The hidden costs CIOs face to make data AI-ready

One of the new challenges CIOs and CTOs must now tackle is proving that their organization’s data is ready to fuel an ever-growing number of AI initiatives. Hurdles to such efforts include significant differences between AI-ready data requirements and traditional data management, said Donie Lochan, CIO at technology services provider Ahead Systems.

For example, Lochan said, most governance frameworks weren’t designed for what AI does in production. “They were built for approvals before deployment,” he explained. Yet with AI, governance must continue after deployment. 

“In traditional governance, when a near miss happens, the process is to convene a review committee,” Lochan said. By the time they meet, however, their AI could have already made thousands of additional decisions.

A better approach, he argued, is to treat every incident as a signal to tighten the architecture itself, update the guardrails, adjust the escalation thresholds, and narrow the decision rights. “Then governance stops being a process and becomes a living system,” he explained.

Related:InformationWeek Podcast: Dealing with strategic tech rejections

AI systems should be governed in the same way we govern production software: with observability, audit trails, escalation procedures, and clearly defined ownership, said Filip Popovic, CTO at sales intelligence and data collection platform ZoomInfo. 

“The reality is that near misses and failures are inevitable, so the goal is not to eliminate them entirely, but to ensure they’re detected quickly,” he advised. Containment of such issues can then be used to improve the system. “Additionally, every AI-driven recommendation or action should be traceable back to the data, signals, and reasoning that produced it.”

AI-driven decisions vs real-world complexities

To address post-deployment AI governance issues, CIOs and CTOs need clear lineage and objective trust signals so they can triage whether the issue stems from the model, workflow, permissions, deprecated data, poor business definitions, or broken data pipelines, said Sam Pierson, CTO with data quality and analytics solutions provider Qlik.

Trust breaks down when people experience inconsistent outcomes or feel that AI-driven decisions aren’t accounting for real-world complexities, said Parijat Jauhari, CTO at SaaS advisory firm LRN. “The most effective organizations recognize that AI governance isn’t just a matter of adoption, it’s also a leadership and culture challenge,” he explained. “When legal, compliance, HR, and technology teams work together, they can evaluate decisions through multiple lenses.”

Related:Oracle’s AI-based layoffs may not be over

If, for example, an AI system generates outreach based on customer intent data, and the legal department raises concerns about compliance or privacy requirements, there needs to be a governance model in which engineering, legal, security, and business stakeholders jointly define acceptable behavior, Popovic said. “Similarly, if the sales department wants to automate actions that the customer success department believes may negatively impact existing relationships, there must be a clear escalation framework.” 

He added that rather than creating new problems, AI often exposes organizational alignment problems that already existed but were previously hidden by slower manual processes.

The slow rhythm of governance frameworks

AI cannot be retrofitted into governance models built for quarterly review cycles, Pierson warned. 

“Traditional governance frameworks were designed for a slower business rhythm, where decisions could be reviewed in batches, exceptions could be analyzed after the fact, and accountability could be documented retrospectively,” he explained. 

Pierson noted that AI doesn’t operate at such a cadence: “It’s continuous, dynamic, and increasingly embedded in live business decisions.” By the time an issue shows up in a quarterly report, the organization will have likely already acted on flawed data, repeated a bad recommendation, or expanded a small governance gap into a larger operational problem.

Related:How to submit an IT leadership column to InformationWeek

The governance perimeter must therefore expand to cover data we never previously curated, said Olga Kupriyanova, director of AI and data engineering at technology research advisory firm ISG. “That doesn’t mean boiling the ocean; it means deciding deliberately which dark data becomes a trusted source, which gets cleaned up, and which gets walled off from AI entirely.”

All of this is brought under the same definitions and boundaries applied to the core.

Handling skepticism with discipline

Transparency is key to combating internal resistance, Lochan said. “If you try to minimize or hide the inconsistency, you only deepen skepticism,” he warned. “What rebuilds trust is showing people exactly what went wrong, what the system did, and what you’ve changed so it won’t happen the same way again.”

The companies pulling ahead aren’t only spending on AI, they’re allocating capital differently, moving fast on what works, and exiting from what doesn’t before it becomes a liability, Lochan said. “Similarly, the CIOs and CTOs who treat governance as a continuous design discipline, rather than a compliance exercise, will be the ones who will ultimately win.”

The organizations that see the greatest success treat AI as a business transformation initiative rather than as a software implementation project, Popovic said. “They invest equally in data quality, governance, organizational alignment, and change management,” he stated. Long-term winners will not necessarily be companies with the most advanced models. “They will be the companies that create the highest level of trust between humans, systems, and AI-driven decision-making.”

Failure as a learning lesson

Leaders should deal with AI failure in the same way they would want a technology analyst to deal with failure, Kupriyanova said. 

“When a good analyst makes a mistake — whether they catch it themself or someone points it out to them — they take it, learn from it, and get better,” she explained. A bad leader, meanwhile, takes offense. “AI has to behave like a good analyst, except at scale, and it has to really learn.”