From BI to AI: Operationalizing Machine Learning Harvey Nash’s CIO Voices brings together technology leaders to explore the trends, challenges, and opportunities shaping the future of enterprise technology. This month’s discussion tackled a topic that many organizations are actively navigating but few have fully mastered: the journey from traditional business intelligence to AI-driven decision-making. To unpack what that transition really looks like, we spoke with Tad Rhodes, Founder of TownVue, Sanjay Bhutiani, Founder of Dreaming Tree AI, and Craig Leland, Market Director at Harvey Nash. Their perspectives revealed a common theme throughout the discussion: while AI may dominate the headlines, successful machine learning initiatives are rarely about the models themselves. They are about data, architecture, organizational design, and perhaps most importantly, people. The conversation raised an important question: what does it actually take to operationalize machine learning at scale? The answer, it turns out, is far more complex than simply adding AI to an existing technology stack. The Shift from Reporting to Discovery For years, business intelligence has served as the foundation for enterprise decision-making. Dashboards, reports, and analytics platforms have helped leaders understand what happened, why it happened, and where performance could improve. But AI introduces a fundamentally different dynamic. According to Tad Rhodes, the distinction comes down to the difference between presenting information and uncovering insights. "BI is really about presenting data. It tells you what already happened. AI is more about interrogating it, surfacing the questions you didn't even know you should have been asking." That observation captures one of the most significant shifts organizations are experiencing today. Traditional BI assumes that humans will interpret information and decide what actions to take. AI changes that relationship by accelerating analysis, identifying patterns automatically, and uncovering opportunities that may never have surfaced through conventional reporting. Sanjay Bhutiani sees this transition through a slightly different lens. In his view, many organizations believe they are moving toward AI-driven decision-making when, in reality, they are simply layering AI tools onto existing BI environments. "The honest answer is that most organizations are not really making the transition." He argues that AI compresses the decision-making loop and fundamentally changes how organizations process information. Systems designed to support executive dashboards are not necessarily designed to support machine learning workflows. Taken together, their perspectives reveal an important truth. The move from BI to AI is not simply a technology upgrade. It represents a shift from reporting what is known to discovering what is not yet visible. Why AI Adoption Is More Human Than Technical One of the strongest themes throughout the discussion was that technology itself is rarely the biggest barrier to AI adoption. The real challenge is organizational change. For Rhodes, adoption begins with demonstrating value at an individual level. He described what happens when employees see AI solve a meaningful problem for the first time. "It's the moment an employee watches AI answer a tough question well, and their eyes of understanding open." That moment of discovery often changes the conversation entirely. Skepticism gives way to curiosity, and curiosity becomes experimentation. Over time, those small moments of adoption spread across teams and departments. Bhutiani agrees that culture ultimately determines success, but he believes the challenge runs deeper than individual enthusiasm. Organizations must rethink how they structure teams, measure performance, and reward innovation. He argues that many enterprises still evaluate technology organizations based on traditional metrics such as uptime, ticket resolution, and project delivery. Yet none of those measurements capture the value AI is intended to create. Instead, organizations need to focus on outcomes such as decisions accelerated, capacity unlocked, and time recaptured. Perhaps most importantly, employees must be rewarded for changing how work gets done. As Bhutiani noted, people are unlikely to redesign their workflows around AI if their performance reviews continue to reward the old way of working. The technology may be new, but the challenge is a familiar one. Organizations that successfully operationalize AI are often the ones that successfully operationalize change. The Foundation Beneath Every Successful AI Initiative If there was one area where contributors aligned most strongly, it was on the importance of data quality, governance, and architecture. In fact, both Rhodes and Bhutiani argued that these elements ultimately determine the ceiling of what AI can achieve. Rhodes offered a pragmatic perspective, noting that organizations often become trapped between two extremes. On one side is the pursuit of perfectly clean data. On the other is the reality of messy, incomplete, and inconsistent information. "Perfectly clean data is a fallacy, but you obviously can't run on a swamp either." His point is an important one. Successful organizations do not wait until their data is perfect before moving forward. Instead, they focus on creating data that is trustworthy enough to support meaningful decisions. Bhutiani reinforced this idea from a governance standpoint. "An organization with mediocre data running sophisticated models will get mediocre outputs, confidently delivered." It is a warning many leaders need to hear. AI systems can make weak data appear convincing. The sophistication of the model does not compensate for weaknesses in the underlying information. Governance plays an equally important role. Without clear ownership, auditability, and accountability, machine learning initiatives can quickly create operational and compliance risks that become difficult to manage later. The organizations succeeding with AI at scale are not treating governance as a final checkpoint. They are treating it as an integral part of the process from the very beginning. Data Architecture: The Hidden Constraint on AI While discussions about AI often focus on models, copilots, and automation, our contributors repeatedly returned to a less glamorous topic: data architecture. Both Rhodes and Bhutiani described architecture as the hidden factor that determines whether AI initiatives accelerate or stall. Rhodes referred to it as the "silent ceiling." Organizations often discover this limitation only after investing heavily in AI tools. Data exists across multiple systems, records cannot be connected, and critical information remains trapped in silos. A customer in a CRM system may appear differently in billing, support, and operational platforms. Without a consistent way to connect those identities, AI lacks the context necessary to generate meaningful insights. Bhutiani echoed this concern. "I have watched organizations invest seven or eight figures in AI tooling on top of data architecture that fundamentally cannot support what they are asking it to do." The result is often predictable. The technology functions correctly, but the outputs remain unreliable because the underlying architecture cannot deliver complete, timely, and connected information. Three architectural principles emerged repeatedly throughout the discussion: Unified rather than siloed data. Real-time accessibility rather than delayed batch processing. Structures that allow AI systems to query and reason across information seamlessly. The lesson is clear. AI does not reward organizations that simply possess large amounts of data. It rewards organizations whose data can actually work together. Centralizing AI Without Slowing Innovation As AI adoption expands, many leaders face an important question: should AI capabilities be centralized or distributed across the business? Interestingly, both contributors challenged the framing of the question itself. For Rhodes, the goal should not be determining who owns AI. The goal should be enabling teams to solve problems. "Ask, collect, structure, expose." His approach focuses on understanding what teams need, building the right foundation, and making information accessible without creating unnecessary friction. He also offered an important observation about so-called shadow AI. Employees often turn to AI tools independently because they are trying to solve real business problems. Rather than viewing this as a threat, organizations should view it as a signal that unmet needs exist. Bhutiani approached the issue from an enterprise governance perspective. His recommendation is straightforward: centralize governance, infrastructure, and shared capabilities. Distribute application and innovation. "The organizations that get this wrong centralize everything and slow themselves down, or distribute everything and accumulate hidden risk." That balance may ultimately become one of the defining operating models of the AI era. The most effective organizations will likely maintain centralized standards while empowering individual teams to apply AI in ways that create domain-specific value. Building Teams for the AI Era The discussion concluded with one of the most pressing questions facing technology leaders today: what does great AI talent actually look like? The answer depends largely on who you ask. For Rhodes, curiosity may be the most important hiring characteristic of all. While technical fundamentals remain important, he believes AI is dramatically accelerating how quickly people can acquire new skills. As a result, individuals who are adaptable, curious, and eager to learn often outperform those relying solely on existing expertise. "A curious person with the fundamentals can pick up domain context fast and let AI handle the rest." Bhutiani takes a different but equally compelling view. He places greater emphasis on domain expertise, arguing that technical skills can often be developed faster than deep business understanding. An engineer who lacks business context may build technically impressive solutions that fail to solve meaningful problems. Conversely, a domain expert who develops technical capabilities can often create solutions that directly impact business outcomes. The most successful teams, he argues, combine senior technical architects with domain experts who are capable of growing into hybrid roles over time. Craig Leland added a practical hiring perspective that many organizations would benefit from adopting. Rather than starting with job descriptions, he encourages organizations to begin by defining outcomes. "The most common advice I provide to clients is first start by defining the outcomes sought after and not just roles or responsibilities." He also highlighted a growing challenge in the market. Organizations unfamiliar with AI hiring often add excessive interview stages in an attempt to reduce hiring risk. Unfortunately, this frequently drives away the very candidates they hope to attract. A clear evaluation process, well-defined objectives, and disciplined decision-making often produce better outcomes than lengthy hiring cycles. The Real Work of Operationalizing AI After listening to Rhodes, Bhutiani, and Leland, one conclusion stands out above all others. Operationalizing machine learning is not primarily a technology challenge. The organizations creating meaningful value from AI are investing just as heavily in architecture, governance, operating models, talent, and culture as they are in algorithms and tools. Machine learning succeeds when data is trusted. It succeeds when systems are connected. It succeeds when employees understand how to use it. And it succeeds when organizations align incentives, processes, and people around new ways of working. The transition from BI to AI is not about replacing dashboards with models. It is about building an environment where intelligence can move from observation to action. For many organizations, that journey is only beginning. A huge thank you to Tad Rhodes, Sanjay Bhutiani, and Craig Leland for sharing their perspectives. Their insights continue to make CIO Voices a valuable forum for technology leaders navigating the future of enterprise innovation.
