The term comes from public-sector benefits management, where the UK Infrastructure and Projects Authority defines a disbenefit as the measurable decline resulting from an outcome perceived as negative by one or more stakeholders. It is a planned part of the benefits register, recorded and measured in the same way as the benefits.
AI deployments produce disbenefits that rarely appear in the business case: more escalations to supervisors, extra review work for a downstream team, slower handling for the customers a model serves badly, or a decline in quality while speed improves. The UK guidance notes that project teams are often reluctant to volunteer them.
A disbenefit can only be measured later if it was baselined first. Listing the likely ones in the baseline record before a pilot starts is cheap, and it stops a net loss from being reported as a gain.