What It Takes to Scale AI: Lessons from the AI Adoption Maturity Model

Most organizations are not short on AI ambition. Budgets are increasing, pilots are running, and tools are being rolled out across teams. In fact, 86% of C-suite leaders plan to increase AI spending in 2026. 

Yet the results don’t match the investment. According to Accenture research, only 21% of organizations are redesigning end-to-end processes with AI at the core, and nearly half of executives say AI has so far delivered little impact on profit. 

The technology isn't the problem. Execution is. 

That gap is what the AI Adoption Maturity Model (developed jointly by Accenture and the Carnegie Mellon University Software Engineering Institute (SEI)) was built to address. Launched in June 2026, the model gives organizations a structured, research-backed way to understand where they stand on AI adoption, identify what's missing, and build a realistic roadmap forward. 

Professor Majd Sakr, Chief Learning and Research Officer at Accenture and a faculty member at Carnegie Mellon, helped bring this work to life. He spends significant time with C-suites and boards on AI strategy and workforce transformation. In those conversations, the same questions keep surfacing:

Are we doing this right?
Where should our big bets go?
What transformations matter?

The model was designed to give leaders a credible, research-grounded way to answer those questions, rather than relying on opinion or gut instinct. Much of what follows comes from the launch webcast, "Rethinking and Maturing AI Adoption," where Majd joined colleagues from Accenture and the SEI to walk through the model and the thinking behind it. The full discussion is worth watching, and the takeaways below capture the parts most relevant to how we think about workforce capability and learning.

Why Most AI Maturity Models Fall Short

Before building this one, the SEI and Accenture reviewed more than 100 existing AI maturity efforts. (A forthcoming SEI study puts the full count at 136 models and frameworks examined.) They found a common problem: most didn't have a technical background, a clear measurement method, or any proof that it worked in real life. Many were narrowly focused. Few could show how they were developed or whether they worked in practice. 

This model took a different approach. It was grounded in roughly 25 executive interviews, surveys of nearly 600 practitioners, and intensive pilots with Fortune 500 organizations, including Accenture's own Global IT organization as the first test case, internally referred to as "pilot zero." The findings were fed back into the model before it was released publicly. It also draws on four decades of SEI leadership in maturity modeling and Accenture's experience across more than 11,000 advanced AI projects worldwide. 

One of the most consistent patterns that emerged from that research: organizations can have strong technical capabilities and still be nowhere near ready to scale AI. The bottleneck was rarely the technology; it was how AI entered the workforce.

The Eight Dimensions of AI Maturity

The model assesses readiness across eight core areas, divided into two categories: organizational change and AI lifecycle engineering.

Organizational Change:

1. Organizational strategy: whether AI efforts are tied to real business goals, not just technology deployment for its own sake. The foundational question is why the organization needs AI to achieve its outcomes, not what AI tools it can deploy. 
2. Workforce and culture: whether people have the capability, confidence, and ongoing training to use AI in their actual work. This is consistently identified as the most underdeveloped dimension across organizations, and the hardest to get right.
3. Workflow re-engineering: whether processes are being redesigned around AI from first principles. Augmenting existing workflows with AI is a starting point; genuinely rethinking how work gets done is the goal. 
4. Risk and governance: whether accountability for AI decisions, data ownership, and failure risk is clearly defined. As agentic AI increases in complexity, governance becomes less optional and more foundational. 

AI lifecycle engineering:

5. Data: whether the underlying data is usable by AI systems. This came up repeatedly in the panel discussion. The assumption that a large language model can work around messy or incomplete data is, as one panelist put it, "absolutely not true." Clean, contextually useful data is the foundation everything else depends on. 
6. Engineering: the technical rigor behind how AI systems are built, integrated, tested, and maintained. 
7. Operations: how AI solutions are monitored, sustained, and improved once they're in production, not just at deployment. 
8. Ecosystem: how well the organization manages dependencies on vendors, platforms, and external partners without creating fragility or lock-in. 

The Misconception That Slows Everything Down

One of the most helpful parts of the panel discussion was talking about what people think about AI adoption, and what organizations think about it that is wrong. Several came up, but one kept returning in different forms: the belief that deploying tools and training people on those tools is the same as building capability. 

Majd addressed this directly. There is a progression that organizations often underestimate. AI literacy is a starting point. Hands-on training comes next. Then role-based learning that connects skills to specific job functions. Then domain-specific application that reflects the actual industry and workflows involved. Only then do people start using AI in how they work every day, not as an extra, but as a real part of how they do their jobs. 

Running a lot of pilots doesn't mean an organization is AI-ready. Having many people using a particular tool doesn't mean AI has been adopted. As Majd noted, having many active AI projects is a sign of experimentation, not necessarily maturity. Organizations that conflate the two often stop investing before they've built anything that scales. 

The Dimension That's Most Often Missing (Workforce and Culture)

When panelists were asked which section of the model was both the most important and the most widely underdeveloped, the answer was immediate: workforce and culture. 

The observation from the panel was unanimous. Every organization will eventually have access to the same AI tools. Technology is not a durable differentiator. What differentiates organizations is what their people can do with it—their judgment, their ideas, their ability to apply AI to real problems. That's not a training problem in the traditional sense. It's a cultural and organizational design problem. It requires:

• Leaders who understand AI well enough to guide their teams through it
• Role structures that account for how work is changing 
• Learning that is embedded in the work itself rather than treated as a separate event. 

The panel also discussed something that doesn't get a lot of air time: cognitive ownership. As organizations push AI adoption, there is a risk that employees will begin deferring to AI outputs without using their own judgment. The goal is not to have people accept what AI produces. It's to have people work with AI while still being able to question it, disagree with it, and use their own knowledge to decide what's right.

Maturity Is Not a Destination

One of the more clarifying ideas from the session was about how to think about maturity levels in the first place. 

The model defines five: Exploratory AI, Implemented AI, Aligned AI, Scaled AI, and Future Ready AI. 

But the point isn't to climb to the top. The model is not a certification, but instead understanding where you are, identifying gaps, and knowing what to do next. AI technology is changing fast. Today's AI is almost different from it was a year ago. The model is designed to change as practices and realities change. Organizations that treat training as a point-in-time snapshot will miss out on continuous reassessment and adjustment. 

As Majd put it in his closing remarks: "This is never a one and done. The organization needs to be in the mindset that we will continuously update ourselves and continuously disrupt ourselves so that we can continue to lead.”

The AI Adoption Maturity Model is publicly available through Carnegie Mellon SEI. You can download it and read the full supporting research at sei.cmu.edu. 

Accenture LearnVantage works with organizations to build workforce and culture capabilities by aligning upskilling with real roles and workflows. Ascendient Learning, part of Accenture LearnVantage delivers hands-on, live training for teams and organizations.

Quick Stats at a Glance from Accenture's research and SEI's Methodolgy:

These are taken from Accenture's press release and the SEI project page.

• 86% of C-suite leaders plan to increase AI spending in 2026 
• Only 21% of organizations are redesigning end-to-end processes with AI at the core 
• Nearly half of executives report AI has so far delivered little impact on profit 
• The model is grounded in 4 decades of SEI maturity modeling and Accenture's experience across more than 11,000 advanced AI projects worldwide
• More than 100 existing AI maturity efforts reviewed 
• Approximately 25 executive interviews Nearly 600 practitioners surveyed 
• Intensive pilots with Fortune 500 organizations, with Accenture Global IT as "pilot zero"