Why CEOs Keep Betting on AI When Half Their Projects Fail

The AI Investment Paradox

The numbers tell a paradoxical story. According to Teneo's annual survey of more than 350 public-company CEOs, 68% plan to increase artificial intelligence spending in 2026. The twist? Less than half of their current AI projects have generated more value than they cost. To outside observers, this might appear to be a case of collective delusion among business leaders, throwing good money after bad. The reality is far more nuanced and reveals critical insights about technology adoption, competitive strategy, and the true nature of transformation.

This investment pattern demands serious examination. Are CEOs demonstrating strategic foresight or succumbing to herd mentality? The answer matters not just for shareholders but for understanding how transformative technologies reshape competitive landscapes.

The AI Investment Paradox

Ben Glickman's Wall Street Journal report on the Teneo survey exposes a tension at the heart of current business strategy. Public companies with revenues exceeding $1 billion are committing substantial capital to AI initiatives that frequently fail to deliver measurable returns. Yet rather than pulling back, leaders are accelerating their investments.

This behavior defies conventional capital allocation wisdom. Standard financial discipline suggests that when investments underperform, prudent managers should reassess, redirect resources, or cut losses. The persistence of AI spending despite poor returns suggests that CEOs are operating under different assumptions about value creation, competitive necessity, and timeline expectations.

The survey data reveals several critical dimensions of this investment pattern. First, there exists a substantial disconnect between CEO and investor expectations about timing. While 53% of institutional investors expect AI initiatives to deliver returns within six months, 84% of large-company CEOs believe meaningful returns require more than six months. This gap represents more than mere disagreement over timelines. It reflects fundamentally different mental models about how transformative technology creates value.

Investors, under pressure to show quarterly results and justify valuations, naturally focus on near-term payoffs. CEOs responsible for long-term competitive positioning understand that certain technologies require extended learning curves, organizational adaptation, and market development before generating returns. The question is whether CEOs are correct in their patience or whether they are rationalizing poor decisions.

The Pattern Recognition Problem

History offers instructive parallels. When companies first adopted enterprise resource planning systems in the 1990s, implementations frequently ran over budget, took longer than expected, and initially disrupted operations rather than improving them. Many early ERP projects failed completely. Yet companies that persisted and learned from failures eventually gained substantial competitive advantages through integrated data and streamlined processes.

Similarly, cloud computing adoption followed a pattern of skeptical early resistance, experimental investments with mixed results, and eventual widespread adoption as capabilities matured and organizations developed expertise. Amazon Web Services reported losses for years before becoming one of Amazon's most profitable divisions.

These historical precedents suggest that poor initial returns do not necessarily indicate bad strategy. The critical question is whether current AI investments represent a similar pattern of learning and capability building that will eventually pay off, or whether they reflect something different.

Research from MIT Sloan Management Review indicates that companies struggle to measure AI ROI because traditional financial metrics capture immediate costs but fail to quantify transformational benefits. How do you value faster decision-making, improved customer insights, or enhanced employee productivity? These benefits resist easy quantification yet represent real competitive advantages.

McKinsey Global Institute's research on AI adoption provides additional context. Their findings show that while 65% of organizations regularly use generative AI, only 23% have achieved cost decreases of 10% or more in specific business areas. However, organizations that implement AI at scale with strong governance structures see returns three times higher than those with scattered, experimental approaches.

This data point illuminates a crucial distinction. The problem may not be AI investment itself but rather how companies approach implementation. Scattered pilot projects without integration into core operations predictably fail to generate returns. Systematic, scaled implementations with clear governance and change management produce substantially better results.

The Application Area Divide

The Teneo survey reveals that AI success varies dramatically by application area. CEOs report the most success using AI in marketing and customer service while struggling to implement it effectively in higher-risk domains like security, legal, and human resources.

This pattern makes strategic sense. Marketing and customer service applications typically involve pattern recognition, personalization, and interaction at scale—tasks well-suited to current AI capabilities. These areas also permit experimentation with lower risk. A slightly imperfect product recommendation rarely causes serious harm.

Conversely, security, legal, and HR applications involve high-stakes decisions where errors carry significant consequences. An AI system that misses a security threat, provides incorrect legal guidance, or makes biased hiring decisions can create substantial liability. Current AI systems, despite impressive capabilities, still lack the reliability and explainability required for these high-risk applications.

This divide suggests that successful AI implementation requires matching technology capabilities to appropriate use cases. Companies that force AI into applications where current technology limitations create unacceptable risk predictably generate poor returns. Those that focus on areas where AI strengths align with business needs see better results.

Walmart provides an instructive example. The retail giant has invested billions in AI for supply chain optimization, inventory management, and logistics—applications where pattern recognition and prediction create clear value. These investments took years to show positive returns but now provide substantial competitive advantages through reduced costs and improved product availability.

Contrast this with IBM Watson's struggles in healthcare. Despite massive investment and aggressive marketing, Watson failed to deliver promised value in clinical decision support. The complexity of medical reasoning, the need for explainability, and the high stakes of healthcare decisions exceeded current AI capabilities. IBM eventually scaled back its healthcare AI ambitions significantly.

The Employment Paradox

One of the survey's most striking findings contradicts prevailing narratives about AI and employment. Rather than expecting AI to reduce headcount, 67% of CEOs believe it will increase entry-level positions while 58% expect growth in senior leadership roles.

This counterintuitive finding warrants careful analysis. If AI automates tasks and improves productivity, why would companies need more employees? Several explanations merit consideration.

First, AI may be creating new work categories rather than simply replacing existing ones. As AI handles routine tasks, organizations discover new opportunities that require human judgment, creativity, and relationship building. AI-powered customer insights might automate data analysis but create demand for strategic customer relationship roles. Automated content generation might reduce routine writing but increase demand for strategic communications professionals who guide AI tools and ensure quality.

Second, organizations implementing AI successfully often expand operations into new markets or service categories that their enhanced capabilities make feasible. This expansion creates net new positions even as AI improves productivity in existing operations.

Third, AI implementation itself requires substantial human capital. Organizations need data scientists, AI ethics specialists, change management professionals, and trainers to implement AI successfully. These roles represent new headcount driven by AI adoption.

Microsoft's experience with AI integration illustrates this dynamic. As the company has embedded AI throughout its product portfolio through Copilot features, it has simultaneously expanded hiring in AI research, implementation, ethics, and customer success. The technology augments human capabilities rather than simply replacing them.

However, this optimistic employment picture deserves skepticism. CEOs may be underestimating AI's eventual impact on headcount, or they may be in an investment phase that precedes eventual workforce reduction. History shows that automation technologies often create jobs initially during implementation but reduce them over longer timeframes as capabilities mature and organizations optimize.

The Size and Sentiment Divide

The survey reveals a striking difference in economic outlook between large and small company CEOs. Optimism among large-company CEOs about economic improvement in the first half of 2026 dropped from 51% to 31%, while smaller-company CEOs remain bullish at 80%.

This divergence likely reflects different exposure to global trade tensions, geopolitical uncertainty, and regulatory complexity. Large multinational corporations face greater impact from tariffs, supply chain disruptions, and international policy changes. Smaller companies with more domestic focus and greater operational agility may feel better positioned to navigate uncertainty.

This sentiment gap has implications for AI investment patterns. Large companies facing economic headwinds might view AI as a efficiency driver to maintain margins during difficult periods. Smaller companies feeling optimistic might pursue AI for growth and market share expansion. These different strategic contexts shape how organizations approach AI implementation and measure success.

The Measurement Challenge

A fundamental problem underlying poor AI returns may be measurement methodology rather than actual value creation. Traditional ROI calculations work well for discrete capital investments with clear inputs and outputs. AI implementations often resist this framework.

Consider an AI system that analyzes customer feedback and identifies emerging product issues. The system costs $500,000 annually to operate. How do you measure returns? You might count prevented recalls, but how do you value early detection of quality issues that would have eventually caused problems versus issues that would have resolved naturally? How do you quantify preserved brand reputation or maintained customer trust?

These measurement challenges mean that reported AI returns likely understate actual value while fully capturing costs. This asymmetry biases ROI calculations downward.

Gartner's positioning of many AI technologies in the "trough of disillusionment" on their Hype Cycle supports this interpretation. Inflated initial expectations have met practical reality, causing disappointment. However, Gartner predicts that 5-10 years forward, mature implementations will deliver substantial value. Organizations that persist through the current challenging period while learning and adapting will be positioned to capture that value.

The Strategic Necessity Argument

Perhaps the most compelling explanation for continued AI investment despite poor returns is strategic necessity. In markets where competitors are investing heavily in AI, choosing not to invest may be riskier than investing without immediate positive returns.

This dynamic resembles an arms race. No individual participant necessarily wants to continue spending, but the risk of falling behind competitors makes continued investment mandatory. The cost of leadership sees as the price of maintaining competitive position rather than an optional growth investment.

Amazon's AI journey illustrates this principle. The company invested in machine learning and AI for over a decade, funding research and applications that often generated unclear returns. These sustained investments eventually enabled the product recommendation engine, logistics optimization, and AWS AI services that now provide massive competitive advantages. Companies that waited for proven ROI before investing now face substantial catch-up costs and capability gaps.

This strategic logic explains why CEOs persist with AI investment despite current poor returns. The question is not whether current AI projects generate positive ROI but whether AI capabilities will be competitively necessary five years forward. If the answer is yes, then current losses represent learning investments rather than wasted capital.

However, this logic has limits. Not every technology that seems transformative actually transforms industries. Some remain perpetually promising without delivering. CEOs must distinguish between strategic necessity and herd behavior.

The Path Forward

Several principles emerge from examining current AI investment patterns:

• First, organizations should match AI applications to appropriate use cases. Focus on areas where current AI capabilities align with business needs and risk tolerance. Success in targeted applications builds capability and credibility for more ambitious implementations.
• Second, companies must develop better measurement frameworks that capture transformational value beyond traditional ROI calculations. This requires identifying leading indicators of AI impact and tracking them consistently over extended periods.
• Third, successful AI implementation requires systematic approaches with strong governance rather than scattered experiments. Organizations should develop clear AI strategies, establish appropriate oversight structures, and commit to learning from both successes and failures.
• Fourth, leaders should communicate realistic timelines to stakeholders. The disconnect between investor expectations for six-month returns and CEO beliefs that value requires longer periods creates pressure for premature abandonment of promising initiatives. Education about realistic AI maturation timelines helps align expectations.
• Fifth, organizations should view current AI investments partially as capability building rather than purely financial returns. Developing organizational competence in AI implementation, data management, and change processes creates long-term value even when individual projects disappoint.

Conclusion

The paradox of continued AI investment despite poor returns reflects the complex reality of transformative technology adoption. CEOs are not simply throwing money at overhyped technology, nor are they demonstrating infallible strategic wisdom. Instead, they are navigating genuine uncertainty about competitive necessity, capability development, and value measurement.

The difference between strategic foresight and expensive mistakes often becomes clear only in retrospect. Organizations that persist with AI investment while learning from failures, matching applications to capabilities, and building systematic implementation approaches will likely look prescient five years forward. Those that chase AI for its own sake without strategic clarity will likely waste substantial capital.

For business leaders, the key is not whether to invest in AI but how to invest strategically. This means accepting that early returns will be disappointing, focusing on appropriate applications, developing robust measurement frameworks, and building organizational capabilities for the long term. The companies that navigate this transition successfully will gain significant competitive advantages. Those that either abandon AI due to poor early returns or invest indiscriminately without learning will find themselves at a disadvantage.

The Teneo survey data showing 68% of CEOs planning increased AI spending despite less than half of projects generating positive returns is not necessarily cause for alarm. It may instead represent rational behavior by leaders who understand that transformative technology requires patience, learning, and persistent investment before delivering value. The critical question is whether individual organizations are pursuing AI with appropriate strategy and discipline, not whether they are investing at all.