You bought the licenses. Most of your team is not using them.
You already have Microsoft 365. Your team uses Outlook, Teams, SharePoint, OneDrive. When Copilot launched, someone on your IT team said this was the obvious next step. So you bought the licenses.
Six months later, most of your team is not using it. The few who are use it for basic email drafts. The CFO wants to know what the $25,000 annual spend is producing. You do not have an answer.
This is not a hypothetical. It is the most common pattern I see when a mid-market company deploys Microsoft 365 Copilot without fixing what sits underneath it.
This guide covers what Copilot actually does in 2026, what it costs for real companies at real scale, why 65% of licensed users are inactive, what it does not replace, and what to do before deploying it or after buying licenses that nobody uses.
What Copilot actually does per app (2026)
Copilot is embedded inside the Microsoft 365 apps your team already uses. The capabilities are real, but the marketing overstates how much of this works out of the box for a mid-market company with imperfect data hygiene.
Where it delivers the most value, fastest:
Teams: meeting recaps with transcription, action item extraction, in-meeting Q&A. This is the highest-adoption feature across every deployment study. If your team runs 10+ internal meetings per week, this alone can justify the cost for a subset of users.
Outlook: email thread summarization, drafting from bullet points, tone adjustment. For anyone processing 50+ emails per day, this is a meaningful time saver.
Word: first-draft generation from prompts or meeting content, document summarization, tone and structure rewriting.
Where it requires more setup to deliver value:
Excel: Python-assisted analysis, formula generation, data visualization. Powerful, but requires clean data and users comfortable asking the right questions.
SharePoint: custom agents grounded in specific document libraries. Requires deliberate information architecture and permission hygiene to work correctly.
Copilot Studio: no-code agent builder for custom workflows. Supports multi-agent orchestration and fine-tuning on company data. This is where mid-market companies can build real operational tools, but it requires dedicated effort to configure, not a default feature you get for free.
The prerequisite nobody talks about: Admin policies must be configured correctly. Transcription must be enabled for Teams recaps. SharePoint permissions must be audited before Copilot surfaces content. Graph Connectors are required for any non-Microsoft data. None of this happens automatically.
What it actually costs for 50, 100, and 200 users
Microsoft does not make this simple. Copilot is an add-on to your existing Microsoft 365 license. The price depends on which base license your company runs.
| Company size | Base license | Copilot tier | Monthly cost | Annual cost |
|---|---|---|---|---|
| 50 users | Business Standard ($12.50/user) | Copilot Business ($21/user) | $1,050/month | $12,600/year |
| 100 users | Business Standard ($12.50/user) | Copilot Business ($21/user) | $2,100/month | $25,200/year |
| 200 users | E3 ($33.75/user) | Copilot Enterprise ($30/user) | $6,000/month | $72,000/year |
These are Copilot costs only. Add the base Microsoft 365 license on top. A 200-user E3 deployment with Copilot Enterprise runs over $153,000/year in total Microsoft spend, increasing to approximately $166,000 after the July 2026 E3 price increase.
Annual commitment is required. There is no monthly billing option. If you buy licenses and your team does not adopt, you are paying until the contract renews.
No minimum seat requirement since January 2024. You can buy a single license and test it with one person before rolling out company-wide. Most companies should do exactly this.
The 5 failure patterns (why 65% of licensed users do not use it)
Microsoft reports 16.1 million paid Copilot seats. Independent research shows only 35.8% of those are active users. That means roughly 10.3 million seats are partially or fully unused.
A Gartner survey of 187 IT leaders found that only 5% of companies that completed Copilot pilots moved to larger deployment. An MIT study cited across the Microsoft partner ecosystem found that 95% of generative AI pilots fail to deliver measurable business impact.
Pattern 1: Buying licenses before fixing the data layer. The most common failure. Organizations distribute licenses before auditing SharePoint permissions, removing stale data, or establishing metadata standards. Copilot surfaces whatever is there, including outdated proposals, draft documents, and content employees were never supposed to see. Users get confused, lose trust, and stop using it.
Pattern 2: Permission sprawl exposed by Copilot. Years of SharePoint and Teams growth without governance become visible the moment Copilot starts surfacing content across the organization. This is not a Copilot bug. It is Copilot working exactly as designed, on a data foundation that was never designed.
Pattern 3: No governance framework. No policy defining which employee segments get licenses first, what use cases are approved, what requires human review, and how outputs are treated for compliance purposes. High-stakes outputs (financial models, legal drafts, compliance reports) get used without the review they require.
Pattern 4: No training tied to real workflows. Generic "how to use Copilot" sessions do not drive adoption. High-adoption deployments use role-specific training built around actual workflows: the sales team's proposal cycle, the finance team's monthly close, the operations team's reporting cadence.
Pattern 5: No measurement of ROI. Most deployments lack a framework to measure what Copilot is delivering. A Fortune 500 energy company deployed Copilot across thousands of users and had no answer when the CFO asked about tangible benefits six months later. If you are not measuring cycle time, rework reduction, and meeting-to-action conversion from day one, you will not be able to justify the spend at renewal.
What Copilot does not replace
Even with a functioning Copilot deployment, these business problems require custom or external AI solutions.
| Problem | Why Copilot cannot solve it | What solves it |
|---|---|---|
| Non-Microsoft integrations (QuickBooks, Sage, HubSpot, Salesforce, WhatsApp) | Copilot only reads Microsoft Graph data natively. Third-party connectors require engineering. | Custom agents via Copilot Studio + Power Automate, or standalone automation using n8n or Make |
| Industry-specific document processing (contracts, blueprints, compliance filings) | Copilot is a general assistant with no training on domain-specific formats or regulations | Custom-trained models or fine-tuned agents |
| Private knowledge base on proprietary data | Cannot be trained on specialized corpora without significant configuration | Copilot Tuning or custom RAG pipeline |
| AI visibility (appearing in ChatGPT, Perplexity, Google AI responses) | Copilot has no capability to optimize how your company appears in AI search | Dedicated GEO strategy: structured content, entity markup, answer-first pages |
| Cross-platform automation (CRM to email to invoice to task) | Copilot works inside M365 apps only | n8n, Make, or Power Automate flows with external connectors |
| Real-time communication AI on WhatsApp, SMS, or client portals | Copilot has no presence on non-Microsoft channels | Custom agents via WhatsApp Business API or web chat |
The companies I work with typically have Copilot handling what it handles well (meetings, email, document drafts) while custom-built systems handle everything outside the Microsoft boundary. They are not competing tools. They solve different problems. For a detailed breakdown, read what Microsoft Copilot cannot do and what to build instead.
What to do before deploying (or after buying and getting stuck)
Whether you are evaluating Copilot, already bought licenses, or bought licenses and your team stopped using them, the sequence is the same.
Step 1: Audit your SharePoint and data layer (4 to 8 weeks). This is the single highest-value step. Identify overshared sites, stale permissions, broken inheritance, missing metadata. A clean data layer improves Copilot relevance by an estimated 40 to 60 percent. For a deep dive, read your SharePoint is why Copilot shows wrong documents.
Step 2: Design the governance framework. Define who gets licenses first (start with high-value roles, not company-wide), what use cases are approved and what requires human review, how prompts and outputs are treated for compliance, and what the escalation path is when Copilot produces incorrect output. This is a business policy decision, not an IT decision.
Step 3: Build role-specific training. Not a generic Copilot overview. Training tied to actual workflows: how the sales team uses it for proposals, how finance uses it for monthly close, how operations uses it for reporting. Measured by specific outcomes, not attendance. For more on why generic training fails, read we bought Microsoft Copilot and nobody uses it.
Step 4: Set up ROI measurement from day one. Use Viva Insights Copilot Dashboard (no extra license needed) and the Impact Report template. Define what you will track: cycle time per workflow, meeting-to-action conversion, rework reduction. Build a CFO-ready reporting cadence at 30, 60, and 90 days. The full measurement framework is in how to measure ROI from Microsoft Copilot.
Step 5: Identify what Copilot will not solve and build those separately. Map every workflow that touches a non-Microsoft tool. Those are the custom build opportunities. Copilot handles the M365 surface. Custom systems handle everything else. The AI Ops Audit is designed to map exactly this.