As enterprises strive to scale AI pilots, many are still finding it challenging to demonstrate tangible business success for their efforts.
According to Foundry’s latest AI Priorities Study, more than 60% of organizations are still in the research or pilot phases of their AI journey.1 Furthermore, analysts IDC say they are seeing AI pilot success rates of just 50-70%.2
AI progress differs across sectors. In financial services, machine learning technologies are already established in areas like risk analysis or anti-fraud, yet generative AI (genAI) projects have yet to gain much traction. It’s a similar story in retail, where AI capabilities are being integrated into major retail platforms, but genAI use is developing at a slower pace.
What’s holding genAI back? According to Laudon Williams, Senior Director, Technology Solutions, EPAM, and Jon Kadis, Managing Principal, Technology Consulting, EPAM, the answer comes down to a lack of in-house expertise as well as challenges around getting the right tooling and infrastructure into place.
Too many organizations in both industries are stacked with bloat and struggling with complexity. They’ve invested in keeping their legacy systems alive at the expense of innovation. As Kadis puts it, “they’re able to run their business, but they’re not able to quickly evolve for new environments and situations.”
Expertise needed
Foundry research confirms this diagnosis. More than a third of enterprises cite a lack of in-house expertise as their principal AI challenge (37%) while a similar number (31%) are put off by the high costs of upgrading their current tech stack.3
The answer isn’t to bolt AI capabilities onto existing architectures or lift and shift workloads to cloud infrastructure as they are.
As Williams notes: “The catch 22 of cloud for a very long time has been this belief that by the simple action of moving my infrastructure from my datacenter to another place, it somehow inherently changes its nature. It doesn’t.”
Old applications and fragmented data still cause issues in the cloud. Instead, enterprises need to modernize.
“It’s hard, it’s complex and most people don’t know how to do it” says Williams. “Many don’t even have the code for some of these old applications that can’t be modernized.”
Develop a vision
Businesses must consider bringing in partners with the expertise to help enterprises modernize – and help them develop a vision to transform infrastructure that can be a springboard for AI success.
What’s more, AI itself is coming to the rescue. Models such as Anthropic’s Claude Sonnet and Claude 4.0 Opus combine a grounding in cloud native technologies with advanced coding capabilities. Potentially, new versions could be employed to make modernization easier, faster and cheaper.
This creates potential for systems that don’t merely replicate the work of legacy systems but reimagine it.
As Williams puts it: “If you’re in the retail space or in the financial space – or frankly, in almost any industry – this is a reenvisioning of how you think about technology as an enabler.”
Modernization and cloud native infrastructure offer what Williams calls “a hyper consistent platform,” where everything runs on the same stack, not only reducing cost and complexity, but making all the data readily accessible for AI.
Kadis adds: “Moving to the cloud-native tooling gets the customer out of the business of managing and patching VMs, and all of that effort to operate the infrastructure reduces significantly or just disappears.’”
Rather than maintain and – at best – react, going cloud native enables CIOs to harness genAI and innovate.
1 AI Priorities Study 2025, Foundry, February 2025
2 IDC Blog, Overcoming GenAI Pilotitis and Acute POC Syndrome, July 2024
3 AI Priorities Study 2025, Foundry, February 2025