Rethinking AI Operating Models

Most leaders add AI to existing org charts. The real advantage comes when decision rights, teams, and control structures adapt around AI.

Companies don’t break because AI fails. They break because the structures AI plugs into weren’t built for machines.

Why traditional hierarchies block AI leverage

AI doesn’t just automate tasks. It shifts who sees information, when it’s surfaced, and how quickly decisions can occur. Operating models designed for human-scale processes—approvals, handoffs, and linear reviews—fragment when a machine bypasses them in a second.

In a 2023 MIT Sloan and BCG study, organizations that gained the most measurable value from AI had redesigned how choices were made. They rewired decision paths that once moved vertically through layers and instead routed them directly to teams closer to execution. The companies that didn’t restructure saw slower outcomes, not from poor models—but from operational drag.

The logic is simple: a great model that needs five emails and two approvals to trigger action won’t beat a decent model that feeds directly into a team with authority to move.

How leading firms updated their operating model

Moderna was never in waiting mode. It organized product teams around AI-native cycles at the height of urgency. These teams worked across functions, accessed common sources of truth, and made production decisions themselves. They didn’t need to surface insights upward and wait. That design helped bring a COVID-19 vaccine to trial faster than legacy processes could have managed.

DBS Bank removed the internal boundaries that stalled action. It grouped engineers, analysts, and product owners into what it called “fusion teams.” These squads didn’t rely on permission from a central data function. They generated value autonomously, with AI embedded into daily choices. CEO Piyush Gupta described the ambition as becoming “AI-fueled at the core.” It wasn’t just tooling. It was a shift in execution power.

In both cases, new capabilities didn’t get layered on after the fact. Structural adaptation came first.

The case against full redesign—and where it fails

One argument still holds traction: AI, like cloud or mobile before it, can be absorbed into standing teams. Many organizations took that route. IT explored tooling. Business leaders scoped use cases. Lines on the org chart stayed put.

This approach has short-term appeal. Change fatigue is real. Leaders already manage stretched teams. A full redesign feels disproportionate, especially when only a few AI pilots are live.

But here’s where the tradeoffs emerge. AI systems don’t just support existing decisions—they challenge how, when, and by whom those calls are made. Routing those calls through legacy steps—compliance reviews, manager signoffs, cross-department escalations—slows the cycle. Useful insights get aged out.

McKinsey’s tracking studies show that the firms stuck in pilot mode tend to preserve traditional workflows. Meanwhile, those capturing broader share gains make structural edits to streamline how recommendations become real actions. Judgment zones get redesigned—not eliminated, but shifted to match the faster cadence AI allows.

Where to start the shift

Org redesign isn’t about whiteboarding a new visual. It starts smaller—with control points. Find where AI triggers a recommendation that humans can’t immediately act on. Use that lag as your signal.

If a model flags customer churn risk but the retention offer needs director approval, that delay is your bottleneck. If demand forecasts run weekly but purchasing decisions operate monthly, the model gets ignored. Follow these patterns and they’ll show you which control structures need realignment.

Ask three questions:

• What judgment calls are still routed through legacy approvals?
• Which teams could act faster if they didn’t need cross-functional permission?
• Where does a good recommendation die because no one has the authority to move?

Moderna answered those questions by embedding decision authority into its cross-functional teams. DBS gave its squads the autonomy to deploy—not report. In both firms, AI found traction where structures changed shape around it.

In most companies, it’s not the AI that’s the bottleneck. It’s the org design that still assumes insight waits for clearance.