Beyond ‘Pilot Purgatory’: What does it take to build AI that works?

AI conversations have moved past the point of curiosity.

Boards and leadership teams are no longer asking what AI might eventually do.

They are asking where it is actually working, what measurable value it is creating - and why so many promising experiments still fail to become durable operating advantages.

Across industries, companies have invested heavily in AI pilots, proofs of concept and impressive demos.

Yet many remain stuck in what I think of as pilot purgatory: the place where a tool works in a controlled environment but never survives contact with the complexity, exceptions and accountability required in production.

The problem usually isn’t the model In my experience, AI initiatives rarely fail because the underlying technology is not powerful enough.

They fail because of how the technology is applied.

A model can be impressive in a sandbox and still be irrelevant to the business if it is not embedded into a real workflow, connected to the right data, governed appropriately and measured against outcomes that matter.

That is why access to AI is no longer a differentiator.

Anyone can buy access to models or integrate a third-party tool.

The real advantage lies in the things that can’t be bought off the shelf: proprietary data, deep domain expertise, and the discipline to continuously improve AI once it is operating at scale.

For us, those principles come together in our Lean AI approach, rooted in a Lean operating model that drives continuous improvement through testing, learning, and acting.

Instead of chasing technology for technology’s sake, our Lean AI approach helps us move AI beyond experimentation and into production, where it can improve service, boost productivity and create real business value.

Production AI requires discipline, not experimentation for its own sake This is where many organizations get stuck.