Deploying enterprise-grade AI across a large organization means navigating complex data estates, guaranteeing compute availability and ensuring systems can talk to each other.
But for many enterprises, technical considerations are among the least of their problems when facing an AI implementation plan, Prashant Prahlad, SVP of product at Capital One Software, told CIO Dive.
Prahlad heads product management for the company’s enterprise B2B software business, which helps prepare customers’ data to scale AI, complete data transformation or carry out cloud migration. He’s often consulting customers on creating consensus within their organizations and building momentum, he said.
“They struggle with organizational rallying for the order on their agenda, which is AI everywhere,” Prahlad said.
Prahlad stressed that an organization’s data foundation will lay the groundwork for how its AI can scale — project failure rates jumped in the last few years due to poor data readiness rather than technology limitations.
How an organization addresses its AI use organizationally, how teams think about ROI and an overload of AI can hurt companies’ AI deployment plans, Prahlad said.
Many organizations have a “mandate to go AI first, but they don’t have the authority,” Prahlad said. They’re often trying to rally different groups of people within the organization without the same incentives, he said.
Employees are more likely to distrust AI than those in higher positions, and they report greater personal risk from autonomous systems because of their lack of control.
The challenging part of ROI for many organizations isn’t just attempting to measure it, but ensuring that all of the proper foundations are in place to get an accurate read, Prahlad said.
Most businesses haven’t justified their spending yet, and say that they need to get AI systems into production and provide proprietary sensitive data to the models to properly test it. It’s a tall order to accomplish while CIOs are expected to report value to the board — and while most still aren’t finding cost efficiency or productivity as much as it is helping create insights or make decisions.
Another challenge facing most organizations is a saturation of AI, Prahlad said. Most organizations are being bombarded by various vendors promising wide-reaching effects, but often existing products that can be prepped to be AI-ready are the better choice, Prahlad added.
“AI springs up in every different location, and sometimes the overexuberance of that, makes it very hard for central teams to then figure out signal through the noise,” Prahlad said.
A single use case
Often, trailing AI through a single use case can teach a lot about what an organization needs in order to set itself up to scale. Creating a “tiger team” of cross-functional specialists to identify and test data is a good start, Prahlad said.
An IT team can help by identifying one use-case that they would then work backward from. For example, a team could deploy a chatbot through one part of the organization, then analyze the data readiness, governance and security needs surrounding that use case.
That’s how a recent client handled their data assessment, Prahlad said, and realized they lacked a centralized data strategy.
“They went in and created a team that would centralize data management across seven different lines of businesses,” Prahlad said. “There are obvious problems there in trying to get the organization aligned, but having a single use case actually helped them understand where the gaps were.”
The starting point for data and AI scaling readiness is all over the map for organizations, and those with large tech teams aren’t automatically better off, Prahlad said.
“It depends on their intentionality in adopting AI and the mandate they’ve had so far,” Prahlad said. “I’ve found that institutions that you would have expected to be at the forefront of AI are not as far ahead.”
