Over the past several months, I’ve had the same conversation with multiple large organizations. They adopted Microsoft Copilot with genuine enthusiasm, assigned licenses, ran internal communications, and waited for productivity gains to show up. For most of them, the gains never materialized in a way they could measure or defend to leadership.
What’s happening now is more significant than a slow rollout. Organizations are actively pulling back. Some are reducing their Copilot seat counts at renewal. Others are redirecting their AI investment toward OpenAI or Anthropic. A few have done both.
The reasons are consistent across industries and company sizes.
The first is that Copilot’s value is hard to quantify in a way that satisfies a CFO. Productivity improvements from AI tools tend to be diffuse and personal. They show up in individual workflows, not in quarterly metrics. When a finance team asks for the business case to renew 5,000 Copilot licenses at $30 per user per month, the answer is often a collection of anecdotes rather than data. That’s a difficult position to defend when budgets are under pressure.
The second issue is that Copilot works best inside the Microsoft ecosystem, and many organizations have hybrid environments. If your data lives in Salesforce, ServiceNow, SAP, or a mix of internal systems, Copilot’s usefulness narrows considerably. The promise of an AI assistant that understands your business only holds if your business runs entirely on Microsoft infrastructure, which most large enterprises do not.
The third issue is token visibility. Organizations that have moved into more advanced AI usage, particularly in development and automation, are discovering that consumption-based AI costs can escalate quickly and without warning. One organization I worked with absorbed over $500,000 in unexpected AI spend during a development cycle because there was no adequate monitoring in place. Microsoft’s tooling for tracking and governing AI consumption is still catching up to the reality of how enterprises actually use these platforms.
What’s drawing organizations toward OpenAI and Anthropic is a combination of model performance, flexibility, and the ability to integrate across a broader technology stack. These platforms don’t require you to standardize on a single vendor’s ecosystem to get value. They also tend to perform better on complex reasoning tasks, which is where enterprise users are increasingly pushing AI tools.
None of this means Copilot is without value. For organizations that are deeply embedded in Microsoft 365 and have strong adoption of Teams, SharePoint, and Outlook, there are genuine use cases where Copilot delivers. The problem is that Microsoft’s pricing and packaging assumes broad enterprise adoption, and the reality on the ground is much more selective.
If you’re heading into a Microsoft renewal in the next six months, the most important thing you can do is get honest about actual Copilot usage before that conversation starts. Usage data, not license counts, should drive the negotiation. And if your organization is evaluating AI investment more broadly, it’s worth separating the Microsoft productivity layer from the AI infrastructure question entirely. They are not the same decision.
The organizations that are getting this right are treating AI investment with the same discipline they apply to any major technology spend. They’re measuring, adjusting, and making decisions based on what’s actually working rather than what was promised in a vendor presentation.
