A contrarian view on the MIT paper – “The GenAI Divide: State of AI in Business 2025”
While the previous three years saw significant success with AI, the surge in concern over ROI, highlighted by the MIT paper on generative AI (GenAI) pilots, is driven by a fundamental shift in both the technology being deployed and the business context surrounding it. This represents a crucial pivot from the cost-cutting, efficiency-focused AI of the past to the more complex, innovation-driven promises of GenAI today.
From a contrarian perspective, the MIT report and the subsequent concern aren’t signs that AI is failing; rather, they signify that the rules for what constitutes a successful AI project have changed. The 95% failure rate may represent an expected, even necessary, part of the process of discovering a new technology’s true potential, rather than a definitive condemnation.
The difference lies in contrasting traditional AI with the new wave of generative AI.
The changing context behind GenAI and ROI concerns
The recent focus on GenAI pilot ROI is driven by the following factors, which create a stark contrast to the previous, more predictable AI successes:
- Hype vs. reality collision: The intense media hype and fear of missing out (FOMO) around GenAI led many companies to launch pilots without a clear, business-centric problem to solve. The MIT report serves as a reality check, forcing companies to move from technological exploration to a strategic, value-driven approach.
- Productivity vs. transformation: As the MIT report found, many companies mistake individual productivity boosts (e.g., faster email drafting) for enterprise-wide transformation. While useful, these small gains often don’t register as measurable ROI at a corporate scale without deep integration into core workflows.
- Measurement lag: The ROI for transformative GenAI applications, which might lead to entirely new products or business models, is not immediate. In contrast, the success of traditional AI could be measured more quickly by tracking operational efficiencies. The new concern reflects a mismatch between traditional, short-term ROI expectations and the long-term nature of GenAI transformation.
- “Shadow AI” paradox: Employees are finding immense value in using GenAI tools on their own. The corporate concern may stem from the fact that this valuable, decentralized AI usage is not being officially measured or captured, leading to a perception that official projects are failing.
- Shifting risk profiles: The GenAI landscape introduces new risks, including the “verification tax” (the need to fact-check GenAI output) and the risk of “hallucinations”. The verification process can negate the very efficiency gains the technology promises, creating a hurdle for scalable ROI.
- Flawed pilot strategy: Many companies rushed into GenAI pilots with an “AI-first” mentality, experimenting with the tool rather than solving a core business problem. The successful 5% in the MIT study did the opposite: they focused on a specific, targeted problem and integrated the solution deeply into existing workflows. The concern is less about the technology itself and more about the flawed strategy used to deploy it.
Building upon the previous analysis and the contrarian view, the initial struggles of generative AI (GenAI) pilots within enterprises can be attributed to the complex nature of the enterprise ecosystem itself. Unlike the more focused, efficiency-driven traditional AI projects of the past, GenAI adoption must navigate a landscape of organizational inertia, cultural resistance, and systemic complexities. The initial phase often resembles a clash between a revolutionary technology and an established, risk-averse environment.
The inherent complexity of the enterprise ecosystem:
1. Legacy systems and data silos
Enterprise systems are built on decades of investment and accumulated data. This infrastructure, often referred to as “legacy systems,” is highly rigid and was not designed to accommodate the dynamic, data-intensive demands of GenAI.
· Data fragmentation: Critical data is often scattered across disparate systems and departments. GenAI requires consolidated, high-quality data to function effectively, making data consolidation a massive and costly undertaking.
· Infrastructure incompatibility: Many legacy platforms lack the real-time processing capabilities, modern APIs, and cloud compatibility required for GenAI workloads. Custom integration efforts become expensive and difficult, slowing progress.
2. Organizational culture and resistance to change
Enterprise culture is often built on risk aversion and maintaining the status quo. This can create a significant barrier to the adoption of a transformative, yet unpredictable, technology like GenAI.
· Fear of the unknown: Employees may fear that GenAI will replace their jobs, leading to active or passive resistance. This fear is heightened by GenAI’s ability to automate complex, creative tasks previously considered safe from automation.
· Misaligned incentives: Employee performance is often measured by existing metrics that reinforce old habits. Without new, aligned incentives, employees have no motivation to invest the time and effort needed to master new GenAI workflows, especially if they perceive the benefit as unclear.
· Lack of experimentation: Many enterprises have a culture that punishes failure. GenAI adoption, which requires a “fail fast, learn often” approach, clashes with this rigid, error-averse mindset.
3. Leadership and strategic misalignment
Effective GenAI integration requires clear vision and strategic alignment from the highest levels of leadership, which is often missing in the initial pilot phases.
· Lack of a clear strategy: Many pilots are driven by hype rather than a clearly defined business problem. Without a strategic roadmap, projects can become aimless experiments with no clear path to scaled ROI.
· Measurement shortcomings: Leaders may apply traditional, short-term ROI metrics (common for cost-efficiency projects) to GenAI pilots, whose value is often longer-term and harder to quantify. This mismatch in expectations inevitably leads to the perception of failure.
· Overlooking the human element: Some leadership teams focus purely on the technology, overlooking the critical importance of change management, employee training, and addressing user concerns.
4. The resource paradox
While large enterprises possess immense resources, deploying them effectively for GenAI is a challenge.
· Resource allocation: GenAI requires a different allocation of resources than traditional tech projects, demanding significant investment in data preparation, high-performance computing, and specialized talent. A poor allocation of the GenAI budget, as noted in the MIT study, can starve high-potential projects.
· The talent gap: Enterprises often lack the specialized data scientists, machine learning engineers, and prompt engineers needed to get the most out of GenAI. Relying solely on third-party vendors can increase costs and reduce in-house expertise.
5. Integration complexity and governance gaps
The very qualities that make GenAI so powerful—its flexibility and creative output—introduce new integration and governance challenges for the enterprise.
· Integration friction: Unlike past AI which focused on targeted automation, GenAI is meant to be deeply integrated into workflows that span multiple departments and legacy systems. This requires a much higher degree of coordination and cross-application support.
· Lack of governance: Many companies launched GenAI projects without clear policies on data usage, privacy, security, and ethics. The risks associated with GenAI (e.g., hallucinations, sensitive data leaks) require new governance frameworks that were not in place during the initial pilot phase.
In essence, the enterprise ecosystem acts as a formidable, complex testing ground for GenAI. The initial pilot failure rate reflects not the inherent flaws of the technology itself, but the predictable, messy process of a conservative, structured organism attempting to ingest and adapt to a radical, disruptive new technology.
Conclusion: Reframing the discussion
The conclusions of the MIT report are not necessarily flawed, but its highly-publicized failure rate is likely misleading and potentially premature. A more nuanced, alternate take on the data would suggest:
- The “95% failure” is a normal part of the innovation adoption cycle, not a condemnation of GenAI itself.
- Initial ROI is a poor metric for a disruptive technology whose real value is long-term and transformative.
- The pilots that “fail” are providing valuable lessons and revealing the crucial areas of friction that need to be addressed before scaling.
In this context, the conversation shifts from one of failure and concern to one of patient, strategic investment and learning. Enterprise adoption of GenAI isn’t a race to the finish line but an exploration of a new continent, and the early, low-return phases are critical for mapping the landscape before building the city.
About the Author
Debiprasad Banerjee is the CEO of Acxelin Agentix, an AI-first company helping enterprises reimagine customer engagement through intelligent automation and digital experiences. Through his writing, he shares practical insights on AI, innovation, and the future of business. Explore more articles from the Acxelin Agentix team for expert perspectives on enterprise AI and digital transformation.