Three Barriers to AI Scaling, According to Capital One Software

Capital One Software's product leader said the barriers to enterprise AI scaling lie less in technology and more in how organizations operate, how they view ROI, and AI overload. In a CIO Dive interview published on October 27, 2026, Prashant Prahlad, Senior Vice President (SVP) of Product, said that for many companies, technical considerations rank among the smallest concerns in their AI adoption plans. The enterprise B2B software business where he leads product management helps clients prepare their data for AI scaling and supports data transformation and cloud migration. He said he frequently advises clients on how to build consensus and gain momentum within their organizations.

What are the three things blocking AI scaling?

Prahlad emphasized that an organization's data foundation underpins how far AI can scale. He noted that the rising project failure rates of recent years stemmed not from technical limitations but from inadequate data readiness. Building on this, he identified three factors that undermine AI adoption plans.

• Organizational operations: Many organizations receive an 'AI-first' mandate without the authority to execute it, he said. He added that such organizations often struggle to bring together multiple groups that don't share the same incentives.
• ROI: He said the hard part about ROI isn't the attempt to measure it, but rather putting all the necessary foundations in place to read accurate figures.
• AI overload: Many vendors rush in promising broad impact, but he said it's often a better choice to prepare existing products for AI rather than adopt new ones. He said the AI hype emerging everywhere makes it extremely difficult for central teams to distinguish signal from noise.

In the same article, CIO Dive reported that employees distrust AI more than those in higher positions do, feeling greater personal risk from autonomous systems because they lack control. Most companies have yet to justify their AI spending, and the article noted that to truly test AI, companies say they need to put AI systems into actual operation and feed their sensitive data into the models. While CIOs face pressure to report value to boards, the article explained that most are still deriving more value from insight generation and decision support than from cost efficiency or productivity gains.

Where to start: one use case

Prahlad said testing AI on a single use case first can teach companies a great deal about what's needed to scale. Here's the approach he suggested.

• Form a 'tiger team' of experts from multiple departments to identify and test data. He said this is a good starting point.
• Have the IT team pick one use case and work backward from there.
• For example, deploy a single chatbot in one part of the organization, then analyze the data readiness, governance, and security requirements surrounding that case.

Prahlad described a recent client that, by examining its data this way, realized it lacked a centralized data strategy and subsequently formed a team to consolidate data management across its seven business units. He said that while aligning an organization in one direction clearly presents challenges, a single use case genuinely helped clarify where the gaps were.

He said that every organization's starting point for data and AI scaling readiness differs, and a large technical team doesn't automatically put a company ahead. That starting point depends on the intentionality behind AI adoption and the mandates received so far, he said, noting that institutions expected to be at the forefront of AI were often not as advanced as assumed.

Implications for AX in Korean Companies

If an AX initiative at a Korean company began with an executive mandate, it's worth examining what accompanied that mandate. As Prahlad noted, an AI-first mandate without accompanying authority makes it difficult to unite departments with differing incentives. A good starting point is checking whether the mandate included budget, data access, or cross-departmental coordination authority.

Organizations without large IT staffs should take note that Prahlad prioritized adoption intent and mandate clarity over technical team size. Rather than directly replicating the case of a client that consolidated seven business units, organizations with fewer business units can start by mapping out which department holds the data needed for a single use case and who manages it. If an adoption review list contains only new solutions, it makes sense to also compare the option of preparing existing systems for AI. That said, his emphasis on data foundations should be read with balance, keeping in mind that he is the product leader of a business that helps clients prepare their data.

There's just one thing to check this week. Pick one ongoing AI initiative, and write down where the data it needs resides and who holds the authority to access it.

Source: How to create a scale-ready AI mandate