AI Transformation: The Costly Mistakes Companies Keep Making

A case-study from Nike's attempt in manufacturing transformation

I (and others) have frequently discussed the challenges of AI adoption in the past. These challenges include translating metrics, developing measures of real-world efficacy, conducting evaluation and validation, and ensuring readiness at various levels—technological, structural, and operational. Other hurdles involve integration, workforce upskilling, and change management (see some links in the concluding remarks below).

AI has significantly improved products, services, and user experiences in many areas. However, most of these advancements have resulted from careful adoption strategies rather than hype. Such strategies include solving the right, feasible, and necessary problems; finding a product-market fit; establishing an informed evaluation and gating mechanism; identifying the risk-reward trade-off to guide investments and track ROI; aligning with the overall business strategy; and assessing both business and technology readiness, among other factors.

Despite many efforts, both internally and externally, the failure rate of AI transformation initiatives has remained high (and I suspect we will see a similar pattern with GenAI). From a GenAI perspective, there has been extensive discussion about technology readiness—specifically: Is GenAI effective? Are the models truly as good as claimed? Do they actually have reasoning capabilities? Can they do more beyond providing a human-like interface? And so on.

Regardless of my personal opinion, the jury is still out on most aspects. However, technology adoption involves far more than just technical readiness. The readiness and effectiveness of any technology are typically subjective. As the well-known saying, widely attributed to George E. P. Box, states: "All models are wrong. Some are useful." GenAI may be useful for certain use cases while remaining unsuitable for others.

The line that demarcates its utility, in simple terms, correlates with the sensitivity (error tolerance) of applications. Since the technology is not yet mature and has persistent challenges in certain areas, a major determinant of the use cases it supports is the requirement for accuracy in predictions and associated error tolerances. Again, these aspects continue to be actively discussed, along with broader ethical and societal perspectives on GenAI’s impact.

However, when it comes to business impacts, due diligence and readiness for transformation have often taken a back seat. AI (and other) “transformation” efforts are typically driven by hype, combined with an enthusiastic coterie of stakeholders. Unfortunately, this enthusiasm comes at a steep price—the lack of proper due diligence.

This due diligence isn't just about understanding technology feasibility (which is often assessed poorly) or a business’s operational readiness. It also involves understanding the broader business vision, priorities, trade-offs, and the necessary alignment of the proposed transformation.

Wall Street Journal’s latest article on Nike’s (and other shoe manufacturers’) experience with robotics automation serves as a textbook case study of why such transformations fail—and, more importantly, why these failures are foreseeable rather than surprising.

Let me summarize the relevant points from the story here:

Around 2015, Nike launched an initiative to automate its manufacturing, leveraging advancements in robotics innovation. (Yes, this was during the period when hype surrounded robotics—often powered by ML—and additive manufacturing, with significant sums of money invested in these areas.)

The Goal, as noted in the article:

Make tens of millions of Nike sneakers at a new high-tech manufacturing site in Guadalajara, Mexico, by 2023.

Other shoe manufacturers, including Adidas and Under Armour, also made significant efforts, but Nike’s was the most ambitious. Nike partnered with the automation company Flex, undertaking a sizeable initiative (emphasis mine):

Nike and Flex established new production lines that used machines commonly seen in electronics manufacturing—but rarely shoemaking—such as a “pick and place” machine that is known for mounting components onto circuit boards. The machines were supposed to build the upper part of a shoe, knit fabric, add logos and glue the sole.

This suggests significant investments and, more importantly, a high likelihood that there was an established trust in the underlying technology supporting this transformation. However:

The effort quickly ran into trouble.

The robots struggled to handle the soft, squishy and stretchy parts that are integral to shoemaking. Shoe fabrics also expand and contract depending on the temperature, while in shoemaking no two soles are exactly alike.

Human workers can adapt to such challenges, but it proved difficult for machines.

…

…factory production never became as automated as envisioned.

At the core of this failure were challenges in adopting and adapting smart robotics capabilities to Nike’s specific needs.

The article notes: "Task after task proved challenging to automate, like the delicate work of gluing soles to the upper part of the shoe." For example, an executive explained, "If you didn’t lay it the right way, there would be a noticeable twisting of the shoe—a misalignment that, aesthetically, would fail quality tests."

Nike grants designers significant freedom (and rightly so). However, the robotics capabilities proved inflexible in adapting to design changes. The issue was so severe that, at one point, "It took the Flex team eight months to figure out how to automate a way to put the Nike swoosh on a shoe, only for Nike to move on to a new shoe line—rendering the method Flex developed obsolete."

Nike faced a tough decision: sacrificing design freedom, material complexity, and variety in favor of scaling and automation. Ultimately, the company chose to preserve its design integrity—a commendable decision that avoided brand damage and potential negative market impact.

Delays and rising costs with no meaningful output first unsettled Flex’s investors, soon followed by Nike’s leadership, leading to the quiet closure of the initiative in 2019—without fanfare.

Analysis

I’d concede that any transformation effort involves risks. I’d also concede that transformations require bold initiatives backed by investments, patience, and a willingness to change. To Nike’s credit, all of these were demonstrably present—entire manufacturing lines were dedicated to the effort, more than four years were invested, and substantial financial and human resources were committed to setting the initiative up for success.

Yet, there was one major flaw. Based on everything in the article, due diligence appears to have been almost entirely absent.

One doesn’t need to be a world-class expert in robotics or AI/ML to recognize that adapting robots to new tasks is a complex challenge (some progress has been made since then, but it remains non-trivial). Having been involved in both manufacturing transformation initiatives and AI/robotics research, I would consider these the first aspects to evaluate and prioritize for feasibility testing. So would any executive or practitioner worth their salt in the space. Technology adoption cannot be based solely on demos or uncorrelated past successes.

It seems that the primary (or perhaps only?) basis for Nike pursuing this large-scale transformation was Flex’s previous success in setting up a complex Texas factory for Apple to manufacture Mac Pros. While specific details aren’t available, it’s clear that these two industries—and consequently, their automation requirements—demand entirely different technological capabilities. Nike appears to have overlooked the feasibility of its proposed automation tasks, as well as the intricacy of adapting robotics to its production processes.

This oversight is reminiscent of the many transformation efforts I have both witnessed and advised against across various industries. Even more puzzling is the justification behind Nike’s substantial investment—dedicated manufacturing lines, capital expenditures, and personnel allocations—all for an initiative that lacked rigorous feasibility assessments.

It is puzzling that Flex failed to anticipate the constraints of its technology. Or perhaps it was overly optimistic, underestimating the complexity of the challenges. Equally unclear is what sort of assessment framework Nike employedto establish the collaboration—not to mention the upfront investment made without clear markers to evaluate the technology’s capacity to scale, adapt, and address the complexity of its business needs.

It is also evident that the initiative was either misaligned with—or worse, entirely unaware of—Nike’s core philosophy and priorities (e.g., design freedom, rapid new model introductions, design complexity, materials complexity, etc.). Nor did it appear that the transformation efforts aligned with the company’s broader vision. If one were to rationalize this misalignment, it could be argued that the initiative supported Nike’s attempt to diversify its manufacturing beyond China, Vietnam, and Indonesia. Still, launching an initiative with such critical priorities while lacking essential information is perplexing.

However, the apparent technical feasibility assessment failures likely stem from deeper issues rooted in company culture. These failures can signal issues such as organizational dysfunction, a lack of internal visibility and transparency, leadership challenges, readiness gaps, and misalignment between the company’s overarching vision and its operational strategy, among other factors.

Some takeaways

Back in 2017, I authored an article on AI transformation, exploring why transformations fail and what should be done to prevent such failures. The article examines the deeper aspects discussed above, and the Nike case study perfectly aligns with those findings.

Reconciling Nike’s story with numerous other cases of failed transformation efforts, here are some additional points worth considering:

Such failures would be entirely foreseeable and avoidable if the requisite due-diligence is done well on all levels, and it factors in the decision making process. The most common reason for why that doesn’t happen are cultural: Company goals diverging from personal (individuals’/leaders’) goals; voluntary conformity (teams conditioned to deliver analyses aligned with the “expectations”), or; involuntary conformity (rigorous due-diligence analyses ignored when they clash with the “expected” outcomes).

Given this, it is difficult to attribute the absence of due diligence to mere oversight or error.

Unfortunately, corporate culture often resists objectivity—let alone constructive dissent. This isn't an exaggeration; rather, these issues are prevalent in a significant number of companies.

Expertise in emerging technologies—and, more importantly, strategic decision-making—cannot be outsourced entirely. These failures underscore deeper organizational gaps. Having the right in-house resources, empowered teams, and leaders capable of making objective, informed decisions is crucial for the success of transformation initiatives.

Technology teams are no longer simply “build-to-specifications” teams. If businesses want to undertake successfultechnology-driven transformation, these teams must have a seat at the table.

Due to the compartmentalized nature of organizations—whether for structural or personnel reasons—there is often a significant informational gap between a company’s core vision, strategy, and priorities across different departments. Addressing these critical organizational challenges is essential to achieving true readiness.

Technology is evolving at a rapid pace, but just as hope is not a strategy, neither is hype. While GenAI faces intense market and competitive pressures, companies should not only focus on understanding the technology—they must also assess its readiness and alignment with their business needs. Additionally, they must entrust capable leaders to provide objective advice, ensuring those insights are meaningfully incorporated into the decision-making process.

So far, the outlook on GenAI transformation efforts across companies appears extremely challenging. The factors discussed above offer some insight into why many organizations struggle with the overwhelming and complex task of realizing GenAI’s ambitious ROI promises. I cover additional aspects here, here, and here. Ultimately, successful transformation depends on far more than just the underlying technology.

Success of business transformations is anchored in the right cultural foundations.