Judgment is the scarce resource in a world where intelligence is becoming a commodity. A business can produce more analysis than its best people have time to consider. Their most valuable contribution may be deciding which questions deserve the effort at all. That choice becomes more consequential as the cost of pursuing every other question falls.
Judgment runs through execution, too: choosing an approach, handling an unexpected failure, deciding whether the result is ready to use. AI can take on more of that work. The business still has to decide where the resulting effort is worth spending.
Consider the person everyone turns to when an analysis looks plausible but something feels wrong. They know which assumption matters, which exception changes the answer, and when another round of research would accomplish little. Their value comes partly from knowing where to stop.
It is easy to fill that person’s day with AI-generated work.
An agent drafts a report showing accelerating customer growth. The figures are cited, the chart is clean, and the explanation sounds reasonable. The expert notices that the earlier figure counts paying customers while the later one includes registered accounts. Both numbers came from the source. They do not support the comparison, and now someone has to reconstruct when the definition changed.
Across projects, the same expert may be needed to clarify definitions, check unusual results, or supply missing context. Each project saves its users time. While the savings appear in separate project reports, the combined demand lands on one calendar.
Checking the work may be necessary, but the person who can spot a flawed assumption in a report may also be the person who should be reconsidering the strategy behind it. An investment in AI quietly becomes a decision about where the organization's best judgment gets spent.
That is why the cost calculation cannot stop when AI produces an answer. The relevant cost includes building and running the system, checking and correcting its output, and supporting the people who use it. The comparison is with the process the business actually had, including its shortcuts and accumulated expertise. AI may save one person's time by creating more work for someone else.
But the value of expert time depends on what that time accomplishes. More expert involvement may be the best result. Someone who used to spend the afternoon gathering information may now spend it investigating the assumption that could change the decision. They still work the same hours, and no jobs have been eliminated. But the business can now investigate important questions it previously had no time to explore.
“How much time did we save?” cannot tell us whether that happened. The second question is: “What did that allow us to do?”
The answer might be work the team could finally take on, a decision examined more carefully, or a problem caught early enough to change course. Those benefits need evidence, even when a reliable dollar value remains out of reach. A faster output is the beginning of the explanation.
But that benefit isn't automatic. If AI adds another review step or leaves experts handling only difficult cases, it may leave little time for deeper investigation. Managers must decide which steps can stop, which cases still need review, and how to make room for more valuable work.
Expert judgment can also create value beyond the immediate task. The mistake in the customer-growth report could become a check that catches the same error in future reports. Capturing and maintaining that knowledge takes effort, but the next analyst no longer has to discover the problem from scratch. That is how judgment compounds.
A worthwhile investment shows where judgment was better spent, why the benefits justified the full cost, and what future work became easier because of it.
That is ROI that survives the second question.
Top comments (0)