AI Decision Matrix

Ina Sebastian, Peter Weill, Thomas Haskamp and Jan vom Brocke built the AI Decision Matrix at MIT CISR from interviews with 30 executives, on the premise that decisions differ in kind. It crosses ambiguity (how far the data determines the answer, and how far people agree on it) with risk (the consequences of an error and the difficulty of correcting it). Every decision splits into framing, acting and learning, and each component can be assigned to a human or to AI.

The four quadrants give AI different roles under human supervision throughout. At One New Zealand, an agent turns marketers' requests into SQL queries to build segments; the time spent fell 60%, and a human reviews before launch. Consequential decisions keep a person in the loop wherever price is involved, because a pricing error is judged unacceptable. Exploratory work, such as marketing content, lets framing move between humans and AI while governance is watched. Strategic decisions stay with human leadership: fifteen agents analyse network data and orchestration agents coordinate them, but trade-offs between reliability, customer experience and cost are not delegated.

Nicolas Suchaud of WEnvision reaches for the matrix for a different reason: Value × Feasibility and its variants stopped sorting anything once an agent prototype takes weeks to build. His answer is to prioritise decisions rather than use cases, and he ranks reversibility first. Beth Stackpole records the authors' own reframing: the question moves from where to use AI to managing the decisions it shapes.