Where AI Creates Value: The Economics of Getting Work Done

Consider a decision that has been sitting on an executive’s desk for weeks. The team has met, circulated slides, requested another analysis, and scheduled a follow-up. Everyone is busy. The decision remains unresolved.

Now imagine that the team could assemble the relevant evidence, compare its options, and prepare a small test in an afternoon. The gain would extend beyond the hours saved. The organization could learn sooner, correct a mistaken assumption earlier, and spend less time pursuing an approach that customers may never want.

That possibility is the starting point for my book, The Economics of Intelligent Work, and for this publication.

The expense between an idea and a result

Much of the cost of organizational work lies in the effort required to move it forward. Someone must locate the information, explain the background, find the right expertise, reconcile conflicting accounts, and make the next step intelligible to another person.

These activities rarely appear together on a budget line. Yet they consume time and attention every day.

AI can reduce some of those costs. Working in ordinary language makes sophisticated computing accessible to people who would previously have needed specialist assistance. A manager can explore a question, examine alternatives, and develop something concrete enough for colleagues to evaluate while the problem is still fresh.

The opportunity is substantial. So is the need to decide which work deserves acceleration.

Start with the behavior you want to improve

A useful AI initiative begins with an observable problem: a customer waits too long for an answer, a handoff loses essential information, or a recurring decision requires the same investigation every month.

People, Process, and Platform provide an order for addressing it. Identify the people doing the work and the behaviors that produce useful results. Examine how those behaviors can be repeated reliably. Then choose technology that supports them.

This sequence matters because speed can magnify existing weaknesses. An organization that rewards volume may acquire a much greater capacity to generate material that nobody has time to review. A team that routinely questions assumptions may use the same capability to investigate alternatives more thoroughly.

The value depends on the practice that surrounds the tool.

Preserve the reasoning that others will need

An important meeting may produce a decision and a polished presentation. A month later, the presentation is easy to find, but nobody remembers which alternatives were rejected or what evidence would justify reconsidering the decision.

That missing history creates work. Someone has to reconstruct it, or proceed without it.

AI can help preserve a usable account of the reasoning: the evidence considered, the unresolved questions, the constraints, and the next action. The goal is selective memory that another person can retrieve and understand. An enormous archive offers little advantage if the relevant lesson remains buried inside it.

Make something that can be tested

Cheap drafts change the economics of beginning. A rough proposal, a prototype, or a sample analysis can give people something specific to challenge.

But a convincing draft is still a proposition. Its usefulness must be tested against the situation it is meant to improve. A customer can assess whether an offer addresses a real need. An operator can discover whether a proposed workflow survives a busy working day. A reviewer can check whether an analysis is supported by its sources.

The form of verification should follow the question. A well-written explanation cannot establish that a system works, just as a functioning system cannot establish that customers value it.

Spend the saved time deliberately

When producing another version becomes easy, deciding whether it deserves attention becomes more important. Endless revision can absorb the very capacity AI was supposed to release.

Executives therefore need stopping rules as well as starting points. What evidence is sufficient for the next step? What remains uncertain? How much more refinement would materially improve the decision?

The practical measure of progress is a change that customers can experience: a more reliable service, a clearer answer, a shorter wait, or a better outcome. Human judgment connects the capabilities of the technology to those results—and remains accountable when they fall short.

For a first experiment, choose one stalled decision or recurring handoff. Use AI to reduce the effort required to move it forward. Preserve the reasoning, test the result, and examine whether the improvement was worth the cost.

That is the inquiry I will pursue here: how organizations can make learning and useful action more affordable.