When AI Transformation Outruns the Operating System
A new artificial intelligence tool arrives, employees begin experimenting with it, and some of the early results are hard to ignore. Drafts take less time. Information is easier to find. Routine analysis that once consumed an hour may take minutes. A few employees find useful ways to automate pieces of their work, and leaders begin imagining what that speed could mean across the organization.
Then the old delays start showing up again.
The report may be finished sooner, but the approval still waits. Analysis may take half the time, but the decision still needs three meetings. One person can move much faster, but then the work reaches a handoff that operates exactly as it did before the technology arrived.
The tool changed. The surrounding work often changes more slowly.
That gap deserves more attention as organizations invest in artificial intelligence. We have good evidence that these tools can improve individual performance. The harder leadership problem begins when we expect those individual gains to become better organizational performance without examining what happens to the work around them.
Artificial intelligence can create real gains
A study published in The Quarterly Journal of Economics examined 5,172 customer support agents using a generative artificial intelligence assistant. Access to the tool increased issues resolved per hour by an average of 15 percent. Less experienced and lower-skilled employees experienced particularly strong gains.
Another randomized experiment involving 758 consultants found that participants using GPT-4 completed 12.2 percent more tasks, finished them 25.1 percent faster, and produced higher-quality work on tasks within the technology's capability range. The same study found a different result when participants faced a task outside that range. People using artificial intelligence were less likely to reach the correct answer.
The two studies give leaders a useful boundary. Artificial intelligence can improve performance when the task is appropriate, the workflow gives the technology something useful to contribute, and people know how to use the output. An organization does not need to redesign everything before an employee can save time or perform a particular task better.
The challenge appears when we assume that faster tasks automatically produce a faster organization.
Coordination moves at a different speed
Researchers recently studied 7,137 knowledge workers across 66 firms. Employees were randomly selected to receive access to a generative artificial intelligence tool integrated into applications they already used for email, meetings, and writing.
Among employees who used the tool during the latter half of the six-month experiment, time spent on email fell by about two hours each week and work outside normal hours also decreased. Yet the researchers did not detect broader changes in the quantity or composition of work simply from giving individuals access to the technology.
Earlier analysis from the study also pointed to an interesting operating pattern. People changed behaviors they could control themselves more readily than behaviors that required coordination with other people.
That fits what I have seen in operations more broadly. Improving one part of a system often moves the constraint somewhere else.
If artificial intelligence reduces the time required to prepare a report from two hours to twenty minutes, the next delay may be approval. Faster analysis may expose slow access to data. A quicker proposal may sit with someone who lacks the authority or information to make the next decision. Automated output may create a new verification requirement before anyone is willing to act on it.
Artificial intelligence did not necessarily create those constraints. Faster work can make them easier to see.
The work may need to catch up with the tool
Current organizational surveys point in the same general direction, although they cannot tell us that workflow redesign alone causes stronger financial results.
McKinsey's 2025 global survey found that workflow redesign had the strongest relationship with reported earnings impact among 25 organizational attributes it examined. At the time, only 21 percent of respondents using generative artificial intelligence said their organizations had fundamentally redesigned at least some workflows.
McKinsey's more recent research found that organizations reporting stronger artificial intelligence performance were also much more likely to report substantial workflow redesign. Nearly three-quarters of the high-performing organizations in its survey reported fundamental workflow redesign, compared with roughly one quarter of other respondents.
Deloitte's 2026 research offers another warning. Only 6 percent of leaders said their organizations were making strong progress in designing how people and artificial intelligence work together, and only 14 percent believed their organizations were adept at shaping those interactions.
I would read those findings cautiously. Successful organizations probably differ from struggling organizations in more ways than workflow design. Still, they give leaders a reason to look beyond adoption rates and license counts.
Follow the work
Before scaling another artificial intelligence tool across an organization, I would pick one important use case and follow the work from beginning to end.
Start with the result you are trying to improve. Identify exactly where artificial intelligence changes the work. Then keep going.
Who receives the output? What decision follows? Where does human judgment remain necessary? Who has authority to act? What information is missing? Where should verification happen? What happens when the output is wrong? Which part of the process becomes slower relative to everything around it?
RISE gives me a practical way to work through those questions. Radiate begins with the result and the capability required to reach it. Innovate examines how the work changes as new capability becomes available. Serve looks at the surrounding conditions, including authority, data, process, workload, and support. Endure asks how leaders will verify performance and reinforce a better way of working over time.
I would not use that exercise to prove that every workflow needs redesign. Sometimes the answer may be that the existing process handles the new capability just fine.
When it does not, buying more technology is unlikely to repair the handoff.
Artificial intelligence can make part of the work much faster, and in the right setting that improvement can be substantial. Once leaders begin talking about transformation, however, adoption is only part of their responsibility. They also have to examine what happens after the faster task is complete.
The practical test is simple enough to run: choose one valuable use case, follow the work all the way through, and find where the gain stops moving.
That's where the next leadership decision should go.
Comments