2025: The Year of AI Execution

The experimentation phase is over.

For three years, enterprises have been piloting AI. Running proofs-of-concept. Building demos. Showing boards what's possible. The question was always "Can AI do this?"

That question is answered. We know what AI can do. 2025 is about whether your organization can actually do it.

The Pilot Purgatory Problem

Here's what the data shows: 72% of organizations have adopted AI in at least one function (McKinsey, 2024). That sounds like progress. But look closer.

Only 26% have successfully scaled AI beyond pilots to generate significant returns (BCG, 2024). The rest—74%—remain stuck in what the industry calls "pilot purgatory." Projects that never quite make it to production. Demos that never become products. Proofs-of-concept that prove nothing except that you can make a convincing slide deck.

This isn't a technology problem. It's an execution problem.

The 10-20-70 Principle

BCG's research reveals something uncomfortable about why AI initiatives fail:

10% of barriers are algorithmic—the AI itself
20% are technological—infrastructure, data, platforms
70% are organizational—people, process, change management

Read that again. Seventy percent.

The algorithms work. The technology exists. What doesn't work is the organizational machinery to actually deploy, adopt, and sustain AI in production environments.

Most AI strategies fail not because the model underperforms, but because the organization can't absorb the change. Employees don't trust the outputs. Business processes weren't redesigned. Nobody owns the outcome. The pilot team moves on. The model drifts. And six months later, everyone's back to the old way of doing things.

What Separates the Scalers from the Stuck

The gap between AI leaders and laggards is now substantial—and growing.

Companies that have successfully scaled AI show 1.5x higher revenue growth and 1.6x greater shareholder returns over a three-year period compared to their peers (BCG, 2024). That's not marginal. That's competitive divergence.

What do these leaders do differently?

They focus. While struggling companies scatter effort across dozens of pilots, leaders concentrate on a few high-priority use cases. BCG found that leaders pursue about half as many initiatives as others—but scale far more of them.

They invest in the boring stuff. Leaders dedicate roughly 70% of their AI effort to people and processes, not algorithms and tools. They fund change management. They train users. They redesign workflows. The model is 30% of the work; making it useful is the other 70%.

They track value obsessively. 86% of high-performing firms track ROI for AI models versus 71% of low performers (Goldman Sachs, 2022). They define success metrics before the pilot starts, not after it's built.

They design for production from day one. The most common pilot failure is building something that can't scale. Leaders architect their pilots with production constraints in mind—data pipelines that work at volume, integrations that fit existing systems, processes that real employees will actually use.

The Uncomfortable Truth About Your AI Pilots

We recently analyzed 598 enterprise AI case studies published over the past two years.

Zero had rigorous evidence. No control groups. No statistical validation. 66% were purely anecdotal—"we implemented AI and things got better."

The industry runs on anecdotes.

This matters because it means most organizations don't actually know whether their AI initiatives are working. They have impressions. They have executive enthusiasm. They have vendor promises. What they don't have is disciplined measurement of business impact.

2025 is the year this catches up with people. As budgets tighten and boards demand accountability, "we're experimenting with AI" stops being an acceptable answer. The question becomes: what value did you capture?

What This Year Demands

If you're still stuck in pilot mode, the path forward isn't more pilots. It's honest assessment of why your existing pilots haven't scaled.

Start here:

Audit your pilot portfolio. How many AI initiatives have you launched in the past two years? How many reached production? How many are still running and delivering measurable value? The answers will tell you whether you have a technology problem or an execution problem.

Kill the orphans. Every organization has pilots that nobody owns anymore. They're not quite dead, but they're not producing value either. End them. Free up resources for initiatives that matter.

Pick one to scale. Not three. Not five. One. Choose the pilot with the clearest path to production, assign executive ownership, fund the 70% (people, process, change management), and see it through.

Define value before you build. For any new initiative, specify what success looks like in business terms—cost saved, revenue generated, time reduced—before the first line of code. If you can't articulate the value, you won't capture it.

The Stakes Are Higher Now

The companies that scaled AI in 2024 aren't going to wait for you to catch up. They're compounding their advantage—building better data foundations, developing internal capabilities, refining their execution playbooks.

Meanwhile, the pilot purgatory crowd is burning budget and burning goodwill. Every failed pilot makes the next one harder to fund. Every abandoned initiative teaches the organization that AI doesn't really work here.

That's the real risk of 2025: not that AI fails, but that your organization concludes it can't execute AI. Once that belief takes hold, it becomes self-fulfilling.

The technology isn't the blocker. The question is whether you can execute.

Have a question? Reply to this email.

References

McKinsey & Company. (2024). "The State of AI in 2024: Generative AI's Breakout Year."
Boston Consulting Group (BCG). (2024). "Where's the Value in AI?" From BCG Global AI Survey.
Goldman Sachs. (2022). "Generative AI: Hype, or Truly Transformative?" Global Investment Research.
Applied AI. (2025). ZenML Case Study Meta-Analysis. Internal analysis of 598 enterprise AI case studies.