Efficiency.

It's perhaps the most overused word in business. It's the word that everyone reaches for when they can't quite put their finger on what they're actually trying to achieve. When it comes to AI transformation, where so much is unknown at the beginning of the process, and even what you're striving towards is fairly unclear, efficiency is presented as the end state: we're going to use AI to become more efficient.

But efficiency isn't just the wrong goal—at least at the beginning of the transformation—it’s a way to ensure failure. This is because transformation is a journey where the path is revealing itself to you the further you travel down it. You often make wrong turns or have to double back, because transformation, especially AI transformation, is more of a direction than a destination. It's not that you won't arrive somewhere, because you will, it's just that you don't know when or where you’ll ultimately arrive.

In the same way that the creative process is filled with waste and inefficiency—a necessary reality of trying to get to the idea or thought or artifact you’re trying to create—transformation is too. It's filled with experimentation and false starts and frustrating outcomes—all classically inefficient endeavors—but each one of those inefficiencies is teaching you and your organization what transformation can, and ultimately will, look like.

The most important part of transformation is not the goals that you set or even the vision that you create. It's that you start—where you are, with the team you have.

What Looks Like Waste

In a 2024 article on organizational learning, BCG argues that companies need to accept “the inefficiencies inherent to experimentation.” To discover what works, people need room to try things that might not.

Imagine a customer-service team experimenting with AI-generated responses. Before using them with customers, employees compare the drafts with their own, discuss what the model missed, and revise the instructions. Initially, this might take longer than answering the questions themselves. But that apparent inefficiency is helping them understand what context the model needs and where human judgment remains essential.

There’s a related issue at the individual level. In research published in PNAS Nexus in 2025, researchers compared learning through AI summaries with learning through conventional web search. Across seven experiments, participants using AI summaries reported shallower understanding and subsequently produced advice that was less detailed and original.

The researchers weren’t studying corporate transformation, but I think the distinction matters here. Getting an answer and developing an understanding are not necessarily the same thing. The seemingly inefficient work of reading different sources, considering competing ideas, and arriving at your own conclusion may be doing something that a ready-made answer cannot do for you.

None of this means every false start is valuable. You still have to examine what happened and use what you learned. But before you eliminate an inefficiency, it’s worth asking what you might be eliminating along with it.