The AI Adoption Maturity Model: Why Half of Executives See No Profit From AI

The AI Adoption Maturity Model is a framework for assessing whether an organization can actually sustain the practices that turn AI spending into business value, released in June 2026 by Carnegie Mellon University's Software Engineering Institute (SEI) and Accenture. It measures eight dimensions of readiness, assigns five levels of maturity, and was built from reviews of more than 100 existing frameworks, surveys of roughly 600 practitioners, and pilots with Fortune 500 companies. The problem it targets is blunt: nearly half of executives report that AI has delivered little impact on profit.

That gap (record adoption, weak returns) is the defining tension of business AI right now. This post explains what the model measures, why the SEI's involvement matters, and how an organization without a Fortune 500 budget can use the same logic this month.

What Is the AI Adoption Maturity Model?

The model assesses an organization's ability to perform and sustain specific technical and organizational practices in pursuit of two high-level goals: organizational change and AI lifecycle engineering. That pairing is the thesis. Getting value from AI is half a people-and-process problem and half an engineering discipline problem, and an organization weak in either half will not convert spending into results.

The pedigree matters here. The SEI is the federally funded research center that created the original Capability Maturity Model (CMM) in the early 1990s, the framework that gave software development its first widely adopted definition of process maturity. Applying that lineage to AI adoption is a statement: the industry has enough experience now to say what good looks like, in writing, with levels. Version 1.0 is a free download from the SEI Digital Library, with Accenture offering commercial assessments built on top of it.

What are the eight dimensions?

• Organizational strategy. Is AI investment tied to named business problems with owners and success metrics, or is it a budget line labeled "innovation"?
• Workforce and culture. Do the people whose work changes have the skills, incentives, and trust to use the systems? Adoption fails in the last metre more often than in the model.
• Workflow re-engineering. Has the process been redesigned around the new capability, or was AI bolted onto a workflow shaped by pre-AI constraints?
• Risk and governance. Who is accountable for AI behavior, what gets reviewed before deployment, and what happens when output is wrong?
• Data. Whether the organization can find, clean, govern, and feed the data its use cases need.
• Engineering. The ability to build, buy, integrate, and test AI-enabled systems rather than accumulate disconnected pilots.
• Operations and sustainment. Keeping systems working after launch: monitoring, maintenance, retraining, cost control.
• Ecosystem. The vendors, partners, and platforms the organization depends on, and how deliberately those dependencies are managed.

Notice what is not on the list: there is no dimension for picking the right model or tool. In the SEI's framing, tool selection is downstream of capability. That inverts how most organizations actually budget, where the tool purchase comes first and the readiness work is assumed to follow. It rarely does.

What are the five maturity levels?

Maturity in each dimension lands on one of five levels, based on how completely the dimension's practices are implemented, governed, and sustained:

The scoring is per dimension, not one blended number, and that design choice is what makes the output useful. An organization can be level 3 on engineering and level 1 on workforce and culture, and that lopsided profile is the diagnosis: the bottleneck is adoption, not capability. Most honest assessments produce exactly this kind of jagged profile.

Why Do Half of Executives See No Profit From AI?

Read against the eight dimensions, the familiar failure stories sort themselves into three patterns.

Pilot purgatory. A proof of concept works in a demo, impresses a steering committee, and never integrates into the workflow where the value lives. In maturity terms, the organization reached Implemented AI in engineering while workflow re-engineering stayed Exploratory. The pilot was real; the adoption was not.

Unmeasured ROI. Nobody baselined the process before the AI arrived, so nobody can demonstrate what changed. This is a strategy-dimension failure: no named metric, no owner, no before-and-after. We made the same argument about Uber's AI spending reckoning: the only number that matters is cost per delivered outcome, and you cannot compute it if you never measured the "before."

Tool-first adoption. The organization bought licenses because competitors were buying licenses, then went looking for problems. Usage statistics look healthy, impact statistics do not exist. This is the strategy and workforce dimensions failing at the same time, and it is the most common pattern among small and mid-sized companies, who feel the pressure to "do AI" without a framework for sequencing it.

How Should a Smaller Organization Use This?

A formal SEI-style appraisal is enterprise machinery. The logic, though, scales down to any size of company, because the eight dimensions are really eight questions, and answering them honestly takes a meeting, not an engagement:

• Strategy: Can we name the one business problem AI should solve first, and what its success metric is?
• Workforce: Will the people in that workflow actually use it, and do they have a reason to want it to succeed?
• Workflow: Are we redesigning the process, or decorating it?
• Governance: Who is accountable when the output is wrong, and what gets human review?
• Data: Is the data this use case needs accessible and clean enough to use this quarter?
• Engineering: Can we integrate this into the systems where work happens, or will it live in a separate tab?
• Operations: Who maintains it in month six, and what does it cost to run?
• Ecosystem: Which vendors are we betting on, and how locked in are we?

Two or three honest "no" answers are not a reason to stall; they are the work plan. Fix the cheapest blocking gap first (usually strategy: pick one problem and baseline it), run one workflow to a measured result, and let that result fund the next dimension. That sequencing is essentially what we walk through in our consultant-versus-in-house framework, and it is the structure behind our free AI readiness assessment.

The release of a CMU-backed maturity model is also a signal about where the market is heading. Frameworks like this become procurement language. Within a year or two, expect "what is your AI maturity level?" to show up in enterprise RFPs and board decks the way CMM levels once did in software contracting. Organizations that can answer with a profile and a plan will have an easier conversation than those answering with a tool list. If you want help building that answer, our AI consulting practice starts exactly there, or book a call and we will map your eight dimensions together.

Frequently Asked Questions