Designing for AI-Driven Organizational Change: Part 1 – Where Are We Now?

Right now, the topic of AI productivity is pretty confusing. Even seasoned students of long-term technology cycles are scratching their heads.

AI seems to be sitting in a middle stage of Carlota Perez’s classic model, laid out in Technological Revolutions and Financial Capital: The Dynamics of Bubbles and Golden Ages.[i] Perez describes each technological revolution unfolding in two main periods – installation and deployment – separated by a turning point that resets the relationship between financial capital and real production.

During installation, speculative money floods into the new technology, inflating bubbles that eventually pop. This might be what is happening now. According to Perez, only after that reckoning does the deployment phase begin, when the technology’s benefits finally spread widely enough to lift productivity and living standards, not just stock prices. The many articles we see about the “AI bubble” reflect the fact that for now, financial capital has uncoupled from actual capital returns on production of wealth versus speculation. In other words, we have investment, but we don’t yet have return in the form of productivity.

The bridge between the uncoupling and recoupling of capital to economic reality often travels through troubled times. Again, this might be what is happening now. Think bubbles, crashes and lots of sorting out across the entire ecosystem.

Of course, if you lead an individual organization through this period, all that may not matter. (Except for the chaos and turbulence that accompanies this bridging.) Whether or not AI “counts” as a separate 50- to 70-year technology wave, there is pressure to transition to AI use and secure its potential productivity gains. So, a leader should focus on what we know about leading successful change with respect to technology and specifically AI.

A technology’s real value depends on one thing: whether people actually change their behaviors – the way they work. Electric motors only sped up production once organizational leaders reorganized factories around them (a transition that took over 20 years). 

Faster trains add little without tracks built for faster speeds. For example, Amtrak’s new Acela train (the Avelia Liberty) is capable of going 186 miles per hour. But for now, the train is mostly limited to traveling 135 miles per hour. Why? Because the tracks along the New York to DC corridor were for the most part assembled in the 1830’s (the tunnel into Manhattan was opened in 1910). 

It’s a new-wine-in-old-wineskins problem: Pour new wine into an old container, and the fermentation bursts it, wasting the wine and wrecking the skin. That’s true at the company level and the economy level. As Michael Hammer puts it in a 1990 Harvard Business Review article on reengineering work: Don’t automate, obliterate. The same logic applies to AI.

Treating AI as a tool you simply bolt onto existing processes is like running a new high-speed train on old tracks. The engine is capable of 186 mph, but the rails, signals and century-old bridges cap it at a fraction of that – so you get incremental speed, not transformation.

AI works the same way: Dropped into unchanged workflows, approval chains, and org structures, it can only go as fast as the “track” beneath it allows. To have a truly strategic commitment to AI means being willing to redesign the operational tracks it runs on. It means first imagining and then building the infrastructure to match the ambition.

But first, a word about failure.

Are we there yet?

In the original research for our book, “Leading Successful Change: 8 Keys to Making Change Work” in 2013 we looked at studies published over the past 20 years. All of them agreed that between 50% to 75% of change efforts fail. We revised the book in 2020, and the updated research reported that 70% of digital transformation efforts failed.

Two recent studies deliver the same depressing news. In 2025, MIT published the NANDA study with the headline finding that 95% of enterprise generative AI pilots fail to deliver measurable profit and loss impact.  Only 5% achieve what the authors call “rapid revenue acceleration.”

The core culprit isn’t model quality – it’s what MIT calls the “learning gap”; organizations deploying AI without redesigning the workflows, decision rights, or workplace design changes that would allow it to deliver value.[ii]

In July 2026, McKinsey published a study where only 11% of 750 respondents reported that their reorganization achieved “reinvention” – the horizon where AI fundamentally rewires how work gets done. The majority remain in what McKinsey calls “enablement” (deploying tools for individual tasks) or “automation” (improving existing workflows). Most importantly, the majority across all three horizons say AI has yet to deliver meaningful enterprise value in terms of business performance, cost savings, employee experience or customer outcomes.[iii]

What does success look like?

A clear majority of organizations, probably over 80%, have implemented some form of AI somewhere in their organization, but only a fraction of them have seen any measurable positive return. Many organizations apparently either can’t accurately measure the return or don’t know the true extent of the use of AI within the organization. 

Some leaders see AI primarily as a way to cut headcount and costs while others see it more as a way to free people up to concentrate on more advanced and ‘human’ work. Some leaders seem to see it largely as a tool while others see it as a precipitator of fundamental organizational change.

What is the difference between these three AI change use cases? 

Example #1 – American Express fraud detection

American Express began applying machine learning to fraud detection in 2010, making it one of the earliest adopters of AI among financial services companies. Its machine-learning-powered fraud detection model now monitors more than $1.2 trillion in transaction value every year, generating a fraud decision in milliseconds for every card transaction worldwide. Rather than relying solely on static fraud rules, the system evaluates transaction anomalies, spending velocity, geographic patterns and merchant behavior in real time — using GPU-accelerated systems that can process fraud decisions within milliseconds [iv] per a Nvidia case study cited by Emerj.

Example #2 – John Deere – See & Spray precision herbicide application

John Deere’s See & Spray technology, developed through its $305 million acquisition of Blue River Technology in 2017, uses boom-mounted cameras to scan over 2,100 square feet of crop area per second. Deep neural networks classify each plant at pixel level – weed or crop – and machine learning algorithms powered by Nvidia Jetson Xavier processors make real-time spraying decisions in the field. In 2024, the technology saved farmers an estimated 8 million gallons of herbicide and delivered average herbicide savings of 59%. For some Iowa fields, savings reached as high as 90.6%.[v]

Example #3 – Blackstone – legal and compliance AI transformation

Blackstone’s Legal and Compliance group faced a structural challenge: Investor-facing material volumes were expected to increase 25% by 2027, but adding headcount at that rate wasn’t viable. The instinct might have been to deploy an AI tool on top of the existing workflow – a point solution.

Instead, the team chose to redesign decision flows first, before touching technology at all. They clarified ownership, codified legal precedent into structured decision rules, and standardized repeatable reviews so that materials which had already passed scrutiny multiple times could advance automatically – with explicit escalation paths for anything requiring senior judgment. Only then did Blackstone embed AI directly into day-to-day workflows as the first-pass reviewer: triaging volume, checking consistency, and detecting anomalies across thousands of documents.[vi]

In Part II, we will revisit these examples using a framework to explore their differences.  And we’ll add a case that illustrates how to go about capturing AI’s biggest organizational payoff:  transformational change, not incremental improvement.

[i] Carlota Perez, Technological Revolutions and Financial Capital: The Dynamics of Bubbles and Golden Ages (Cheltenham, UK: Edward Elgar Publishing, 2002).

[ii] Challapally, Aditya, Chris Pease, Ramesh Raskar, and Pradyumna Chari. The GenAI Divide: State of AI in Business 2025. Cambridge, MA: MIT NANDA (Project NANDA), July 2025.

[iii] De Smet, Aaron, Drew Goldstein, Holly Price, and Tanguy Catlin. “From Adoption to Impact: Three Horizons of AI Transformation.” McKinsey & Company, July 8, 2026. https://www.mckinsey.com/capabilities/people-and-organizational-performance/our-insights/from-adoption-to-impact-three-horizons-of-ai-transformation.

[iv] “Artificial Intelligence at American Express,” Emerj,  June 2026,  https://emerj.com/artificial-intelligence-at-american-express-2/

[v] “Artificial Intelligence at John Deere”, Emerj, October 2025, https://emerj.com/artificial-intelligence-at-john-deere/

[vi] Blackstone’s Legal & Compliance AI Transformation Started With Technology. It Succeeded Because It Put People First,” McKinsey & Company, 2026, https://www.mckinsey.com/capabilities/people-and-organizational-performance/how-we-help-clients/blackstones-legal-and-compliance-ai-transformation-started-with-technology