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.
- 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.
- 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.
- 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.
- 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.