages. I pay €800 a year for MS licenses in Partner Success Core and that doesn’t give me a single M365 Copilot seat. I have to pay the full price of €337 per year to get a chance to use the premium Copilot experience in my tenant for my solopreneur Power Platform advisory business.
This, of course, makes it easier to evaluate whether I get value from the license or not. While I acknowledge that I’m not the typical user persona for M365 Copilot, it has given me just a fraction of the value that my ChatGPT Plus subscription does. At a lower price point. Since I don’t have to sell and promote this product to customers, like many MS partners do, I’m free to tell everyone about this experience. As I’ve done in this newsletter for the past year on many, many occasions.
It could have been just me who’s a cranky old CRM consultant shouting at the AI cloud. Well, based on these adoption numbers for Microsoft 365 Copilot, I’m not in the minority. The vast majority of people out there who use Microsoft cloud tools don’t see Copilot as giving them enough value to justify the $30 per month cost.
Remember: even though these aren’t individual users buying things for themselves, we’re not talking about just another MS license. This is AI! It’s the thing that’s “no longer optional” at so many companies — thanks to all the AI CEOs forcing their employees to adopt it or else. We read about these initiatives from media of how business leaders want to transform their company via AI on a daily basis.
There must be loads of companies who have gone and ticked the AI box by choosing to purchase M365 Copilot licenses. And despite all this blind support from management, this is as far as Microsoft has gotten with the paid Copilot product sales.
Agents aren’t doing any better
Another interesting figure from Ed’s newsletter was about SharePoint agents:
❝ “I’m also hearing that less than SharePoint — another popular enterprise app from Microsoft with 250 million users — had less than 300,000 weekly active users of its AI copilot features in August.”
Hmm, let’s see now. What other numbers do we have of agents that we could compare this with? Oh, right! The 3 million agents in FY25 story. Let’s revisit a screenshot I shared in my Numbers designed to please article a couple of months ago:
My theory (as well as Suhail’s) was that most of the 3M agents would be in SharePoint, and they wouldn’t be in production use. Weekly usage could well be considered a threshold for an AI agent being in production, so that would give us 10% when comparing it to the 300k figure in Ed’s newsletter.
But that was about users. Unlike Copilot, which is supposed to be a personal AI assistant, agents are meant to be tools that many individuals tap into with their Copilot. Now, let’s ignore for a moment the crazy fact that you can’t use SharePoint agents in the Microsoft 365 Copilot UI. Out of the 300 million users of SharePoint Online (I don’t know why Ed said only 250M), 0.1% of them interacted with an agent during a week in August 2025. One user out of a thousand.
Can you feel the AI transformation already?
Where does Microsoft go from here?
If this were a normal software product, the industry would have moved on already and invented something else to pitch to customers. In practice, it is anything but normal. GenAI is the only story left for Big Tech to tell when chasing for ever greater growth percentages. And yet the only thing growing are the capex investments.
Getting people to use AI services that they either want to (ChatGPT, all the AI coding assistants) or can’t avoid (Google search AI overviews, free Copilot features in MS products) is one thing. Making them voluntarily open their wallets and insert a credit card to purchase some of the AI magic dust is a whole different game. That’s the game where everyone has been losing so far, as no one besides NVidia or the IT consulting companies is making profits from GenAI at this point.
It’s not like all those GPUs could be left underutilized, though. The growth of especially the US economy relies on AI related activities. If they were to be removed, the bubble would burst. So, rolling back AI features is not an option, understandably. The only possibility is to reimagine the monetization model and hope that it works this time. And that’s exactly what Microsoft is actively pursuing.
First of all, the idea of bundling more of Copilot into the licenses customers already have is the obvious way to boost AI adoption. Which may not bring in any new money, yet will be a critical number for MS to show to investors. Who could soon begin questioning why the capex spend is necessary if the GPU utilization for Microsoft 365’s enterprise Copilot is barely scratching 60%.
Enter Copilot Chat in Office Microsoft 365 apps. Announced on September 15, “Microsoft 365 Copilot Chat and agents are rolling out in Word, Excel, PowerPoint, Outlook, and OneNote for all users—creating a unified chat experience across the apps millions of people use every day at work.” What was once the main demo scenario of why contextual AI inside the business productivity apps is crucial for worker productivity is now something all 440 million M365 users get for no extra cost.
Copilot Chat pane in Outlook, PowerPoint, Excel, Word. For everyone.
The nice side effect from this replacement of app specific Copilots with the generic Chat has been that premium M365 Copilot users seem to have lost all of the smarts that their apps used to have. Even M365 Copilot MVPs on social media are anxious about the update that makes Copilot responses dumber by default. I of course raised my concerns about the same useless AI assistants issue a couple of months ago already. I’m glad it’s not just me who is too dumb to get Copilot in Outlook to work with email and calendar.
Then, there’s the next push: vibe working. What has been disguised as a marketing campaign to ride on the popularity of the vibe coding concept is actually an indirect admittance of the fact that GPT based M365 Copilot has been a disappointment for everyone. That’s why Microsoft is now teaming up with OpenAI’s biggest competitor Anthropic and building its new Office Agent features on top of Claude:
❝ “Today we’re introducing Office Agent, a multi-agent system that builds upon an open-source stack, Anthropic's Claude model, and a new taste-driven development (TDD) paradigm to deliver polished PowerPoint presentations, ready-to-use Word documents, and soon Excel spreadsheets.” Office Agent – “Taste driven” multi-agent system for Microsoft 365 Copilot
All of that is preview stuff, of course. When MS announced they were expanding model choice in M365 Copilot and Copilot Studio, all the Modern Work consultants on LinkedIn were quick to point out that enabling Claude breaks the MS trust barrier for customer data. Well, tech industry reporters have been talking about how Charles Lamanna & crew are embracing Claude for a while now, so I see it as only a matter of time before models like the new Claude Sonnet 4.5 are running on Azure in addition to AWS. Because otherwise the Office Agent won’t be as good with spreadsheets and docs like competing products that tap into the best LLM for the job.
How good is Office Agent then? Well, I did a quick test drive of the Agent Mode in Excel. Meaning, I installed the Excel Labs add-in that seems to not require any Frontier program sign-up, contrary to what MS documentation says (that Copilot Frontier switch was erroring out in M365 Admin center for me, so I said “f*** this” and tried the tools without it). Then, I gave it a task: “build me a 5-year investor-ready financial model for a SaaS startup.” Like a user of our FinModeler SaaS app might do.
What was the result? 12 minutes of work, timing out after endless verifications and “finalizing” steps. Some numbers and formulas were added to the Excel workbook, sure. Would I trust this to be the type of financial model that I’d use in finding investors for my business? Hell no. So, it’s definitely a “vibes” kind of an LLM output that looks neat but you can’t build on top of.
The introduction of all these agents that are not yet part of the official M365 Copilot product yet gives an indication of the direction where Microsoft is taking their AI narrative. After forcing consumers to pay for Copilot, redirecting the Office home page to AI chat instead, and making Copilot a toxic brand along the way, now’s the time to double down on agents. “Oh, Copilot, yeah that’s just the UI for AI, but to do anything useful with it you’ll need the latest agents.”
That would be perfectly in line with the upcoming Microsoft 365 Agent product announcement that’s supposed to take place at Ignite 2025. Given how quickly things change in the MS AI roulette table, it might of course not arrive in the format that Charles Lamanna was telling in the internal memo in August. But one thing that I’m increasingly confident of is that we will see a per-agent monetization model emerge. Simply because the per user M365 Copilot commercial model has failed so dramatically.
The latest announcement of “Bring Your Own Copilot” model with the new personal and family Microsoft 365 Premium SKU is not a sign of the premium value of Copilot. On the contrary, the basic AI features in Office apps are now such a commodity that MS is simply creating a new model to allow individual employees to sneak in a paid Copilot license via their personal MS account into work context inside M365 tenants. Since only 2% of the workers had premium Copilot licenses anyway, getting a bigger audience opening the AI tools through any means necessary is now on the table.
And so we get the “BYOAI” model, something that will irritate every single enterprise IT admin, of course. But the thing is: a personal M365 Premium won’t unlock any of the premium features like MS Graph grounding for Copilot. Which should lead everyone to asking: “what exactly is premium anymore”? The answers from this blog post aren’t very detailed nor convincing:
❝ Q: What features are available with personal Copilot subscriptions?
A: Features like Researcher, Analyst, Photos Agent, and Actions may be available depending on the subscription tier. Additionally, users have access to the many in-app Copilot features when using the Microsoft 365 apps like rewrite, summarization, or discussion insights in Word; design suggestions and narrative builder in PowerPoint; and more. Employees can bring Copilot from their personal Microsoft 365 plans to work - what it means for IT
“Copilot Actions?” You mean that Ignite 2024 keynote highlight that was deprecated and removed from M365 Copilot by May 2025 already? Because it was a half-assed attempt at chat driven automation, yet still resulted in a Microsoft 365 environment being provisioned for users of Copilot scheduled prompts today? Which now seem to be superseded with the recent Flow Builder agent that is actually Power Automate in disguise? Which is a product waiting to get reimagined, so that users wouldn’t think automation without AI was ever possible?
"Oh cool! Another way to create automations in M365! And now in a secret Power Platform environment you can't see in Maker portals!"
At this point, no one at Microsoft knows what their plan is — because they don’t have one. The masterplan of transforming becoming The Copilot Company, announced at Ignite 2023, has backfired commercially. Satya’s big bet is not working and now he is inviting Judson Althoff to be the CEO of all things commercial at MS while he focuses on the technical challenges of AI:
The announcement is framed as a “tectonic AI platform shift”. There isn’t any concrete context given on what specifically is not working in the current mode of operation at Microsoft, just the general notion that everyone at the company needs to do better in these unprecedented times. Aside from the key thing about Nadella making way to Althoff, it’s pretty shallow. Here’s how the internal memo ends:
❝ This will also allow our engineering leaders and me to be laser focused on our highest ambition technical work—across our datacenter buildout, systems architecture, AI science, and product innovation—to lead with intensity and pace in this generational platform shift. Each one of us needs to be at our very best in terms of rapidly learning new skills, adopting new ways to work, and staying close to the metal to drive innovation across the entire stack!! This isn’t just evolution, it’s reinvention, for each of us professionally and for Microsoft. Satya
See that “this isn’t X, it’s Y” part there? Yup, that’s classic GPT writing. In practice, Satya Nadella isn’t even writing these announcements that are sent to all MS employees anymore. It’s all just AI talk. If would be much more honest if the memo would have been signed by Copilot instead.
While we wait for it to be reimagined as the CEO Agent, with Copilot as the UI for it, of course.
“This is not a failure!”
After posting this newsletter issue on LinkedIn, I got a lot of comments from Microsoft partners who disagreed with my interpretation of what the 8 million figure means. Hardly a surprise, given how much the business success of people selling and consulting M365 Copilot depends on the product being seen as something desirable.
I’m not saying all is lost for MS. My main point is that whatever they tried to pull off with the $30pupm M365 Copilot license backfired dramatically. Redmond will eventually “reimagine” things in a way that will let them put this failure behind them without ever actually addressing it. But we, the people working with Microsoft tech every day, deserve to know what happened and remember it the next time a similar stunt is played.
To illustrate my point, and give it a perspective that’s not just “my opinion with no facts whatsoever” (as some people on the professional network said), I asked Claude Sonnet 4.5 to crunch the same numbers. To do an analysis of the news sources and factors at play with M365 Copilot as a commercial offering, then evaluating whether it was success or spin: