The AI Dilemma for Executives — And What to Do About It
Artificial Intelligence presents a uniquely difficult dilemma for today’s executives – CEOs, CFOs, CIOs, COOs, and Boards alike.
On the one hand, there is no shortage of studies, articles, and industry research showing that the vast majority of AI initiatives fail to produce meaningful return on investment. Depending on the source, estimates range from 70% to as high as 95%. Gartner has repeatedly warned that most AI projects never make it into production or fail to scale. McKinsey has published similar findings, noting that while experimentation with AI is widespread, only a small fraction of companies capture sustained business value. Deloitte and Accenture echo the same conclusion: lots of pilots, lots of excitement — and very little durable ROI.
On the other hand, AI clearly cannot be ignored. This is not a passing technology fad that executives can simply wait out.
There are numerous predictions from knowledgeable people about how deeply AI will transform both society and business. Elon Musk has warned about the risks of AI if mishandled. Sundar Pichai has compared AI’s impact to that of electricity or fire. Sam Altman has spoken openly about AI reshaping entire categories of work. While some of these predictions may be overblown in the short term, given enough time they are almost certainly directionally correct.
Compounding the dilemma further, AI is not just about incremental personal productivity gains around the edges — though it is already doing that. AI has the potential to fundamentally transform entire industries and core business processes, including how companies plan, operate, serve customers, manage risk, and allocate capital.
Another complicating factor is time scale. Some pundits are talking about radical transformations occurring over the next several years, not decades. Whether those timelines ultimately prove optimistic or not, executives are under growing pressure to act now.
And the stakes for failure are unusually high. Failed AI initiatives can result in:
- Public embarrassment or loss of credibility
- Legal and regulatory ramifications
- Significant sunk costs
- Brand damage
- Loss of customer trust that can take years to rebuild
Adding yet another multiplier is the FOMO factor — fear of missing out. Almost every day, executives read about companies announcing layoffs due to “AI-based efficiency.” Boards ask questions. Investors ask questions. Employees ask questions.
A brief but important aside: both of these things cannot be true at the same time — that up to 95% of AI initiatives fail to deliver ROI, and that countless companies are achieving massive efficiency gains from AI. There is growing evidence, or at least a strong hypothesis, that many of these layoffs would have occurred anyway, and AI has become a convenient cover for making them. Executives should take such announcements with a healthy degree of skepticism.
This is the AI dilemma: Executives must act, yet failure is rampant, the stakes are high, and there is no established playbook for success.
Why Most AI Initiatives Fail
Understanding why AI initiatives fail is a prerequisite to succeeding with AI-based transformation. In our experience, failure rarely comes down to a single issue. Instead, it is usually a combination of several common factors.
- Underestimating the Need for Clean, High-Quality Data: AI runs on data. Yet organizations consistently underestimate what it takes to prepare their data for AI success. Data that may be “good enough” for reporting or analytics is often not fit for AI and machine learning. Poor data quality, inconsistent definitions, missing history, and fragmented sources undermine even the most sophisticated models.
- Proof of Concept Is Not Production: A one-time proof of concept (POC) is not a reliable indicator of production success. During POCs, teams often go to extraordinary lengths to manually collect, scrub, cleanse, and shape data. These efforts cannot be easily replicated at scale or sustained on an ongoing basis. Additionally, real-world environments change over time. Data patterns shift, business conditions evolve, and models degrade — a phenomenon known as model drift. Many organizations fail to plan for this reality.
- Lack of Rigor Around Use Cases and ROI: Too many AI initiatives begin with vague aspirations or “magical thinking.” AI is not magic. It is a powerful new tool and capability, but it does not eliminate the need for standard business discipline. Expected ROI must be quantified upfront, with clear assumptions, success metrics, and accountability. Without this rigor, AI initiatives drift and ultimately fail.
- Failure to Think Through Productionalization: The optimism and energy required to complete a POC often crowd out more practical questions:
- How will this AI solution be used day-to-day?
- How will data be refreshed regularly?
- How will models and rules be updated?
- How will workflows change?
When these questions are not addressed early, promising initiatives stall.
- Over-Reliance on AI Capabilities: Some organizations overestimate what AI can do and design processes that rely too heavily on automation without sufficient safeguards. Over-reliance on AI without appropriate controls, checks, and balances has led to several high-profile failures.
- Underestimating the People Aspect: AI “blindly doing its thing” without people involved is dangerous — at least with today’s technology. Humans must remain involved in oversight, interpretation, and decision-making. That requires training, change management, and organizational readiness, which are often underestimated.
- Lack of Alignment to Strategic Objectives: AI initiatives frequently fail the “so what?” test. Just because AI can do something does not mean it should. If AI initiatives are not aligned to an organization’s strategic objectives and core purpose, they become distractions rather than enablers.
- Getting Ahead of the Technology: AI capabilities are advancing rapidly, but not all technologies are ready for large-scale production use. Some approaches work well in controlled environments or at small scale but break down in real-world conditions. Moving too far ahead of the technology introduces unnecessary risk.
The AI Playbook for Executives — How to Solve the AI Dilemma
Despite these challenges, AI success is achievable. What is required is a practical, pragmatic, and disciplined approach. Below is an executive-level playbook for navigating AI-based transformation.
- Get Educated, Be Invested, and Provide Leadership: Executives must go beyond surface-level familiarity with AI. In addition to personal learning, organizations should work with trusted partners who are willing to be honest about what AI can and cannot do. Many vendors stretch the truth. Executives should tread cautiously.
AI also requires visible leadership. Leaders must lean into the technology and be part of the solution.
- Identify Use Cases Aligned to Strategic Objectives and ROI: At this stage, no commitments are made. Executives and leadership teams simply begin “kicking the tires” on potential AI use cases — focusing on those that align to strategic objectives and could produce ROI.
- Realistically Assess AI Readiness: Once promising use cases are identified, organizations must assess readiness across multiple dimensions:
- Data
- People
- Technology
- Training
- Change management
This step is about honesty, not optimism.
- Quantify the ROI: For the most promising use cases, rigorously quantify ROI. Include all costs, including technology, data engineering, governance, training, and ongoing support. Be realistic about benefits.
- Start Small, Tread Cautiously, and Keep Humans in the Loop: When it comes time to select AI use cases for implementation, start small and get some wins. Do not attempt your most ambitious project first. At the same time, avoid turning AI loose without oversight. Some of the most visible AI failures have come from organizations deploying AI without sufficient human intervention. For now, keeping humans involved is both prudent and necessary.
- Create Policies for Personal AI Usage: Employees are often unsure what they can and cannot do with AI. Clear policies reduce risk while encouraging responsible experimentation and productivity gains.
- Provide AI Training and Guidance: Broad-based AI success requires broad-based understanding. If only a handful of people understand AI, scale will remain elusive.
- Focus on Productionalizing AI Solutions:
Force the organization to think beyond the POC:- How does this fit into daily workflows?
- How are data pipelines built and maintained?
- How are models updated?
- How do business processes change?
- Don’t Just “AI-ify” Existing Processes — Transform Them: When computers first became widespread, companies “computerized” paper processes rather than rethinking them. Do not repeat that mistake with AI. Use AI to transform processes, not simply automate old ones.
- Establish a System for Continuous AI Improvement: AI success is not about one-off initiatives. It requires a systematic approach for continuously identifying use cases, evaluating ROI, staffing initiatives, implementing solutions, and measuring results. This does not need to be bureaucratic — but it must be intentional.
Closing Thoughts
The AI dilemma facing executives is real. AI is not a fad and is not going away. It is difficult to succeed with AI, many have failed, the stakes are high and doing nothing is not an option.
While this may sound dire, it is not. Despite the mystique and hype surrounding AI, it is ultimately just another tool — albeit a powerful one — that can be used to improve how businesses operate.
That is why the first step is education. Demystify AI, and organizations will be far better equipped to make informed decisions about how and where to deploy it. There is a great deal of hype in the market today, often accompanied by unrealistic timelines. The best approach is to tune out the noise and take a practical, pragmatic, and disciplined path forward.
Apps Associates is a leader in leveraging technology — including AI — to transform businesses and deliver tangible ROI. We would welcome the opportunity to speak with you about how to implement AI-based transformation at your organization.
