Vast mountains of often unstructured data can become easier to surmount with AI’s assistance. AI helps CIOs move faster by processing large volumes of data and accelerating insight into how business actually operates, said Michael Wetzel, CIO at risk and compliance firm Netwrix. “The value comes when AI is applied to real business problems, not technology for its own sake.”
One problem AI can help tackle is observability across the organization. When AI is grounded in identity and data security, leaders can see how people, systems, and data interact, Wetzel noted. “Since identity defines how employees show up, collaborate, and contribute, starting with identity allows CIOs to better understand risk, access, and behavior across the organization.”
AI at work for the enterprise
AI has become very effective at identifying interesting patterns and anomalies in data, something that was difficult to do earlier without building specialized analytics tools, said Vaibhav Kumar Bajpai, a group engineering manager at Microsoft Core AI. “Today, leaders can ask AI to predict outcomes based on existing data patterns and get answers in minutes instead of days.”
It’s important to remember that AI should never be used to replace people — it should augment them, Wetzel said. “Humans bring context, intuition, and judgment, but they can’t analyze data at the same scale or speed as AI,” he explained. “AI can continuously process information and surface patterns, allowing users to focus on higher-order thinking, decision making, and problem-solving.”
Humans are brilliant, but they get tired, observed Eric Poff, CTO at AI technology developer Vurvey Labs. “AI is tireless and consistent,” he said. “AI can help eliminate the volume-versus-quality trade-off by operating at scale while maintaining quality.” This also means that AI insights are generally more consistent. Humans can have bad days, leading to inconsistent results. “AI, however, is capable of providing consistent, deterministic outputs when constructed properly,” Poff said.
Added value to business
Where AI adds clear value is in its ability to synthesize large volumes of data into concise, well-structured insights quickly, Bajpai said. “It can organize detailed data points, highlight key drivers, and present explanations in a consistent format that would take humans much longer to produce.” In many cases, AI-based insights are grounded in recognized data sources and references, which helps establish credibility. “Used correctly, AI complements human expertise by accelerating analysis rather than replacing it,” he said.
The biggest mistake CIOs make is using AI without establishing a clear business purpose or understanding how it will impact people, Wetzel said. “Some organizations focus too much on security controls or technology while losing sight of the employee experience,” he noted. Other enterprises may move too fast, leading to the creation of shadow AI tools lacking appropriate visibility or governance. “The right approach is to start with how people work, then layer in security and AI thoughtfully,” Wetzel advised.
For the last few years, the AI narrative was dominated by unstructured data and generative tasks such as summarizing emails, writing code, or generating images, Poff said. “AI can unlock new insights by performing translations between business intent and structured data; effectively bridging the ‘what’ and the ‘why,'” he explained. Poff added that structured data tells users what happened, while unstructured data tells them why. “AI can look at both simultaneously, providing insight you simply can’t get from a static dashboard.”
Training AI matters
The usefulness of AI tools depends heavily on how well they are trained on the right data and how rigorously they are evaluated against real-world use cases, Bajpai said. “Having worked on AI evaluation for many years, I’ve seen that the real value of these tools comes from thorough testing, such as A/B testing, in production scenarios tied to specific user needs,” he said. “In practice, an AI tool using less powerful models — but tested extensively for a defined use case — can outperform a more powerful large language model that has not been evaluated with the same rigor.”
