Skip to content
Fast Company Executive Board

The Fast Company Executive Board is a private, fee-based network of influential leaders, experts, executives, and entrepreneurs who share their insights with our audience.

Enterprise AI’s 5% success rate is a governance problem

The case for redesigned workflows and functional governance.

[Photo: Getty Images]

Profile photo of David Talby

CEO · John Snow Labs

Organizations were not built to absorb enterprise generative artificial intelligence. A 2025 MIT NANDA study found that just 5% of generative AI pilots deliver measurable financial returns.

Workflows meant to connect AI and governance haven’t been redesigned either. Closing that gap is now the central problem in enterprise AI. Technology stacking alone won’t fix it.

For years, companies have treated AI governance as a safeguard—a regulatory backstop separate from the systems it oversees. That framing breaks down once AI is embedded in everyday operations.

Many enterprises have invested heavily in governance programs. They have written policies, defined approval processes, and dedicated oversight committees. Yet, those programs sit alongside the AI systems they govern rather than inside them. The result is a familiar pattern. AI projects stall, teams work around controls, and the gap between what’s documented and what’s deployed keeps widening.

WHEN GOVERNANCE LIVES OUTSIDE THE SYSTEM

A different model is emerging at companies that have moved past pilots. Instead of treating governance as a checkpoint at the end of the development cycle, they build it directly into how AI systems are designed, tested, and run. Controls are embedded into pipelines. Monitoring runs continuously. Compliance evidence is generated as a byproduct of normal operations rather than assembled in a panic before an audit.

This shift makes AI more usable. Compliance becomes less burdensome. Teams stop pausing to interpret policy at every step because the guardrails are already part of the operating system. As models change, the controls change with them, avoiding the lag that comes with static documentation and quarterly review cycles. The point is to replace governance theater—forms, templates, and after-the-fact attestations—with functional governance.

The clearest examples come from regulated industries. HSBC’s December 2025 partnership with Mistral AI, as Reuters reported, brought generative AI into internal workflows under the bank’s existing responsible-AI framework. It included self-hosted models, controlled deployments, and structured integration into how teams already work.

Citigroup took a different angle. According to a Business Insider report, the bank built a 4,000-person internal network of “AI Champions and Accelerators” embedded across 84 countries. These “champions and accelerators” are peer guides who help colleagues use approved tools inside redesigned processes, replacing central trainers handing down policy from the top.

WORKFLOW FRICTION

Even with stronger governance, many AI initiatives struggle for a simpler reason: Existing workflows weren’t designed for them. AI changes more than which tasks get automated. It moves decision points and shifts where responsibility sits.

When that shift goes unaddressed, confusion follows. Teams lean too heavily on automated outputs and miss moments when they should intervene. Other teams hesitate to rely on AI at all, unsure who is accountable when something goes wrong. Shadow AI fills the vacuum, creating more risk than the original system was meant to solve.

Fixing this is an operational problem. AI needs to function alongside redesigned workflows that include clear decision points, defined human oversight, and well-understood escalation paths. Without that, even technically sound systems take longer to deliver value, or fail to deliver it at all.

The security side of the same problem is also now visible. The EchoLeak vulnerability disclosed in mid-2025 (CVE-2025-32711, CVSS 9.3) showed how an attacker could exfiltrate sensitive data from Microsoft 365 Copilot through a single crafted email, with no user interaction required. A follow-up exploit, Reprompt, was disclosed in early 2026. A different vector was targeted with a similar result

Neither cyberattack was a traditional security breach. Both happened because governance had not yet adapted to how AI interprets and responds to language inside enterprise data flows. Similar issues continue to surface in deployed chatbots, where prompt injection and weak output controls have produced data exposure and reputational damage.

FROM COMPLIANCE EXERCISE TO BUSINESS ENABLER

Many organizations can show that their AI systems meet compliance requirements. Far fewer can point to consistent, measurable business impact from those systems.

Treated as a cost, governance competes with the push to expand AI adoption. Tied to performance, governance serves a different purpose. It ensures AI systems are reliable, deployable at scale, and aligned with business goals from the start.

This requires a shift in how success is defined. AI initiatives require connection to business outcomes, with clear ways to track usage, measure impact, and feed those insights back into how systems are governed. Better governance produces better performance. Better performance generates better data on what governance should require next. The two reinforce each other when they are built into the same system.

AI AS A SYSTEM, NOT A SET OF PROJECTS

A new operating model is emerging—one where governance, lifecycle management, workflow design, and business metrics function as a single system rather than a collection of disconnected efforts.

Teams spend less time reconciling competing requirements. Decisions move faster because expectations are clearer. Successful use cases replicate more easily because the underlying foundation is consistent.

It also creates resilience. As regulations evolve and AI capabilities advance, organizations with integrated approaches can adapt without rebuilding their processes from the ground up. Governance is part of how organizations enable AI to work, not just how they prevent it from failing.

Trust is the connective tissue. When systems are well-governed, leaders are willing to deploy them more broadly. Employees are willing to rely on them. Customers and partners are comfortable engaging with them. When trust is missing, progress slows, regardless of the capability of capable the underlying technology.

Companies producing real results from AI aren’t ignoring governance. They also aren’t drowning in it. They’re rebuilding how governance functions—moving it out of the policy binder and into the operating system.

The work is harder than buying another tool, and slower than running another pilot. But the 95% who can’t get past pilot are giving up their ability to deploy AI responsibly, at scale, with confidence. It’s the strongest competitive advantage available right now and one no company needs to lose

David Talby is CEO of John Snow Labs and Pacific AI.