Strategy

Most AI Rollouts Fail Before They Start. Here's the Structural Fix.

Melissa Reeve's new book argues that companies aren't failing at AI because of the technology — they're failing because they never changed the organization underneath it.

Written by Elena Brooks · Bureau Business · Jun 18, 2026

The Problem Isn't the Tool

Most companies approaching AI transformation are making the same category error: they're treating it as a technology initiative. Buy the platform, run the pilot, train the team, declare victory. What they get instead is what Melissa Reeve calls "random acts of AI" — isolated wins that don't compound, speed at the edges while the middle stays slow, and adoption curves that plateau.

Reeve, co-founder of the Agile Marketing Alliance and formerly the first VP of marketing at Scaled Agile, lays out the structural argument in her new book *Hyperadaptive: Rewiring the Enterprise to Become AI-Native*. Her core claim is blunt: the operating models most companies run on were built for the industrial era, and they cannot support AI-native work.

The 10-to-90 Gap

The number that anchors Reeve's argument comes from Brad Miller, Moderna's chief information officer during its AI transformation. Miller told Reeve that 90 percent of companies attempting generative AI fail — not because the technology doesn't work, but because they haven't built the organizational mechanisms to absorb it.

Moderna is in the 10 percent. In early 2023, CEO Stéphane Bancel set a target that looked impossible: 15 new drugs to market in five years, against an industry baseline of one drug per decade at roughly $2 billion per development cycle. Moderna hit 100 percent generative AI adoption across the organization within six months. The method wasn't a better model selection process. It was sustained investment in training, coaching, process redesign, and a culture that treated AI fluency as a non-negotiable capability.

Nine Dimensions, Not One

Reeve's framework identifies nine organizational dimensions that must move together for AI transformation to hold. Three stand out as the most consistently neglected:

**Incentives.** If reward systems still pay people for being right rather than for learning fast, the organization will not adapt. AI work involves unknowns and iteration. People need to feel safe failing.

**Decision rights.** AI compresses decision hierarchies. A junior analyst with the right model can now make calls that previously required three layers of approval. Organizations that haven't redistributed decision authority leave speed on the table.

**Organizational structure.** Most companies are still organized around functions and permanent teams built for work as it existed 20 to 40 years ago. AI-native work often requires dynamic teams organized around value streams.

The lesson from prior transformation cycles — Toyota's production system, Agile, DevOps — is consistent: progress stalls when organizations move one dimension without moving the others.

Learning Can't Be a Catalog

AI models update faster than any corporate training curriculum can track. Reeve's answer is what she calls a bidirectional AI learning flywheel: cross-functional pods that run experiments and capture what works, internal champions who carry that learning to the front lines, and a feedback loop that pushes front-line discoveries back up for refinement and redistribution.

PwC operationalizes a version of this through what it calls "prompting parties" — cross-functional groups working through real business problems with AI and teaching each other in real time. The learning is social, specific, and faster than any learning management system.

The Workforce Question Is a Business Question

The World Economic Forum projects 92 million jobs displaced by AI by 2030. It also projects 170 million new jobs created in the same window — a net gain of 78 million roles. The macro pattern from every prior technology transition, from electrification to personal computing, is net positive growth. The business question is who pays for the bridge.

Reeve points to Unilever as a company that has made the calculation explicitly: the cost of displacing a workforce and rebuilding it — recruiting expense, lost institutional knowledge, damaged customer relationships, reputational hit — exceeds the cost of reskilling. Unilever uses AI to match existing employees to emerging roles and treats the investment as long-term strategy.

The companies that don't make that choice will make it eventually, just at higher cost and lower leverage.

Key takeaways

Only 10 percent of companies that attempt generative AI transformation succeed, according to Moderna's CIO Brad Miller — the gap is organizational, not technological.
Reeve identifies nine dimensions organizations must move together; the three most neglected are incentive structures, decision rights, and how teams are organized around work.
Moderna reached 100 percent generative AI adoption in six months by treating AI fluency as a core capability and investing in training, coaching, and process redesign — not just tool deployment.
The World Economic Forum projects 92 million jobs displaced by 2030, but also 170 million new jobs created — the business question is whether companies invest in reskilling or absorb the reputational and financial cost of replacing their workforce.
Static training curricula can't keep pace with AI's update cycle; Reeve advocates for bidirectional learning flywheels where front-line discoveries feed back into organizational knowledge systems in real time.

FAQ

What does Melissa Reeve mean by a 'hyperadaptive' organization?

Reeve defines hyperadaptive organizations as those architected to sense changes faster, learn continuously, and make smarter decisions than any individual could make alone. They've replaced the operating model underneath their AI tools — including how people, processes, and culture move together — rather than layering AI on top of an unchanged structure.

Why do most AI transformations fail, according to the book?

Moderna's CIO Brad Miller, cited in the book, estimates that 90 percent of companies attempting generative AI transformation fail. Reeve's argument is that failure is organizational, not technological: companies skip the workforce mechanisms — training, coaching, process redesign, cultural change — that make adoption stick.

What is a bidirectional AI learning flywheel?

It's Reeve's term for a learning system where knowledge travels in both directions: cross-functional pods capture what works and push it to the front lines, while front-line discoveries feed back up for refinement and redistribution. The goal is a system that updates itself as AI models update, rather than a static curriculum that goes stale.

How should companies think about job displacement from AI?

Reeve frames it as a business decision with a calculable cost. The World Economic Forum projects net job growth from AI by 2030, but the transition period creates real displacement. Companies like Unilever are investing in reskilling because the financial and reputational cost of replacing a workforce exceeds the cost of retaining and retraining it.

What are the most neglected dimensions of AI transformation?

Reeve identifies incentive structures, decision rights, and organizational design as the three dimensions most companies ignore. Specifically: reward systems that penalize failure rather than rewarding learning speed, approval hierarchies that haven't been redistributed to match AI's compression of decision-making, and team structures built around outdated functional silos.

Citations

Your business doesn't need random acts of AI. Here's whyMelissa Reeve's five key insights from 'Hyperadaptive: Rewiring the Enterprise to Become AI-Native,' including the 10-to-90 adoption gap cited by Moderna's CIO and the nine organizational dimensions framework.
Hyperadaptive: Rewiring the Enterprise to Become AI-NativeModerna reached 100 percent generative AI adoption in six months after CEO Stéphane Bancel set a target of 15 new drugs to market in five years, supported by sustained investment in training and process redesign.
Future of Jobs Report — World Economic ForumThe World Economic Forum projects 92 million jobs displaced and 170 million new jobs created by 2030, a net gain of 78 million roles.