Error at Scale: The Economics of Automating Bad Decisions

Business cases for enterprise AI typically count the savings from producing recommendations faster and leave out the expected cost of acting on recommendations that are wrong. This paper argues that AI systems automate bad decisions through three channels (outdated sources, inconclusive sources, and bad data) and that automation multiplies the resulting losses by removing the human checkpoints that once contained them. Drawing on illustrative scenarios and six public cases, it introduces planning formulas for expected excess loss, the net value of automation, and the break-even threshold for remediation, and shows how a defect in a shared source turns independent errors into correlated, systematic risk. It applies value-of-information theory to explain why decision-specific remediation outperforms enterprise-wide data programs, and identifies an internal externality that biases organizations toward over-automation. The paper proposes Technology → Value, a four-stage framework with gates at Intent, System, Operations, and Value, as a structure for future validation.