Emily Lewis-Pinnell, Driving AI Adoption at Evaila.
We’ve seen the AI adoption numbers. Organizational adoption has reached 88%. MIT’s NANDA lab found that 95% of generative AI pilots deliver zero measurable P&L impact. McKinsey and BCG tell versions of the same story from different angles: a small minority of organizations are capturing real value from AI, while most are not.
These numbers have been circulating for months. What has been missing is a clear picture of what the organizations succeeding with AI adoption are actually doing differently.
A new study from Stanford’s Digital Economy Lab offers the most useful answer I have seen, and it documents what I have observed repeatedly working with clients on AI adoption. The Enterprise AI Playbook, published in April 2026 by Pereira, Graylin and Brynjolfsson, examined 51 deployments that delivered measurable value across 41 organizations, seven countries and over a million employees. Rather than surveying sentiment, they studied what worked and reverse-engineered the patterns.
Technology Not The Hard Part
When Stanford’s researchers asked practitioners across those 51 successful deployments to identify the hardest challenge, 77% pointed to invisible costs: change management, data quality and process redesign. Not the model. Not the infrastructure. Even amongst the organizations that got it right, the hardest work was organizational, not technical.
For 42% of implementations, the foundation model itself was fully interchangeable. The durable advantage was in the workflow design, not the AI.
And 61% of those successful deployments included at least one prior failure. The earlier failures shared a pattern: teams treated AI as a technology project rather than one of process redesign. They applied AI to broken workflows and expected the model to compensate.
How This Looks In Practice
One sales organization I worked with had invested in enabling new Salesforce AI features and rolled out an enterprise ChatGPT license. On paper, the team had access to everything they needed. In practice, usage was thin and uneven. Reps were not sure what to do with the tools, managers were unsure of what to measure and the workflow remained unchanged. The AI sat alongside the existing process rather than inside it.
A program that started with the team, not the tool, shifted the outcome. Training that framed AI around what reps actually wanted, which was less time on admin and more time talking with customers. Then, critically, modeling the specific new workflows that would make this possible: AI-drafted follow-ups that reps reviewed instead of writing from scratch, automated call summaries that fed directly into opportunity updates, prep briefs generated before customer meetings rather than assembled manually. Each workflow was designed, tested and tied to a metric the rep already cared about.
The tools were the same before and after. The workflow was not.
Workflow Over Technology
McKinsey tested 25 organizational factors against EBIT impact from AI. The single highest-correlated factor was workflow redesign. Their research found that 55% of high performers had fundamentally redesigned workflows around AI, versus only 20% of other companies.
Stanford’s data now explains why this gap is so consequential. Similar use cases, the study found, took weeks at one organization and years at another. The difference was never the model. It was executive sponsorship, existing organizational processes and end-user willingness. The workflow, not the technology, determined the timeline and the outcome.
Where To Start
In my own practice, I use a five-phase approach I call the CHART Framework to move organizations from experimentation to operational impact. The sequence matters more than any individual step.
1. Clarify Before You Build
Define goals, map stakeholders and establish the real AI baseline. Most organizations skip this and start shopping for tools before they understand where they actually stand.
2. Highlight Where AI Will Matter Most
Not every process needs AI. Surface the highest-value opportunities across functions and workflows and be honest about where it will not help.
3. Architect A Phased Plan Across People, Process, Technology
Sequence for operational reality, not a lab environment. Stanford’s data confirms this: the organizations that succeeded designed for how work actually moves, not how it looks on a diagram.
4. Ready Your People
This is the phase most organizations skip entirely, and it is where adoption lives or dies. Practical training, fluency and change management are not overhead. They are the work.
5. Track Outcomes, Scale What Works
Tie results back to baseline metrics. Expand from there, not before.
The Model Is A Decision—The Workflow Is The Discipline
The conversation at most organizations is still organized around model selection, vendor evaluation and pilot count. Stanford’s data suggests that conversation is oriented around the wrong variable. For nearly half of the use cases studied, the model was a commodity. What is not a commodity is the work of redesigning how teams operate, where decisions are made and how accountability flows when AI enters the sequence.
The 88% adoption figure was supposed to be the finish line. It turned out to be the start. What separates the leaders from the rest is not better AI, but the discipline of redesigning the work around it.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
