Walk into any boardroom and you’ll likely hear “we need an effective AI strategy.” Budgets are being approved, pilots are spinning up everywhere and leaders are talking about AI transformation. But when I talk to IT practitioners, the technical professionals who implement AI responsibly and securely, the story is very different. These people are frequently challenged with stalled projects and ballooning costs, and the promised business impact of AI in many cases simply doesn’t materialize.
This dichotomy raises an interesting dilemma. Enterprise AI often fails not because the models are insufficient. It fails because of misalignment amongst people: the combination of our habits, incentives and aspirational thinking.
I think of these challenges as the seven deadly sins of AI. Underneath all of them lies a core myth: if we just add more data, more tools, more spend, then success will follow. In reality, however, AI success comes from precision, not excess — engineered systems, deliberate data selection, and measurable outcomes.
Let’s explore each of these “sins” – and how to prevent their allure – in more detail.
Sin #1 Gluttony: Collecting All the Data and Using Very Little of It
For years, the mantra has been “more data equals more intelligence.” The result is data gluttony in which we ingest and hoard everything, just-in-case.
In practice, most enterprises use only a fraction of what they store, while paying to ingest, replicate, secure, and govern data that never meaningfully affects a model, a decision or a business outcome. There’s a widespread assumption linking intelligence to sheer volume, and entire storage systems are built around this idea. The result is a kind of digital pantry stocked with items that never become a meal.
Years of building large-scale data platforms have shown a clear pattern. The operational information that exposes process delays and workflow bottlenecks often sits underused. Yet those details often contain the strongest signals to help organizations act faster or spot inefficiencies.
At the same time, many companies pour money and time into retaining telemetry and logs that do little beyond adding weight to infrastructure. The waste grows quietly in the background until costs begin to overshadow value.
To cut the fat, instead of defaulting to “collect everything,” require every dataset to answer two questions before ingest:
1) What decision, model, or workflow will this data improve?
2) How will we measure whether the data added value?
If you can’t answer those questions, thus identifying high-signal operational data tied to a specific workflow or outcome, then don’t ingest it. The end goal is to map operational workflows and identify the data needed to optimize those workflows.
Sin #2 Envy: Copying AI Leaders Without Understanding Their Reality
Currently, we’re witnessing many organizations attempt to mirror the AI strategies of their strongest competitors. Some have implemented entire architectures that looked impressive on paper, or even on stage at conferences, yet in practice, they created friction within their own walls.
Without understanding the underlying ‘why’ behind the strategy, these initiatives often overwhelm budgets, and teams are not set up properly to operate systems at scale. The lesson that emerges from these situations shows the importance of choosing areas where your organization intends to build out unique strengths and operational advantages that will become your competitive advantage.
An enterprise only gains leverage when it focuses on the capabilities it can perform better than anyone else.
Sin #3 Sloth: Waiting to Modernize and Losing Control of Costs
AI requires infrastructure that responds quickly and scales predictably. Many leaders continue using platforms that appear to function, even as new workloads begin to overwhelm them. One of the earliest signs is a rising tide of analytics activity that feels impossible to tame. AI workloads consume enormous amounts of compute (training LLMs consumes up to seven times the energy of conventional cloud computing tasks) and significantly increase the query load of earlier tools.
I have heard stories of companies spending millions of dollars to replace tasks that saved only a fraction of that amount. The pattern resembles letting a brand new system generate a monthly bill that no human worker could possibly justify. Organizations fall behind slowly at first, as query delays increase and governance exceptions multiply. By the time the true cost becomes visible, any modernization effort feels intimidating.
A practical starting point begins with gathering clear performance metrics and identifying the largest inefficiencies in data movement and query patterns before committing to wholesales replacement. From there, consolidation around a smaller and more tunable environment creates the foundation for sustainable deployment.
Sin #4 Pride: Assuming Your Current Stack Can Absorb AI
During the early rise of the internet, I worked in enterprise IT and saw firsthand how deeply a single innovation can reshape the entire technology environment. Email systems could not communicate with one another. Address directories did not align. Entire workflows required reinvention. The arrival of the internet did not allow for a simple add-on. It pushed every organization to revisit the fundamentals.
AI introduces the same level of disruption. Whenever leaders insist that their existing systems only need minor adjustments, the distance between expectation and reality grows larger. A useful point of clarity often comes from evaluating whether the current environment can keep pace with competitors who fully embrace AI-driven efficiency. Once that conversation begins, the need for broader transformation becomes obvious.
Sin #5 Greed: Accumulating Too Many Tools Without Improving Efficiency
The rise of AI tools mirrors the early era of office software. I once led a computing standards group at Martin Marietta (Lockheed Martin) and saw employees rely on a mix of spreadsheets, databases, and word processors. Everyone had a personal favorite, and every team needed support for all of them. Eventually, the company had to standardize. The decision saved money, simplified training, and dramatically reduced support burden.
The same pattern is emerging in AI. Many organizations now support dozens of overlapping tools that serve individual needs but vary in collective value. Tool sprawl forces IT teams to maintain a patchwork system with uneven security and unclear integration paths. Consolidation improves reliability and reduces cost, and it also strengthens the organization’s ability to operationalize AI consistently.
Sin #6 Lust: Chasing Shiny Demos
New AI demos often deliver captivating moments. The images and interactions create excitement and appear transformative, yet they rarely reveal underlying operational complexity. The quickest way to evaluate a demo is by examining the business outcome it intends to influence and quantify measurable improvements it can deliver. If a team cannot articulate the expected result in practical terms, the demo remains performative rather than a step toward production value.
Many leaders feel pressure to adopt these technologies to signal innovation, but steady, thoughtful evaluation provides a stronger foundation. Each new tool or capability must fit within the larger environment that handles security, governance, and cost control. Without that alignment, even the most impressive demonstrations drift into the category of novelty.
Sin #7 Wrath: Blaming People When Systems Fail
AI systems can magnify organizational gaps. When something goes wrong, blame often lands on the person closest to the output. Analysts, accountants, and operators become the target even when they never had the expertise or authority to manage the underlying risk.
In reality, the real issue usually lies in system design and overall accountability. AI requires clear ownership that includes IT, security and governance as part of the system design from the start. When these elements work together, teams avoid surprises and respond quickly to errors. When they do not, individuals shoulder responsibility for issues that may span far beyond their reach or expertise.
A healthy culture treats failures as opportunities to improve architecture and processes. Teams begin to understand AI as an engineered system rather than a magical assistant, and this shift produces better outcomes and steadier progress with AI across the organization.
The Way Forward
Every major shift in computing has followed a similar arc. A portion of the value comes from improving what already exists, and a larger portion comes from discovering entirely new use cases. AI accelerates both paths.
Organizations that thrive approach AI with precise engineering, steady efficiency, and a commitment to strong architecture that allows agents, analytics, and governance to work as a unified system. They choose tools deliberately. They modernize before systems reach a breaking point. They build environments that allow AI to operate safely and predictably.
Ambition continues to drive interest in AI, and discipline will determine which companies turn that ambition into a lasting advantage.
