Root out hidden data debt to improve AI decision-making

AI learns to act before enterprises learn to govern

AI can deliver faster analysis and, over time, automate numerous day-to-day business decisions. And there will come a day when AI delivers ambient intelligence, where digital systems continuously process information and make decisions without being asked, according to Steve Prewitt, vice president of data and AI at Genpact, a professional services and technology firm.

"The idea of a human-readable data pipeline, one where the human need to take actions will be obsolete, will be here sooner than we think," Prewitt said.

But many organizations are implementing AI to assist with decision-making without the data and governance foundations necessary to ensure reliable outcomes. Grant Thornton found that 75% of boards had approved major AI investments, but 48% had not set AI governance expectations and 46% had not integrated AI risk into ongoing oversight. Those gaps amplify traditional data-driven decision-making (DDDM) risks while also introducing new ones.

"We're in a crazy place right now where most companies aren't ready for the AI transformation that was here six months ago, let alone ready for the new way of working that's here right now," Prewitt said.

However, Genpact's 2026 AI report estimated that Global 2000 companies could unlock $18 trillion in value by addressing enterprise debts related to data, processes, technology and talent. The report estimated this action would produce 8% faster annual revenue growth and 16% annual cost reduction.

How AI switches from analysis to action

Executives and workers have used documented available information to make business decisions since the advent of record-keeping. Over the past 70 years, constantly evolving technology has expanded the speed and accuracy of DDDM.

AI is the latest technology in a long chain of progress, including mainframes, spreadsheets, BI platforms and advanced analytics applications. But AI's distinguishing features can produce results that exponentially surpass those older tools.

"With AI, you're able to get a much better, more rounded picture, and get it faster, for almost any analysis you want to do," said Chris Hutchins, founder and CEO of Hutchins Data Strategy Consulting.

Yet AI is not only accelerating DDDM; it's transforming it. Computers are no longer just analyzing the data. Some organizations are handing more power to AI systems to recommend or automatically execute actions that could affect the enterprise.

"The point of data-driven decision-making has always been the decision part. The knowing isn't the point; doing is," Prewitt said. "Now we're outsourcing that doing part to AI systems."

But the speed and reach that AI brings to DDDM increases the consequences of longstanding risks and could introduce new ones while complicating accountability. An AI system can execute a decision, but the organization and the people who authorize this automation are ultimately responsible for them, experts said.

"AI is transformative because it can compress the decision cycle, but it can’t take accountability for that decision-making," added Doug Leal, vice president of data and analytics consulting at CGI.

AI turns bad inputs into big consequences

AI requires the right data, governed processes, technology and people to provide reliable insights, Prewitt said. He noted that Genpact's research found only 6% of organizations have resolved deficits in those areas.

The same weaknesses in those four areas are the same ones limiting the effectiveness of conventional DDDM technologies, including poor data quality, asking the wrong questions, legacy IT architecture that limits access to data and cognitive bias. Those lead to distorted recommendations when using BI -- and the wrong actions when using agentic AI.

"AI makes those challenges bigger, and if you put an AI agent that's making decisions on top of it, then the impact is even bigger," Leal said. "AI amplifies the range of the blast radius."

The size of that blast radius depends on a variety of factors, such as the number of connected systems, how reversable the actions are and how quickly the controls can find a deviation.

AI-driven decision-making can introduce or increases risks, such as hallucinations, biases and model drift, Hutchins said. Executives might also run into problems if they give AI the wrong objective. Hutchins cited a utility that used AI to place critical assets in an area prone to severe weather. Because the AI system focused on route optimization rather than reliable access, it failed to account for possible road closures due to bad weather which could make these assets inaccessible during a storm.

Trust begins when the audit trail starts

Organizations that succeed in the AI era won't be the ones with the most agents deployed, but the ones that know what their agents do and why, Leal said.

Insightsoftware's 2026 AI survey found that only 51% of the data and analytics leaders trust AI-generated insights. They listed the following as the biggest barriers to gaining confidence in those outputs:

• Security and governance concerns.
• Data quality concerns.
• AI hallucinations and fabricated information.
• Little or no audit trails for AI-generated answers.

Leal said the remedy requires executives to improve data quality, establish necessary governance and guardrails, gain visibility and observability, create processes to validate AI outputs, set clear objectives and educate workers on AI use.

"The challenges that we've been facing for years to build data-driven organizations still exist. And if you don't address those foundational challenges, AI will just amplify them and make them more visible," Leal added.

Mary K. Pratt is an award-winning freelance journalist specializing in enterprise IT, cybersecurity strategy and data management.