I am an operations consultant first. I work with businesses to assess, strengthen, and streamline their internal systems before any technology enters the conversation. AI is not my starting point. It is a tool I consider only after I determine whether a business is positioned to use it effectively, whether implementation is necessary, and whether the cost of adoption will yield a meaningful return. When those conditions are not met, my recommendation is to wait.
That position is not popular in a market saturated with AI hype. It is, however, supported by a growing body of research that should give every business leader pause.
What the Research Actually Shows
MIT's NANDA initiative analyzed 300 public AI deployments, surveyed 350 employees, and conducted 150 interviews with organizational leaders. Their finding: 95% of generative AI pilots at companies are failing to deliver measurable impact on profit and loss. The core issue, according to the report, is not the quality of the AI models. It is flawed enterprise integration. AI tools built for individual flexibility stall in enterprise environments because they cannot learn from or adapt to existing workflows.
A separate PwC survey found that only 12.5% of CEOs reported AI delivered both cost savings and revenue growth simultaneously. NTT DATA found that between 70 and 85% of generative AI deployment efforts are failing to meet their desired ROI. An analysis of AI adoption patterns found that 42% of companies abandoned most AI initiatives by mid-2025, up from just 17% the prior year. And 88% of AI pilots never make it to production.
These numbers reflect a systemic pattern of organizations moving faster than their infrastructure can support.
The root cause consistently cited by enterprise data leaders?
Not model accuracy, computing costs, or talent shortages. 73% identified data quality and completeness as the primary barrier to AI success. Companies with strong data integration achieve 10.3x ROI. Those with poor data connectivity see only 3.7x. The gap is not about the AI. It is about what the AI has to work with.
The VUCA Amplification Effect
Organizations operating in Volatile, Uncertain, Complex, and Ambiguous (VUCA) environments face an amplified version of this risk. In VUCA conditions, the environment shifts unpredictably. AI systems trained on historical data patterns can rapidly become misaligned with current operational realities.
In these environments, AI without a strong operational infrastructure creates a dangerous feedback loop:
- Unstable inputs from VUCA conditions feed into AI systems built on fragile data pipelines.
- Degraded AI outputs are used to make high-stakes operational decisions.
- Poor decisions worsen the operational environment, increasing VUCA intensity.
- Infrastructure gaps prevent the identification of the AI as the source of the problem.
For organizations serving clients in VUCA environments — including healthcare, corrections, crisis response, and emergency operations — this loop can translate directly into harm to people.
My position on AI implementation in VUCA environments has been, and remains, consistent: no AI deployment should proceed without a thorough operational infrastructure assessment, clearly defined governance structures, a human in the loop at every critical decision point, and a plan for what happens when the system is wrong. In high-stakes sectors, that bar must be higher, not lower.
AI Readiness Is Not About Enthusiasm. It Is About Infrastructure.
Readiness for AI deployment requires an honest assessment across five dimensions: purpose, people, process, platform, and performance. Most organizations rushing to implement AI have not worked through all five. They have the enthusiasm. They may even have the budget. What they often lack is the operational foundation that gives AI something solid to work with.
Before any AI implementation, these questions must have clear answers:
- Is there a defined business application this AI will serve, or is implementation being pursued because it feels urgent?
- Does the organization have the in-house technical expertise to manage, audit, and correct the system?
- Is the data infrastructure clean, governed, and reliable enough to produce trustworthy outputs?
- What are the compliance and cybersecurity implications, and are there qualified professionals overseeing them?
- What is the human oversight protocol when the AI produces an error?
The last question matters more than most organizations acknowledge. AI systems make mistakes. They still hallucinate and misinterpret context. They perform well on training data and poorly on edge cases. The presence of a human in the loop is not a concession that the technology is inadequate. It is a recognition that no automated system should operate without accountability.
Where AI Makes Sense, and Where It Does Not
There is a meaningful difference between deploying a chatbot on a business website to answer frequently asked questions about services and embedding AI deeply into HR processes, financial decision-making, or sensitive client operations.
The former is appropriate for many businesses. A well-configured conversational AI that handles routine inquiries, captures leads, and reduces response time is a low-risk, high-utility implementation. The human in the loop is never far away, the stakes of an error are low, and the operational lift is manageable.
The latter is where I see the most risk. HR and finance are areas where AI should not have access to personal information or financial data until the level of risk has been fully assessed and mitigated. These are high-stakes domains where errors are not just operational problems. They are legal, ethical, and human problems. The rush to automate performance reviews, compensation modeling, or financial forecasting with AI tools that have not been properly integrated, audited, or secured is a liability that most organizations are not equipped to manage.
The Role of Software Engineers and Cybersecurity Professionals
One of the most overlooked elements in the AI implementation conversation is the human talent required to make it work safely. AI tools do not manage themselves. They require software engineers who understand how the systems integrate with existing infrastructure and can identify when something is failing. They require cybersecurity professionals who can assess exposure, protect sensitive data, and respond when vulnerabilities emerge.
Real-world examples bear this out:
- McDonald's shut down its AI voice-ordering pilot with IBM after the system repeatedly misunderstood orders, adding items customers never requested.
- Australia's Commonwealth Bank eliminated 45 customer service roles in anticipation of AI handling the workload, only to rehire those employees after the AI chatbot failed to reduce call volumes.
- U.S. Immigration and Customs Enforcement discovered its AI resume-screening tool had fast-tracked unqualified applicants into law-enforcement training programs simply because their resumes contained certain keywords.
In each case, the technology failed not because AI is inherently unreliable, but because the implementation outpaced the infrastructure and oversight required to support it.
A Note on Cost
There is also a financial reality that is not discussed enough. The assumption that AI implementation will reduce costs is not a given. Third-party API usage, licensing fees, and the ongoing cost of maintaining AI integrations are expenses that compound over time. Organizations that adopt AI tools without a clear understanding of their total cost of ownership frequently discover that the ROI they projected never materializes.
This is part of why I assess not only whether AI can serve a business need, but whether the economics make sense. A tool that costs more to maintain than it saves is not an efficiency gain. It is a drain.
What I Actually Help Businesses Do
My work is not to position AI as unnecessary. It is to position it correctly. I help businesses understand which AI tools exist and where they could genuinely add value. I circulate that information through my newsletter as part of keeping clients informed about the landscape. When a business is ready for a more specific implementation conversation, we work through the assessments together.
The goal is never to be the last firm to adopt AI. It is to be the firm that adopts it in a way that holds.
AI implemented without operational readiness fails expensively, publicly, and sometimes in ways that affect the people an organization is supposed to serve. The research supports a measured, infrastructure-first approach. So does every case study of an AI rollout that went wrong.
If your business is curious about where AI could fit, that conversation starts with an honest look at what your systems can currently support. That is where I start, every time.
Frequently Asked Questions
Why are most generative AI pilots failing?
According to MIT's NANDA initiative, 95% of generative AI pilots are failing to deliver measurable impact on profit and loss. The core issue is not model quality — it is flawed enterprise integration. AI tools designed for individual flexibility cannot learn from or adapt to existing enterprise workflows. The failure is an infrastructure problem, not a technology problem. Organizations are deploying AI faster than their operational systems can support it.
What is an AI readiness assessment and what does it cover?
An AI readiness assessment evaluates whether a business has the operational foundation required to deploy AI effectively. It covers five dimensions: purpose (is there a clearly defined business application?), people (does the organization have the technical expertise to manage and audit the system?), process (are workflows clean and documented?), platform (is the data infrastructure reliable and governed?), and performance (what are the compliance, cybersecurity, and human oversight protocols?). Without clear answers across all five, AI deployment carries significant and often invisible risk.
Why is data quality the biggest barrier to AI success?
73% of enterprise data leaders identify data quality and completeness as the primary barrier to AI success. Companies with strong data integration achieve 10.3x ROI from AI, while those with poor data connectivity see only 3.7x. The gap is not about the AI models — it is about what the AI has to work with. AI systems trained on incomplete, inconsistent, or ungoverned data produce outputs that cannot be trusted, no matter how sophisticated the model.
What is the VUCA amplification effect on AI?
In Volatile, Uncertain, Complex, and Ambiguous (VUCA) environments, AI without operational infrastructure creates a dangerous feedback loop: unstable inputs feed into fragile data pipelines, producing degraded outputs that inform high-stakes decisions, which worsen the operational environment further. For organizations in healthcare, corrections, crisis response, or emergency operations, this loop can translate directly into harm to people. No AI deployment in a VUCA environment should proceed without a thorough infrastructure assessment, defined governance, and a human in the loop at every critical decision point.
When does AI make sense for a small business?
AI makes sense for a small business when the implementation is low-stakes, clearly defined, and does not require access to sensitive personal or financial data. A well-configured conversational AI handling routine inquiries, capturing leads, and reducing response time is a sound implementation for most businesses. What requires more caution is embedding AI into HR processes, financial decision-making, or sensitive client operations — domains where errors are not just operational problems but legal, ethical, and human ones. The question is not whether AI can be useful, but whether the infrastructure and oversight protocols are in place to support it.
Do you need a software engineer or cybersecurity professional for AI implementation?
Yes. AI tools do not manage themselves. They require software engineers who understand how systems integrate with existing infrastructure and can identify when something is failing. They require cybersecurity professionals who can assess exposure, protect sensitive data, and respond when vulnerabilities emerge. McDonald's, Commonwealth Bank, and ICE all share a common failure pattern: implementation that outpaced the infrastructure and oversight required to support it.
Sources
Ready to Assess Your AI Readiness?
Before your business invests in AI tools, it needs an honest look at what your current systems can support. That assessment is where every engagement begins. If you want to understand where AI could genuinely add value — and where it would only add risk — let's talk.