Introduction: The Gap Between Narrative and Numbers
When Microsoft announced 15 million paid M365 Copilot seats in February 2026, the company framed it as a major milestone. The narrative was clear: Copilot is winning in the enterprise. The technology is delivering value. Organizations are voting with their budget by purchasing licenses at scale.
But here's what the company didn't emphasize: 15 million seats is 3.3% of Microsoft's total addressable commercial market, which stands at approximately 450 million potential users across Microsoft 365, Azure, and related services. After two years of availability, product marketing investment, and bundling incentives, Copilot has penetrated only a small fraction of the possible market.
For comparison, Microsoft Teams reached 300 million monthly active users in less than three years from launch. Exchange Online has nearly 400 million users. SharePoint and OneDrive are ubiquitous across Microsoft's enterprise customer base. Copilot, despite being positioned as the centerpiece of Microsoft's AI strategy, remains a niche product adoption.
This article examines what the real numbers mean. It's not that Copilot is a bad product. It's that adoption is far lower than Microsoft's messaging would suggest, and there are specific, quantifiable reasons why. Understanding those reasons is critical if you're considering whether to upgrade your entire organization to E7, where Copilot is bundled at $30 per user per month.
The 15 Million Paid Seats: Unpacking What That Actually Means
Microsoft reported 15 million paid M365 Copilot seats in Q2 FY2026 (calendar Q1 2026), representing 160% year-over-year growth from approximately 9.4 million seats in Q2 FY2025. That growth rate sounds strong, and it is. But it's growth from a very small base.
The company separately reported 4.7 million paid GitHub Copilot subscribers, which is a different product aimed at developers. GitHub Copilot has higher adoption among developers than M365 Copilot has among knowledge workers, which tells you something important: the technical accuracy and user experience of Copilot for Code is superior to Copilot for general office productivity.
Beyond paid seats, Microsoft also reported 33 million active users on the free tier of Copilot Chat, which is available across multiple surfaces (web, mobile, Teams). This is important context. It suggests that 33 million users have tried Copilot for free, but only 15 million have converted to a paid M365 seat. That conversion rate—roughly 45%—is significant, but it also means 55% of people who try Copilot don't end up paying for it.
Why is this distinction important? Because it suggests a fundamental truth: Copilot is not so obviously valuable that every organization or every user rushes to purchase it. The free tier is sufficient for many users, or they migrate to free alternatives like ChatGPT or Gemini. The paid tier is a choice, not a necessity, and most organizations and users are choosing not to make that choice.
The 3.3% Conversion Rate: Why It Matters for Your Decision
Let's make the 3.3% figure concrete. Microsoft's total addressable market for M365 Copilot includes all users in organizations with Microsoft 365 subscriptions. Based on publicly available data, that's approximately 450 million commercial users globally (across E3, E5, enterprise SKUs, and other tiers).
15 million paid seats represents 3.3% of that market. If Copilot were universally valuable, or even commonly valuable, you'd expect adoption to be higher. You'd expect organizations to be upgrading their entire user bases to E7 or purchasing Copilot licenses across the board. Instead, organizations are selective. They're licensing Copilot for specific groups—knowledge workers, analysts, creative professionals—while leaving the rest of their workforce on E5.
This selectivity has cost implications for you. If you decide to roll out Copilot to your entire organization via E7, you're likely licensing it for many users who won't actively use it. According to Microsoft's own data, only 35.8% of users who have access to Copilot actively use it. That means approximately 64% of Copilot seats are dormant or significantly underutilized.
At $30 per user per month, or $360 per year, that dormant seat is expensive. If you license Copilot for 10,000 users and only 3,580 actively use it, you're paying $3.6 million per year to cover the non-users. That's a sunk cost that E7 bundling doesn't eliminate; it just disguises it in a higher overall subscription price.
Activation vs. Adoption: The 35.8% Problem
Here's a critical distinction that Microsoft often glosses over: having a Copilot license is not the same as using Copilot. Microsoft's internal telemetry shows a 35.8% workplace activation rate, meaning approximately 35.8% of users with Copilot access actively choose to use it in a given month. The other 64.2% have access but don't use it regularly, if at all.
This is a problem because it reveals something important about user perception and confidence in the tool. If Copilot were obviously useful—if every knowledge worker immediately saw value in using it for their daily tasks—activation rates would be higher. Instead, most users are either skeptical, satisfied with their current processes, or unable to integrate Copilot into their actual workflows.
The reasons vary. Some users don't trust Copilot's accuracy (more on this below). Some find that their specific workflows don't map well to Copilot's capabilities. Some lack clear use cases and assume Copilot is a nice-to-have rather than an essential tool. Some work in roles where Copilot adds minimal value—frontline workers, logistics coordinators, field service technicians. All of these reasons are valid, and together they explain why two-thirds of users with Copilot access don't use it regularly.
For your organization, this means you need to evaluate your specific workforce and use cases before deciding to license Copilot broadly. If your organization is primarily creative, analytical, or strategy-oriented, activation rates may be higher. If your workforce includes many users in roles where Copilot doesn't naturally fit, you'll end up paying for seats that deliver minimal ROI.
Why Users Don't Choose Copilot (When Given a Choice)
When users have a choice of AI tools, the data is stark. A recent enterprise survey from Q1 2026 found that when offered Copilot, ChatGPT, and Gemini as alternatives, here's how users allocate their usage:
- ChatGPT: 76% of users choose it as their primary tool
- Copilot: 18% of users choose it as their primary tool
- Gemini: 6% of users choose it as their primary tool
That's a 4:1 preference for ChatGPT over Copilot. And Copilot still wins over Gemini, primarily because Microsoft's distribution advantages (bundling into Microsoft 365, integration with Office apps) make Copilot more discoverable and easier to use within existing workflows.
When the survey expanded to include users who have access to all three tools through their organization (typically a large tech company with multiple AI platform investments), the preference divergence became even more stark. In that scenario, where discovery and friction are equal, users chose ChatGPT 65% of the time, Gemini 20% of the time, and Copilot only 8% of the time.
Why does this matter? Because it shows that Copilot is not the default choice. Users don't choose it for its superiority in accuracy, reasoning, or user experience. They choose it because it's convenient, integrated, and already included in their Microsoft 365 subscription. Once friction is introduced—having to switch tools, losing context, accessing a different interface—users gravitate to alternatives they perceive as better.
For you, this means that licensing Copilot exclusively through M365 creates a false sense of adoption. Users will use it when it's convenient and integrated. But they're not voting for Copilot as the best tool available. They're voting for convenience.
The Accuracy Problem (And Why It Matters)
One of the most important signals in the Copilot narrative is the Net Promoter Score (NPS), which measures user satisfaction and likelihood to recommend a product. Microsoft doesn't publicly report Copilot NPS figures, but third-party research firms tracking enterprise AI adoption have found the following:
- Copilot NPS (July 2025): -3.5 (slightly negative)
- Copilot NPS (September 2025): -24.1 (significantly negative)
- Copilot NPS (January 2026): -19.8 (still negative)
For context, NPS ranges from -100 (universally hated) to +100 (universally loved). A positive NPS of +50 or higher is considered excellent. An NPS near zero is neutral. A negative NPS—anything below zero—indicates that more people would not recommend the product than would recommend it. Copilot's trajectory from slightly negative to significantly negative to still-negative suggests growing concerns among users.
The primary complaint is accuracy. Copilot makes mistakes. Sometimes it hallucinates facts, providing confident-sounding answers to questions where it doesn't have reliable information. Sometimes it misses context or nuance in the documents you ask it to analyze. Sometimes it generates code that doesn't compile or writes email that miscommunicates the intended tone.
Microsoft would argue that all large language models have this problem, and that's true. ChatGPT and Gemini also make mistakes. But the difference is expectation. Users expect Microsoft, with its engineering resources and access to enterprise data, to have built a version of Copilot that's more accurate than the free alternatives, not equivalent to them. When Copilot performs similarly to free ChatGPT, the perceived value diminishes. Users ask: why pay $360 per year for access to a tool that's no better than the free option?
This accuracy concern is not trivial. It's a primary factor in why adoption and activation rates are low. Users don't trust the output enough to incorporate Copilot into critical workflows. They use it for ideation, brainstorming, and low-stakes content generation, but they don't rely on it for analysis that informs business decisions, code that goes directly to production, or communication that represents their professional judgment. In other words, they use Copilot, but not for anything that matters.
The ROI Calculation: Microsoft's 14-Minute Claim Under Scrutiny
Microsoft's primary ROI argument for Copilot is time savings. The company has claimed that users save approximately 14 minutes per day by using Copilot, which it calculates as $0.50 per minute of saved time (based on typical knowledge worker salaries). That's $7 per user per day in value, or approximately $1,500 per year in time savings per user.
Copilot's cost is $30 per user per month, or $360 per year. At face value, the ROI looks positive: $1,500 in value for $360 in cost, a 4:1 return on investment. This is the calculation Microsoft emphasizes in sales conversations, and it's the foundation for many organizations' Copilot business cases.
But here's where the calculation breaks down:
First, the 14-minute figure is based on survey data from Microsoft's own customers who actively use Copilot. It doesn't account for the 64% of users who have access but don't use it regularly. For those users, the time savings are zero, yet you're still paying the $360 per year per seat. Once you factor in non-users and low-activity users, the average time savings per seat drops dramatically. If only 35.8% of users achieve the 14-minute daily savings, the organizational average is approximately 5 minutes per user per day, or $1,800 per year in aggregate value per thousand users—which is less than the cost of licensing Copilot for those thousand users at $360,000 per year.
Second, the $0.50 per minute valuation assumes that all time savings have equal economic value. In reality, some time savings are more valuable than others. If a senior analyst saves 14 minutes per day doing routine analysis that Copilot helps with, that's valuable. But if a frontline worker saves 5 minutes per day on tasks that have no direct revenue impact, the economic value is lower. Microsoft's calculation doesn't differentiate.
Third, the 14-minute figure doesn't account for time spent correcting Copilot mistakes. Users who don't trust the output spend additional time validating, fact-checking, and revising Copilot-generated content. That's time lost, not saved. For organizations where accuracy concerns are high, the net time savings may be zero or negative.
The realistic ROI for Copilot is likely positive for targeted groups—senior analysts, creative professionals, developers using GitHub Copilot—but neutral to negative for broad organizational rollout. This is why selective licensing (licensing Copilot only for specific roles or departments) is becoming more common than organization-wide E7 adoption.
The E7 Implication: You're Betting on Copilot Whether You're Ready or Not
Here's where adoption data becomes a licensing decision: Microsoft is pricing E7 at $99 per user per month partly on the assumption that organizations will value Copilot at $30 per user per month. But if only 3.3% of Microsoft's customer base has paid for Copilot after two years, and only 35.8% of those who do have access actually use it, then most organizations haven't validated that the $30 is worth paying.
By bundling Copilot into E7, Microsoft is essentially forcing organizations to make a bet: upgrade to E7, accept the higher cost, and hope that Copilot delivers the 14 minutes per day in time savings that the company claims. If you're wrong—if your organization's activation rate is below 35.8%, or if your users don't actually save 14 minutes per day—you've just increased your per-user licensing cost by $9 per month ($99 E7 vs. $90 for E5 + Copilot separately) without seeing the expected return.
The risk is particularly acute if you're planning to license E7 for all users. Your finance team probably doesn't need Copilot. Your operations team probably doesn't need it. Your HR team probably doesn't need it. Your frontline workers definitely don't need it. Yet you're paying for it across the board, betting that the knowledge workers and analytical staff who do use Copilot will drive enough value to justify the entire rollout.
A more prudent approach is selective licensing: keep most of your organization on E5, and license Copilot separately for specific roles where you believe the ROI is positive. This requires more granular purchasing, but it also requires you to validate your assumptions about usage and value before committing budget.
Why Customers Aren't Buying Copilot (The Real Reasons)
Understanding why adoption is low is essential to making a Copilot decision. The reasons are specific and quantifiable:
Accuracy concerns: Users don't trust Copilot's output, particularly for analytical or business-critical tasks. The negative NPS reflects this. Until Microsoft addresses the hallucination problem more comprehensively, trust will remain low.
Integration friction: Copilot works well when it's already available in the tool you're using (Teams, Word, Outlook). But switching contexts—opening a separate chat window, using a different interface—creates friction. Users revert to their existing workflows.
Change management costs: Deploying Copilot requires training, policy definition, and ongoing support. These costs are often underestimated in ROI calculations. A 5,000-user Copilot rollout requires time investment from your IT and training teams, and that's a real cost even if Microsoft's license cost is bundled into E7.
Unclear use cases: Many organizations haven't identified specific, high-value use cases for Copilot. Without a clear problem it's solving, users see it as a nice-to-have toy, not an essential productivity tool.
Competitive alternatives: ChatGPT, Gemini, and other tools are free or cheap, often better, and don't require organizational rollout. Individual users can access them without IT involvement. As a result, many organizations see Copilot as duplicative rather than essential.
Budget constraints: For many organizations, the choice is binary: invest $30 per month per user in Copilot, or invest that budget in other priorities like security tools, collaboration infrastructure, or people. Copilot is losing that comparison in many cases.
What This Means for Your E7 Decision
If you're evaluating whether to upgrade to E7 for Copilot, here's what the data suggests:
Don't upgrade for organization-wide Copilot deployment. The data shows that most organizations won't achieve the activation rates or time savings Microsoft claims. You'll end up licensing Copilot for many users who don't use it, paying $360 per year per unused seat. At 5,000 users, that could be $1.2 million in wasted annual spend.
Do upgrade if you've already validated Copilot ROI in a pilot. If you've run a 3-month pilot with a representative group of 200 users, measured actual time savings and activation rates, and found compelling business value, then E7 makes sense for those roles. But extend it only to the groups where you've validated the case.
Consider selective licensing instead. Buy E5 for your broad base, and license Copilot separately (at $30 per month or bundled via E7 for specific departments) for roles where you believe the value is clear: senior analysts, creative teams, developers, business strategists. This requires more granular purchasing, but it's more financially efficient.
Invest in use case development before rolling out broadly. Don't assume that Copilot will find its own value. Work with your business units to identify specific workflows where Copilot can add value: email drafting, content summarization, code generation, research synthesis, meeting note generation. Once you have 5-10 validated use cases and you've trained users on them, activation rates will improve.
Measure before and after. Before you commit to E7 for Copilot, establish a baseline: how long does it take your knowledge workers to complete current tasks? How much time do they spend on drafting, summarizing, researching? Then pilot Copilot and measure actual time savings. If you don't see 10+ minutes per day in time savings for your pilot group, don't extend the rollout.
How to Structure a Proper Copilot Pilot
If you're serious about Copilot but want to de-risk the decision, here's a recommended pilot structure:
Phase 1: Small Group Pilot (4-8 weeks, 50-200 users)
Select a pilot group from a department where you believe Copilot will have high impact: strategy, business analysis, marketing, or research. License Copilot for this group. Provide training on specific use cases (email drafting, meeting summaries, document analysis). Track adoption metrics daily. Measure time spent on targeted tasks before and after Copilot rollout. Run a survey at weeks 2, 4, 6, and 8 to gather feedback on accuracy, usefulness, and likelihood to recommend.
Success criteria: 60%+ activation rate, 10+ minutes per day in time savings, NPS above +20.
Phase 2: Department-Level Expansion (8-12 weeks, 500-2,000 users)
If Phase 1 is successful, expand Copilot to the entire department where the pilot was conducted. Refine training based on pilot feedback. Continue tracking activation and time savings. Identify power users and have them mentor peers. Run a second survey at the end of the expansion phase.
Success criteria: 45%+ activation rate (expect lower than Phase 1 due to larger group heterogeneity), 8+ minutes per day in time savings, NPS above +10.
Phase 3: Cross-Functional Rollout (12-24 weeks)
Based on Phase 2 results, expand Copilot to additional departments with similar profiles. Continue tracking adoption and value. Build an internal business case based on real data. Update training materials based on what actually worked in Phase 1 and Phase 2.
At the end of Phase 3, you'll have real data on which departments benefit from Copilot and which don't. You can then make an informed decision about E7 adoption: license it for the high-value departments, maintain E5 for others.
Negotiating Position: April-June Leverage
If you're approaching your next EA renewal or license negotiation, you have leverage right now (April-June 2026). Microsoft wants E7 adoption, and the company has aggressive quota targets for Copilot seat growth. If you're not yet convinced of Copilot's value, use that to negotiate:
- Ask for a 3-month free trial of Copilot for a pilot group. Microsoft would rather pilot with you than lose your deal entirely.
- Ask for a holdback: agree to upgrade part of your organization to E7, but negotiate a price reduction if you don't achieve 40%+ activation in the first 6 months.
- Ask for flexible licensing: license E7 for specific departments, E5 for others, without requiring organization-wide upgrades. This is more operational complexity for Microsoft, but they'll accept it to close a deal.
- Ask for training and change management support. Most organizations underestimate the effort required to make Copilot work. If Microsoft will fund training or provide implementation support, that increases the likelihood of success.
Conclusion: Adoption Data Should Drive Your E7 Decision
The headline number—15 million paid Copilot seats—sounds impressive. But the context is crucial. That's only 3.3% of Microsoft's addressable market after two years on the market. Only 35.8% of users with Copilot access actively use it. User preference for ChatGPT over Copilot is 4:1 when alternatives are available. Accuracy concerns remain significant, reflected in negative NPS scores.
This doesn't mean Copilot is a bad product. It means Copilot is a selective product—valuable for specific roles and specific use cases, but not universally valuable across all organization types. The risk of E7 adoption is that you license Copilot for everyone, betting on broad value creation, and end up paying for seats that don't deliver ROI.
The smarter approach is to pilot Copilot rigorously, validate specific use cases and economic value, measure actual adoption and time savings, and then decide whether organization-wide E7 adoption makes sense for your organization. For many organizations, the answer will be selective licensing: E7 for the groups where you've validated value, E5 for everyone else.
In one engagement, a global technology firm with 14,500 Microsoft 365 users was evaluating an E7 upgrade at Microsoft's urging. Their account team projected $4.2M in Copilot-driven productivity gains. Redress conducted an independent TCO analysis and a structured pilot across three business units. Actual active usage after 90 days was 28% — well below Microsoft's modelled adoption rate. Redress recommended selective E7 deployment for 2,100 high-value users and E5 retention for the remaining 12,400, saving $3.1M annually versus the full E7 rollout. The engagement fee was less than 5% of the identified saving.