Enterprise AI has a proof problem. Budgets are committed, tools are deployed, pilots are running in every corner of the org. But when someone at the board level asks, “So what did we get for all that?” the room falls silent.
The cause usually isn’t the technology. We surveyed more than 200 enterprise leaders across the US, Canada, and Europe, and four elements showed up wherever AI actually produced impact: Leadership × Talent & Culture × Tools × Governance. Zero out any one of them and the equation collapses.
The AI adoption plateau
Only 26% of leaders said more than half of their AI pilots made it to production. And yet 83% described their execution goals as “realistic” or “ambitious-but-achievable.” Leaders believe scaling is possible. They’ve approved the business cases. They just haven’t built the infrastructure to make it real.
This is the adoption plateau. Pilots launch, show promise, and then stall. The blocker usually isn’t technical. AI that can do real work in the systems that matter is still concentrated in a handful of technical teams, while the rest of the org waits in line for someone to build something for them.
From productivity signals to business outcomes
Most organizations measure AI success by tracking usage: licenses activated, logins counted, tokens consumed. A recent New York Times piece on ‘tokenmaxxing’ showed how absurd this can get: employees racing to rack up billions of tokens like a leaderboard competition, with little connection to actual business value. Usage numbers tell you people are touching AI tools. They don’t tell you the business is better off.
Our research shows a clear shift in what leaders want to measure. 45% now rank tangible business outcomes as their top AI metric: pipeline acceleration, conversion improvement, churn reduction. Productivity gains still matter, but leaders are focused on whether the business line moved, not just whether activity happened around it.
Here’s what actually unlocks the next round of budget: 88% said internal proof drives further investment. Not a vendor pitch. Not competitive pressure. Evidence from their own teams that something changed. We see this internally at Zapier. The numbers that move our leaders show what shifted in the work itself: shipping faster, fewer handoffs, happier customers.
That proof comes from AI showing up in the workflows that run the business. A support ops team building its own escalation automation and shaving 5–10 minutes off every ticket. A marketing team scoring and routing leads in real time without waiting on engineering. The evidence is credible because it’s coming from the people doing the work, in the systems they already use.
Why pilots stall (and what’s actually blocking scale)
Our framework identifies four bottlenecks. They’re more interconnected than most leaders realize.
Integration is the single largest constraint. 46% of leaders named it their top bottleneck, almost as many as named the other three combined. AI models can summarize, classify, and reason, but they still need to connect to CRMs, ticketing systems, ERPs, and whatever else runs the business. The companies moving fastest figured out this isn’t a problem you solve one API at a time. It’s a platform problem.
Leadership owns the vision but no one owns the handoff. Our research found that 81% of leaders say they can scale a pilot within a year. Meanwhile, 91% of practitioners say their pilots stall after launch. Leaders are enthusiastic about new ideas but can’t create the momentum to go from pilot to production. Without one accountable owner coordinating across departments, promising pilots quietly stop moving.
Culture defaults to firefighting. 95% of practitioners say most of their time goes to operational triage: debugging broken workflows, chasing data-quality problems, manually handling exceptions AI can’t cover. That leaves no room for experimentation. The teams that break the cycle automate the repetitive operational work first.
Governance hasn’t caught up with how fast people are moving. 58% of practitioners say their governance structure hinders execution. But 63% admit to using AI tools without formal approval anyway. Employees want to be productive, and current approval paths can’t keep up.
If your AI approval path means escalating, waiting, and reviewing for weeks, people will route around it. The better model is governance built into the platform itself. IT sets the boundaries once (which apps, which actions, which data) and the platform enforces them automatically.
What separates the companies that scale from the ones that stall
The framework maps to five moves. None are technical. They’re organizational decisions a CIO or AI leader can make this quarter.
- Widen the builder base. The biggest predictor of AI ROI in our data is breadth of adoption. Strategic AI users are 3.8x more likely to describe their organizations as consistently forward-moving. At one of our customers, a global media company, non-technical teams have shipped more than 500 workflows, with 15% coming from citizen builders. That’s the kind of scale that generates board-ready proof.
- Run a hackathon, then protect the time to keep going. When Zapier rolled out AI internally, we held a week-long, company-wide hackathon. Everyone participated, regardless of technical background. The point wasn’t polished output. It was hands-on experience with what was changing. That single week shifted the conversation across the company. The data backs this up: roughly 30% of leaders and practitioners say peer-to-peer learning and hackathons are the best ways to build new AI skills. What people build with AI matters more than what they know about it in the abstract.
- Make governance invisible to builders but ubiquitous for IT. 70% of leaders now see governance as a competitive advantage, but only 4% expect full governance in place by 2026. The companies closing that gap fastest are building governance into the platform itself: workspaces scoped to departments, role-based permissions on apps and actions, real-time audit logs, and guardrails enforced at the point of execution. Builders don’t feel the governance. IT gets a clear picture.
- Solve integration before you add more pilots. Nearly a third of leaders cite integration and workflow skills as their top workforce gap, ahead of baseline AI fluency. LLMs have created a perception that connecting systems is trivial: describe what you want and it happens. Reality is more nuanced. Integration works best when it runs through a platform with managed authentication, enterprise-grade connectivity, and IT visibility into every action.
- Don’t let pilots become permanent. With 43% of enterprises investing $5 million or more in AI this year, you can’t afford initiatives sitting in pilot purgatory. Set clear timelines. Kill what isn’t working. Scale what is.
What comes next
The enterprises that pull ahead this year will be the ones that can walk into a board meeting and say, with evidence, “Here’s what this is doing for us.” That evidence comes from AI running in real workflows across every department, built by the people closest to the work.
The technology is there. The budgets are committed. 46% of leaders say promotions and pay will depend on AI fluency in 2026. This isn’t a side project.
The harder part comes next: empowering every team to build what they need, under governance IT trusts. The companies that solve that equation are the ones that will have something to show the board.
Zapier is an AI orchestration platform that connects more than 9,000 apps to help companies automate workflows and improve productivity. Since 2012, millions of users have trusted Zapier to automate everything from lead routing and data synchronization to customer conversations, all without writing code. By turning complex integrations into simple, point-and-click workflows, Zapier empowers teams of all sizes to focus on strategic work. From startups to Fortune 500 companies, organizations worldwide trust Zapier to streamline operations, reduce errors, and accelerate growth through intelligent automation.
