Univé builds an AI-ready workforce

Most enterprise AI initiatives fail not because the technology doesn’t work, but because organizations never figure out how to make their people actually use it. The pattern is familiar: leadership announces an AI strategy, IT procures licenses, a handful of enthusiasts build impressive demos, and then adoption flatlines. Six months later, the CFO asks why utilization is at 12% and nobody has a good answer.

This piece is for the operations leader or IT executive who has already seen one AI rollout underwhelm and is trying to understand what went wrong—or what will go wrong next time if they don’t change their approach.

The Three Ways AI Workforce Initiatives Break

A recent case study from a Dutch insurance cooperative illustrates how one organization navigated AI adoption across 3,000 employees. But most organizations won’t replicate their results because they’ll skip the unglamorous groundwork that made it possible. The failure modes are predictable:

Failure Mode 1: The Pilot That Never Scales

The organization runs a successful pilot with 50 enthusiastic early adopters. Leadership sees promising results and announces company-wide rollout. But the pilot succeeded precisely because it was small, self-selected, and supported by dedicated project resources that won’t exist at scale. The typical pilot-to-production cost multiplier is 3–5x, and most of that cost is change management, not technology.

Failure Mode 2: Governance Theater

Legal and compliance create a 40-page acceptable use policy that nobody reads. IT blocks external AI tools but provides no sanctioned alternative with clear guidelines. Employees either ignore the policy entirely or avoid AI altogether because they’re afraid of making a career-limiting mistake. One large financial services firm we worked with found that 60% of their knowledge workers were using unsanctioned AI tools despite explicit prohibitions—not out of malice, but because the approved path was too unclear or too slow.

Failure Mode 3: Top-Down Use Case Mandates

Leadership identifies “high-value use cases” and assigns teams to implement them. But the people doing the actual work often have better insight into where AI would help—and where it wouldn’t. Top-down mandates generate compliance without conviction. The use cases get built, checked off, and quietly abandoned within two quarters.

What the Successful Minority Does Differently

The organizations that achieve broad, sustained AI adoption share a few characteristics that have nothing to do with which vendor they chose or how sophisticated their models are:

• They invest in governance that enables rather than restricts. The policy isn’t “don’t use AI for anything sensitive”—it’s “here’s exactly how to use AI for sensitive work safely, and here’s who to ask if you’re unsure.”
• They create formal roles for internal champions who aren’t IT. These are functional experts—someone in underwriting, someone in claims, someone in HR—who understand the work deeply enough to identify real applications and credible enough to bring skeptics along.
• They measure adoption behavior, not just license utilization. Active users per week matters less than whether those users are doing substantively different work. A 40% utilization rate with shallow use is worse than 25% utilization with deep integration into workflows.
• They budget for ongoing enablement, not just initial training. The two-hour kickoff session teaches tool mechanics. The follow-up office hours, peer learning sessions, and prompt libraries six months later teach actual productivity gains.

The Hidden Cost: Change Management at Scale

When organizations model AI costs, they typically account for licenses, integration, and initial training. What they miss is the sustained investment required to move from early adopters to mainstream users.

A reasonable estimate for a 1,000-person organization: the first-year technology cost (licenses, infrastructure, security review) might run $200K–$400K. The change management cost to achieve meaningful adoption—dedicated program management, internal champions with protected time, content development, workflow redesign support—typically adds another 40–60% on top. Organizations that skip this line item end up with expensive shelfware.

The math changes further when you factor in the productivity dip during transition. Even successful AI adoption creates a 4–8 week period where workers are slower, not faster, as they learn new patterns. If your deployment plan assumes immediate productivity gains, your ROI model is wrong.

Governance That Actually Works

Effective AI governance for workforce tools isn’t about control—it’s about clarity. Employees need to know three things:

• What data can and cannot be used with AI tools, stated in concrete terms they can apply without calling legal.
• What decisions require human review before acting on AI output, and what the review process looks like.
• Who owns the policy and how quickly they can get answers when edge cases arise.

The organizations that get this right typically assign a responsible AI lead outside of IT—someone with enough organizational authority to make judgment calls and enough domain knowledge to understand context. This isn’t a full-time role in most mid-market companies, but it needs to be someone’s explicit responsibility with dedicated hours.

One pattern that works: a tiered classification system. Green-light use cases that require no approval. Yellow-light use cases that need a quick check with the responsible AI lead. Red-light use cases that require formal review. Most employee questions fall into the first two categories, and having clear answers for those frees up governance bandwidth for the genuinely complex situations.

The Leadership Visibility Problem

In most failed AI initiatives, senior leadership announces the strategy and then delegates execution entirely. This signals that AI is an IT project, not a business priority. Employees—rationally—conclude that their core work matters more than learning new tools.

The successful organizations we’ve observed make leadership use visible. Not performative announcements about “our AI journey,” but actual demonstrations: the COO showing how they used AI to prepare for a board meeting, the VP of Sales sharing a prompt that improved their pipeline analysis. This isn’t about executives becoming power users. It’s about signaling that the organization’s most time-constrained people find this valuable enough to invest in learning.

The secondary effect: when leaders use the tools themselves, they encounter the friction points firsthand. They understand why the governance policy is confusing, why the approved tool doesn’t integrate with the CRM, why the training materials assume technical knowledge employees don’t have. That firsthand experience drives better resource allocation than any project status report.

The technology for AI-assisted work is mature enough. What breaks is the organizational infrastructure around it: governance that clarifies rather than restricts, change management that persists beyond launch, distributed ownership that survives the departure of the original project sponsor. Organizations that treat AI workforce enablement as a sustained capability-building effort—budgeted and staffed accordingly—will see returns. Those expecting tool deployment to generate behavior change will join the 70–80% that quietly write off their investment within two years.

The question for the next quarter isn’t which AI tool to buy. It’s whether you’re willing to fund the unglamorous work that makes any tool actually get used.