“Companies are investing in the structure and the walls and the roof and the architecture, but they’re ignoring the foundation,” Transcend executive says.
Illustration: Anna Kim, Photos: Adobe Stock
Everyone’s crazy about AI, but what does that mean in practice?
A new survey by Transcend, shared exclusively with IT Brew, finds that, for all the hype, one-third of enterprise AI initiatives are either delayed, scaled back, or abandoned. It’s a real problem for organizations, Transcend’s CIO and CISO in residence Aimee Cardwell told IT Brew.
“Companies are investing in the structure and the walls and the roof and the architecture, but they’re ignoring the foundation,” Cardwell said. “It’s not that the AI isn’t smart or isn’t able to do the use case, it’s that your data isn’t ready for it.”
Check your work. Often this manifests itself in the details of how AI models are deployed. Researchers found that, of the 70% of organizations providing a dedicated budget for AI, 30% don’t have a specific budget line item for the data prep necessary for feeding AI models. That kind of oversight has led 81% of the 220 tech industry leader respondents to report at least one initiative hitting a bump in the road, even as 49% report that AI is widely used and deployed in their organizations.
For IT pros, these stalled initiatives mean more work, not less—the very problem AI was supposed to solve. But issues with the technology aren’t limited to such initiatives; as IT Brew has reported, AI rework is an overarching issue for tech teams.
The data privacy platform’s research puts the problem into stark relief, with engineers reportedly devoting less than a quarter (23%) of their time to building profitable AI functions while spending the rest of their workday on the nuts and bolts of infrastructure. As Cardwell explained, “The other 77% is maintaining data infrastructure, fixing data quality problems, managing and monitoring governance.”
“As an AI engineer, you expect to be working on building AI tools,” Cardwell said. “But in many cases, that’s not necessarily what you’re doing.”
On sequence. Better data management could go a long way toward solving the problem, Transcend’s research suggests, because 93% of the time during the AI life cycle, organizations face data issues related to permissions or governance. Those problems, including data sequencing, are part of the growing pains of using a new technology, Cardwell told IT Brew.
“Hopefully, a year or two from now, we’ll be over this particular challenge and companies will have figured out that the sequencing is important,” Cardwell said. “And then we’ll be releasing a different study with a different set of conclusions or or realizations.”
