AI Decision Matrix

Same decision, every time. How much can you hand to AI?

Rate one business decision on two dimensions, ambiguity and risk. The matrix places it in one of four types, and each type splits the work between people and AI differently.

Based on research by

  • Massachusetts Institute of Technology (MIT)
  • MIT Center for Information Systems Research (CISR)
  • MIT Sloan School of Management
  • University of Münster
  • European Research Center for Information Systems (ERCIS)
MIT CISR Research Briefing

The framework

Four types of decisions

Ambiguity describes how clearly data determines the answer. Risk describes the consequences of getting the decision wrong. Together they give four decision types: routine, consequential, exploratory and strategic.

Each type splits framing, acting and learning differently between people and AI.

Graphic: MIT Sloan School of Management. Read the article

The AI Decision Matrix. Ambiguity on the vertical axis, risk on the horizontal axis. Routine (low ambiguity, low risk): automate and monitor; AI frames, acts and learns. Consequential (low ambiguity, high risk): manage to ensure reliability; AI frames and acts, humans learn. Exploratory (high ambiguity, low risk): enable experimentation to learn fast; humans with AI frame, AI acts, humans learn. Strategic (high ambiguity, high risk): lead with continuous oversight; humans with AI frame and act, humans learn.

Place one decision on the matrix

Describe a decision in a few sentences, or answer 20 short statements. The matrix moves with every answer, and you get a short report at the end.

Before you start

  • Taking part is voluntary. You can answer the questionnaire without agreeing to anything. Your result is then calculated in your browser only, and nothing is stored or sent.
  • If you agree, we store your answers, your description (if you write one), the optional details about your organisation and the date. We do not ask for your name or email address.
  • AI analysis: your description and answers are analysed by Claude, an AI model by Anthropic. Anthropic does not use this data to train its models. To avoid paying twice, AI results are kept in a cache for 30 days.
  • Research: Launch Labs and MIT CISR evaluate the stored data anonymously to improve the tool and the research. The data is stored in the United States. Privacy notice

AI analysis is off. Agree above to describe your decision and let AI pre-fill the questionnaire, or simply .

LowAmbiguityHigh
Exploratory
Strategic
Routine
Consequential
The matrix updates with every answer.
LowRiskHigh

Designing Decision Rights for AI

Leaders have to decide how people and autonomous AI share decisions. The briefing recommends managing AI as a portfolio of business decisions rather than a collection of use cases. Routine and consequential decisions pay off now in efficiency and scale. Exploratory and strategic decisions pay off later, in innovation and better judgment.

Read the briefing
  • AuthorsIna M. Sebastian, Peter Weill, Thomas Haskamp, Jan vom Brocke
  • SeriesMIT CISR Research Briefing No. XXVI-6
  • PublishedJune 18, 2026
  • Basis30 interviews with 27 executives at nine global companies in 2025 and 2026

Cite as

Sebastian, I. M., Weill, P., Haskamp, T., & vom Brocke, J. (2026). Designing Decision Rights for AI. MIT CISR Research Briefing, XXVI(6).

The authors

  • Portrait of Ina M. Sebastian

    Ina M. Sebastian

    Research Scientist, MIT CISR

  • Portrait of Peter Weill

    Peter Weill

    Chairman and Senior Research Scientist, MIT CISR

  • Portrait of Thomas Haskamp

    Thomas Haskamp

    Assistant Professor, University of Münster · Academic Research Fellow, MIT CISR

  • Portrait of Jan vom Brocke

    Jan vom Brocke

    Professor and Chair of Information Systems, University of Münster · Academic Research Fellow, MIT CISR