Enterprise AI’s biggest graveyard isn’t filled with failed models. It’s littered with perfect, multi-year plans that never launched. Haunted by data security horror stories and the ghosts of past data projects, companies chase flawless strategies for so long that they never launch anything at all.
Mladen Milanovic sees this as a fatal miscalculation. As VP of Automation at leading technology services and solutions provider Presidio, he argues that the only winning move is to stop waiting and start building.
Haunted by history: "Any time we bring up AI, the pushback is, 'Our data lake isn’t ready,'" says Milanovic. "Most customers are kicking the can down the road, stuck in a multi-year journey to unify everything—which, more often than not, doesn’t succeed." The biggest obstacle to embracing AI, Milanovic says, is hesitation driven by past failures and fresh concerns about data security.
The low-hanging fruit: His counter-intuitive solution is to sidestep the grand data project entirely. Instead of trying to solve for everything at once, he advises narrowing the scope to a small, manageable, and secure set of information. "We don't need to expose the entire data set," he explains. "We can take lower-level information—like SOPs for processing a claim—and give those documents to an AI agent to prepare work for a human to review."
This pragmatic approach succeeds because modern AI platforms offer a way to bypass the biggest security risk. "When you and I use ChatGPT, our data goes to OpenAI for retraining," he says. "But enterprise users get a copy of the LLM in their own secure tenant. The data stays within that tenant, fully controlled and managed by you."
To counter fears of obsolescence, he also champions a layered architecture that separates the AI model from the workflow. Though not without its challenges, the design lets companies swap in new models without rebuilding the system. "The execution part isn't going to significantly change in the next 24 to 36 months, but the models will," he says.
Fueling the future: The quick wins and ROI from these initial projects create a virtuous cycle. It frees up employees from mundane tasks, allowing them to be redeployed to help accelerate the very governance initiatives that were once stalled. "We're not saying put the governance project on the back burner," Milanovic clarifies. "We're saying work in parallel and use the wins from discrete AI applications to excite the people working on the bigger project. Now they can see, 'Oh, when we finish this, we'll be able to do so much more.'"
Survival of the fastest: "If you don't embrace AI as part of your business process, you won't have a business process left to automate," Milanovic warns. "Your business will be gone, replaced by competitors who did. This isn't the first time we've seen this; it happened with cloud-native companies, and it's happening again now." The lesson is clear: it's not the big that eat the small, but the fast that eat the slow.
