Forrester argues that most AI products fail at the problem-definition stage, before any building starts. Analyst Zhi Ying Barry writes that few firms can link AI spending to revenue, customer outcomes or profit. Teams that count releases, features or usage fall into what she calls the build trap. She advises finding where value leaks across customer journeys and internal processes first. That means unnecessary effort, delay, cost or risk for customers and for the firm. Root-cause work should come before solution choice, using methods such as the five whys and fishbone analysis. She cites Starbucks, which sped up drink preparation when the real cause was order sequencing. She also quotes Cat Wu of Claude Code on defining the buyer, the problem and the top use cases. A problem definition template sits in an accompanying report for Forrester clients.