Corporate AI research is moving fast right now. New surveys land almost every week, each with its own headline numbers, and it can be hard to know which ones deserve your attention.

We gathered recent reports and compared what they found. Across very different methods and samples, the reports keep pointing toward the same conclusion: access to capable AI tools is only one part of the challenge. The harder work increasingly lies inside the organization—in governance, skills, workflows, accountability, and people’s readiness to use AI effectively.

Boston Consulting Group’s fourth annual AI at Work survey asked nearly 12,000 employees across 14 markets what AI is actually doing to their jobs, and the answer is that the tools are producing savings faster than organizations are redesigning work around them. Among frontline employees who regularly use AI, 42% reported saving at least eight hours a week. Yet 66% said they receive limited or no guidance on how to use that time, and more than half said they are not redirecting it toward more strategic work. BCG also found that strong strategic clarity was associated with a 25-percentage-point difference in reported measurable impact, compared with only five points for strong access to tools without clear direction.

The Society for Human Resource Management asked 5,875 US workers about their day-to-day experience, and the answer was less about software than about culture. Nearly 40% of workers said their workplaces had run practical workshops on day-to-day AI skills. However, the finding SHRM chose to lead with is that culture matters as much as adoption. Where workers trust how the organization is approaching AI and believe human judgment still matters, engagement and organizational commitment tend to be stronger. Where trust and communication are weaker, organizations risk lower confidence and less purposeful adoption.

Smarsh and FTI Consulting found the same gap from the compliance side. While 55% of enterprises are actively deploying AI, only 26% said their governance frameworks are fully aligned with the pace of implementation. The risk becomes especially clear around shadow AI: just 30% of organizations said they have comprehensive systems for detecting and managing employees’ use of unauthorized AI tools. That can leave companies with limited visibility into where sensitive data is going, how AI-generated content is being used, and whether required records are being preserved. If you read our earlier piece on shadow AI, this is the same story now showing up in compliance data rather than anecdotes.

KPMG surveyed more than 2,000 senior leaders for its quarterly Global AI Pulse, and the numbers show that adoption is accelerating faster than measurable returns. The share of organizations embedding AI across their operations rose from 13% to 22% in a single quarter, the largest shift anywhere on KPMG’s maturity scale. Yet only 7% of leaders said their organizations had established a measurable return on their AI investment, even as nearly a quarter faced growing pressure from investors to demonstrate value. One factor was strongly associated with better results: organizations with full visibility into their AI operating costs were five times more likely to report established returns than those without it. The lesson is not simply to deploy more AI, but to track what it costs and whether it is producing measurable value.

Read together, these reports are pointing a flashlight at the same spot. Not the model, but the people and the systems around it. The gap between a company that has AI and a company that gets value from AI is not primarily a technology gap. It is a question of whether the work around the tool has been done.

For anyone setting priorities for the fall, that becomes a simple gut check. Look at where your AI budget and attention are actually going. If most of it is aimed at tools and very little at the people and workflows using them, these recent reports suggest the balance may be backward.

Closing that gap is the work we do at UnconstrainED. If your fall planning includes rethinking how your people work with AI, we would love to be part of that conversation.