More than 80% of enterprises have already deployed AI initiatives within their organization, but in 2025, over half of generative AI projects were abandoned after failing to produce a valid proof of concept.
The age of AI pilots is here, but the age of AI results appears to be lagging.
These two stats, taken together, can tell the wrong story. A company kicks off a pilot, it works, teams learn lessons and leadership is pleased. But when the next iteration begins, similar questions arise and the same lessons have to be learned all over again. Integration remains cumbersome, governance continues to slow things down and organizational capabilities don’t grow.
The pilot didn’t work after all or, even worse, AI can’t deliver the depth that was promised.
This is the wrong read. A pilot can easily succeed on its own terms, but still fail to improve the organization’s capacity to perform in the next iteration, one of the key problems facing leaders who are scaling AI in their enterprise.
Why successful AI pilots still leave enterprises stuck
Plenty of companies can point to at least one AI initiative that “worked.” Documents were summarized; records were classified; meetings were notated; workflows were improved; briefs were more thorough. These wins matter, but leave an important question unanswered: did the organization get any better at doing it the next time?
Leaders need to keep this question top of mind when evaluating the results of previous AI projects and the goals of future ones. When these pilots aren’t treated as building blocks toward a greater goal, they create value for individual teams without increasing organizational capacity.
Businesses have been using new technologies to increase capacity for years, since long before the dawn of AI. Electronic spreadsheets didn’t just make the process of creating spreadsheets faster, they increased what businesses were able to track in the first place. Assembly lines made it possible to make more cars in a shorter amount of time, yes, but more importantly, they made more kinds of cars possible to make.
This is the central tension in scaling AI for business. Getting better at launching AI pilots looks good in an investor report, but it doesn’t increase what the company is capable of doing. When AI initiatives are treated as building blocks instead, they leave behind something reusable: clearer governance, recorded rationales, better inputs and less rework. They leave something for the next iteration to pick up and use and, in turn, build upon for the iteration after that.
This is the difference between launching and compounding, and it’s the first step in solving many of the AI adoption challenges facing industry today.
The real obstacle in enterprise AI adoption is the reset loop
AI projects don’t fail to compound on their own. Teams do, in fact, learn and grow during the pilot: individual capabilities increase, data is gathered faster, insights are put to use more effectively. The issue is not what happens during the pilots, but in what happens (or doesn’t) after.
How did this integrate with existing systems? Who owned it after launch? What evidence was preserved? What approvals were required? What happens when initial conditions aren’t consistent across iterations?
If these questions aren’t anticipated at the start of the project, then they won’t be answered at the end and the next team will have to spend time and resources learning the same lessons all over again. This is the pilot reset loop, and it’s the first visible sign of a failing AI culture within a company.
Pilots produce outcomes. It is up to organizations to ensure that the structure, context and decision-making that lead to those outcomes are logged so the next initiative isn’t starting again from zero.
AI implementation challenges are organizational challenges
The tools, models and technology used in stalled AI efforts are important, but AI implementation challenges arise from an organization’s approach to documenting what happens. When teams rely on context shared in conversations instead of systems, the result is the aforementioned restart from zero.
Real AI scaling looks like new initiatives building on the preservation of previous projects’ learnings: improved governance, better inputs, documented decision-making and identified bottlenecks. Until teams start planning for what they’re going to leave behind for the next team, the pilot reset loop will continue to spiral.
It’s not about whether the pilot worked, it’s about what it left behind
To truly win enterprise AI adoption, leaders have to start thinking about what the next team will pick up and run with. Pilots prove AI has value, but on their own, they do not prove it has value to the organization.
As with everything, the tool is only as good as how its wielder uses it.
Our eBook, Converting Tribal Knowledge into Institutional Intelligence: An AI-Enabled Operating Model, explains the reset loop further, examines why it persists and offers a model for moving beyond it. It provides a way for organizations to apply learnings forward instead of starting from zero each time.
