Business imagination is the scarce asset now

What the paper develops

You chair the portfolio review that decides which AI ideas receive scarce capacity. One team brings a working assistant. Another has a vendor demo. A third shows time saved on drafting and analysis. All three can produce something. None has shown which work should change, what result would matter, or why AI is the right investment.

Cheaper execution is useful. Models can draft, summarize, classify, compare, search, and code. A field study of customer-support agents found a 14 percent average productivity gain in one defined workflow. Novice and lower-skilled workers gained the most. At enterprise level, McKinsey reported broad AI use but much less reported impact on earnings. A defined task can improve while a portfolio still lacks a sound way to choose among investments.

That gap has a cost. Capacity can follow the polished prototype, the familiar vendor, or the metric that is easiest to count. A promising operating problem may stay hidden because no one has framed it as a choice. Another team may carry a demo forward without showing which business result it can change.

The first constraint is the judgment to frame the work

A portfolio needs evidence about the work and its result. Counts of users, prompts, documents, code suggestions, and pilots may all rise. Cycle time, quality, customer experience, risk, and financial results may not move. McKinsey separates five levels of evidence: technical performance, user adoption, operating results, strategic outcomes, and financial impact. No single activity count proves the full case.

Failure can start before a pilot. Gartner predicted in 2024 that at least 30 percent of generative AI projects would be dropped after proof of concept by the end of 2025. It cited poor data, weak risk controls, rising costs, and unclear business value. RAND interviewed 65 experienced practitioners. They most often pointed to confusion or poor communication about a project's intent and purpose as the reason AI projects failed. RAND advised leaders to focus on the problem before the technology.

Cheap execution creates more ideas than a review can frame and compare. The first limit is often the judgment needed to turn one operating problem into a testable choice. This paper calls that judgment business imagination. It is the practical skill of turning an operating problem into a choice that a portfolio can compare, test, and fund.

Put one opportunity brief before the tool decision

Require one short opportunity brief before a material AI idea receives funding or scarce delivery capacity. Use it as well before the result becomes part of recurring work. The brief names the work, what happens now, how value may appear, what evidence earns more investment, who owns the result, which decisions stay with people, and which other routes the team should compare.

Consider a proposal to “improve portfolio reporting with AI.” That phrase is too loose to fund. The brief may reveal a narrower problem. Analysts spend hours fixing inconsistent fields before an executive review. The late work delays the decision. The team expects a shorter preparation cycle and a more reliable report. It should compare better data checks and a fixed rule before it proposes a model.

Compare routes before committing to a tool

Model access is only one route. A short test with a leading model may fit a valuable question with an uncertain answer. Stable, repeated, low-risk work may fit a cheaper model or fixed-rule automation. Poor source data may need repair first. A high-stakes judgment may need a human expert. A weak idea should wait or stop. The review weighs value, uncertainty, risk, repeatability, and the strength of the evidence.

Connect the pilot to the business result

Keep the brief with the idea after approval. Test five links. Did the tool work well enough? Did people use it in the workflow? Did the starting measure move? Did that change matter to the business? Did the result justify the full cost and risk? Early signs can stay visible without becoming a financial claim too soon.

At the next review, bring back the same brief with the evidence filled in. The portfolio can then explore, redirect, fund the next defined step, fold the idea into existing work, wait, or stop. Before funding the tool, require the team to name the work, value, evidence, owner, human judgment, and route that fit the job.