Breaking Free from Pilot Purgatory: Lessons from Albemarle’s AI Transformation

At AVEVA World, Jonathan Alexander, who leads Albemarle’s global AI and analytics efforts, shared a candid and deeply practical look at why so many industrial AI initiatives stall out—and how his team has driven $150 million in annual improvements by breaking through pilot purgatory.

“Over the past five years, we’ve started a global digital transformation journey where we branded AI as Albemarle Intelligence with the goal of upskilling our local engineers and our operators to realize the value from AI and analytics. To date, we have generated over $150 million in annual improvements.”

Alexander’s message to industrial AI leaders was clear: Scaling AI is about more than technology—it’s also about business value, culture, and persistence over years, not months.

10 Reasons You’re Stuck in AI Pilot Purgatory

Throughout his presentation, Alexander covered reasons why organizations get stuck in their attempts to scale AI. He also walks through ways enterprises can move beyond pilots and make AI an effective technology across the organization. 

1. Think Long-term Before You Start Small

Alexander cautioned against chasing technology for technology’s sake.

“Think about the problems that your enterprise has and then find a technology that can solve those problems. People tend to overestimate what you can do the first three to six months, but they wildly underestimate what you can do over a longer period of time like three to five years, 10 years even.”

2. Fix the Data Foundation

“Garbage in is garbage out,” said Alexander. “Our manufacturing processes have lots of data. Some of our sites have tens of thousands, hundreds of thousands of instruments. And some of our sites are 50+ years old, those instruments are held together by duct tape and chicken wire. How do you do AI at scale when you have that type of data and that infrastructure?”

Rather than replacing every instrument, Albemarle invested heavily in contextualizing data using the AVEVA PI System. The result: over 70,000 instruments contextualized, 1,200+ machine learning models trained, and 200+ improvement projects delivered. “And those improvement projects are what have delivered the $150 million in annual savings,” said Alexander.

From Emeritus

3. Solve One Problem Everywhere

Many organizations repeat the “solve one problem, then start over” cycle. Albemarle flipped the model: identify a common, high-value problem and solve it across the enterprise.

By scaling statistical process control (SPC) and targeted machine learning across all sites, Albemarle reduced variability and captured major savings—like half a million dollars annually from one batch process improvement.

4. Move from Dashboards to “Action Boards”

Data visualizations can become another art piece on the wall if they aren’t tied to operational change.

Alexander said, “When we talk about dashboards, it actually makes me feel kind of a little sick to my stomach. We say action board. If you have an insight but you don’t have an action plan afterwards, then it doesn’t go on the action board.”

By integrating insights directly into operational workflows, Albemarle ensures analytics translate into measurable change.

5. Avoid Shiny Objects

Alexander warned against overcomplicating things with too many new tools too fast, and stressed the importance of integration. Albemarle uses modern tools to accelerate proven methods like Lean Six Sigma, applying rapid problem-solving cycles daily and tackling longer-term projects faster. The focus is on execution speed and repeatability, not novelty for novelty’s sake.

6. Fragmented Technology, Fragmented Results

“Data is the new gold,” said Alexander.

But miners don’t find gold efficiently without the right tools. Alexander likened Albemarle’s engineers to gold prospectors wading through rivers of process data—there’s value there, but it’s hard to reach without the right resources and infrastructure.

The solution is a unified data environment. Albemarle’s “unobtainium dataset” aggregates cross-plant data so they can run high-dimensional models and reuse them for multiple problems without rebuilding pipelines each time. This removes a major bottleneck in scaling AI.

7. Success Doesn’t End at Launch

From Oil Now

Many projects fade after go-live because adoption stalls. Albemarle combats this by extending engagement long after initial deployment. Alexander said, “You've probably had some sort of installation, some sort of training, you maybe do a little bit of configuration, then you move on to the next site. But you forget that the journey afterwards, the adoption journey is exponentially longer.”

Deep engagement after go-live, strong change management practices, and cultural adaptability are all essential. Sometimes that means rethinking your approach entirely. For example, after a launch, Albemarle transitions from “Project AI” to “Operation AI” to exaggerate the long-term time horizon of their efforts. They work to embed solutions into daily work, coach operators and engineers, and maintain a global steering team where site leaders share progress and best practices.

8. If You Build It, They Won’t Come

Even the best tech won’t deliver value without people on board. Albemarle invests heavily in change management, using models like Prosci’s ADKAR and the Kübler-Ross change curve to guide adoption.

Alexander’s point: Engineers often overinvest in the tech and underinvest in people. Changing mindsets takes as much planning as changing systems.

9. Culture Eats Strategy for Breakfast

Even with strong change management, culture can throw curveballs. Alexander used a family anecdote: his kids refused gumbo—until he sprinkled goldfish crackers on top.

He said, “When you're approaching your digital transformation and your global problems, you're going to run into issues. You’re going to have to sprinkle on some goldfish.”

The takeaway: adaptation is key. If a strategy isn’t landing, adjust it to fit the local culture, even if the change feels unconventional.

10. AI is a Teammate, Not the Terminator

AI should support—not replace—human expertise. Albemarle’s generative AI assistant, built on Microsoft Azure OpenAI, answers process troubleshooting questions by drawing from the company’s own manuals and SOPs, citing the sources directly.

Alexander said, “AI for us at Albemarle is about the people—people empowered by technology unlocking the value of information.”

By positioning AI as an ally to operators, Albemarle fosters trust and accelerates adoption to turn AI into a teammate on the plant floor.

Breaking the Cycle of Pilot Purgatory

Alexander’s 10 reasons aren’t just a checklist—they’re a roadmap for moving from scattered pilots to enterprise-scale impact. Across every point runs a consistent theme: Sustainable AI value comes from aligning technology with business problems, integrating it into daily work, and adapting both processes and culture along the way.

Whether it’s rethinking how you measure success (action boards replacing dashboards), building a resilient data foundation instead of chasing shiny tools, or sprinkling goldfish on your change initiatives to fit local culture, the lesson is clear: Scaling AI is as much about people and process as it is about algorithms and models.

For industrial leaders, the journey out of pilot purgatory isn’t a straight sprint—it’s a long, deliberate climb. But as Albemarle’s results show, with persistence, structure, and adaptability, those small pilot sparks can become a global transformation engine delivering measurable, compounding returns year after year.

Watch the Full Session

For the complete “Top 10 Reasons Why You’re Stuck in Pilot Purgatory, Preventing ROAI & Economies of Scale” presentation from Jonathan Alexander, visit AVEVA’s session page.