AI Effectiveness
Start Here Home Frameworks Journal Labs Subscribe

Framework

IRDA

Information → Recommendation → Decision → Action

Four stages where AI can sit in an enterprise workflow. Most production AI lives at Information or Recommendation. The interesting and still-unsolved work is at Decision and Action.

  1. I

    Information

    Does AI surface the right facts at the right moment?

    Worked example

    Healthcare: a hospital publishes its Machine-Readable File (MRF). Information is there but unreadable. AI structures it.

    Failure mode

    Polished retrieval demos that look great on five clean documents and crater on real internal sprawl.

  2. R

    Recommendation

    Does AI propose the right next move?

    Worked example

    Healthcare: given a patient and a price comparison, AI recommends which in-network option to consider.

    Failure mode

    Recommendations that ignore policy, context, or downstream side-effects on other parts of the workflow.

  3. D

    Decision

    Is AI the one making the call — or supporting a human who does?

    Worked example

    Healthcare: a procurement team decides whether to negotiate using the AI-surfaced comparison data. The AI is not deciding.

    Failure mode

    Pretending AI is deciding when no one wants it to; or pretending humans are deciding when they've been quietly automated out of the loop.

  4. A

    Action

    Does the recommended action actually get taken — and tracked?

    Worked example

    Healthcare: the negotiated rate is captured in the contracting system; outcomes are auditable.

    Failure mode

    AI shipping recommendations into a black box where no one can tell whether the action ever happened or worked.

IRDA composes with Decision Maturity into a 4×4 grid — see the periodic table for the full topology. The 1-page Decision Effectiveness Scorecard is a printable version of this lens.