The Missing Piece in the AI Productivity Puzzle

More than 50,000 U.S. companies say they have adopted Claude, Anthropic’s AI tool. (Michael M. Santiago/Getty Images)

Artificial intelligence becomes more sophisticated and pervasive every quarter, but businesses are struggling to keep pace.

In a recent survey of finance leaders, two-thirds of respondents said that internal change management is their biggest unsolved issue in the AI transformation. Buying the technology is easy. Rebuilding the work around it is the hard part.

I have spent three decades advising leaders of the world’s largest financial institutions through change. I have never seen a new capability arrive this fast, and with such tailored applications for finance, achieve industry adoption so unevenly. Most firms haven’t worked out how to remake themselves to take full advantage of AI’s promise. While senior-level usage has risen, especially since the beginning of the year, analysts and associates are still the heaviest users inside large banks and asset managers.

Almost every AI tool shows this pattern at the start of its rollout. Use of the new tool is uneven at the level just below where business and process decisions get made. So the general processes of the business stay the same, regardless of how much faster or efficient any single step along the chain of processes might have become.

It is in translating AI agents into clear enterprise value that the challenge arises. Oftentimes, solving one problem just creates a bottleneck around the next problem. Anyone who has experience in a factory knows if you speed up one machine on the line, you don’t get a faster factory. You just get a new bottleneck somewhere else on the line.

New capacity is only created when a firm changes who does the work, in what sequence, against which data, and to what standard of proof. That is less a matter of whether or not the 20-somethings in the office are using the latest AI model and more so an organizational act that moves considerably slower than AI innovations or model releases.

An example: When a company is being sold, a banker pitches for the transaction, writes up the Confidential Information Memorandum, decides which buyers to approach, runs them through diligence, and negotiates to a close. The CIM was the obvious first thing to automate; using an AI model to create a fast draft takes a fraction of the time it once did.

But the deal won’t close weeks earlier just because one constraint was resolved. The CIM is only as good as the financials underneath it, which are still assembled by hand. Fix the financials, and the constraint just moves further down the chain to the list of potential buyers and the work of narrowing hundreds of names to a dozen or so that show real promise.

The way to correct this bottleneck effect and then see actual returns from AI adoption is to ask: What would this process, which may have been implemented 10 or 15 years ago, look like if redesigned with today’s technology? If a buyer list can be assembled and screened in an afternoon, should targeting still happen after the financials are built? To answer that requires senior workers who understand the technology well enough to know what it can and can’t do.

Most firms today look like pyramids. A large fleet of juniors doing grunt work sit below narrower levels of senior and midlevel folks doing more client-forward work.

I think we will need to see them shift toward a “skyscraper” shape. Partners and managing directors with deep client relationships and a wealth of experience will still be at the top but with a slimmer, longer layer of AI-fluent workers running complex workflows beneath them and a fleet of specialized AI agents at the base. In this type of organization, more humans will sit higher up on the value chain.

For that to work, however, there needs to be AI training at every level, not only for analysts and associates but for managing directors and partners, too. Senior people will need to be evaluated on their AI fluency and leadership. They will be the ones who have to drive the redesign and adoption of new workflows, after all.

While I am concerned about this challenge, there are examples of broad-based transformation already taking hold.

One of my favorite is a leading international bank that freed capacity by using AI. The bank built its technology, data, and AI capability, retrained the people it freed up, and sent them after midmarket commercial customers. This cut costs, which is great. But it also grew revenue and market share, which is arguably more important. That sort of opportunity will only be available to firms whose senior people understand the technology well enough to redirect it.

The AI-native financial firm isn’t a distant possibility, but it requires a deep rethink of how organizations operate. The firms that will define the category are the ones redesigning how work is done and making AI adoption a requirement at every level.

Guest commentaries like this one are written by authors outside the Barron’s newsroom. They reflect the perspective and opinions of the authors. Submit feedback and commentary pitches to ideas@barrons.com.

Kevin Buehler is the chief innovation officer at Rogo.

Copyright ©2026 Dow Jones & Company, Inc. All Rights Reserved. 87990cbe856818d5eddac44c7b1cdeb8