Set Participation Rights for AI Refund Decisions
Diagnose quality, information and commitment needs before selecting a feasible participation mode with a named owner.
21 minutes
Scenario
An ecommerce company is defining who should participate when an AI service recommends changing refund thresholds. The customer-operations director owns the policy, but finance, fraud, legal, support agents and the model team hold different information and will carry out the change.
- Your role
- Programme manager designing the decision process
- Method
- Vroom-Yetton-Jago Decision Model
Evidence pack
Quality
A poor threshold could increase fraud loss or unfairly deny legitimate refunds.
Owner information
The director knows service targets but not current fraud clusters or model calibration limits.
Distributed knowledge
Finance holds loss data, fraud holds abuse patterns, legal holds consumer-rights constraints and agents know exception behaviour.
Commitment
Agents can technically follow a new threshold, but past changes produced workarounds when rationales were not explained.
Goal alignment
All groups accept fair, fast refunds; finance and service disagree about acceptable fraud exposure.
Conflict
There is substantive conflict about trade-offs but no personal dispute.
Time
A regulatory deadline allows ten working days for the policy decision.
AI role
The model can simulate threshold effects but cannot own policy, stakeholder commitments or legal accountability.
Constraints
- One human role remains accountable for the decision.
- Participation is chosen from evidence, not as a reward or default cultural preference.
- The process must fit the ten-day deadline and include an escalation route.
Case steps
Work through each prompt using the evidence pack. Answers and rubric weights stay protected in the interactive flow.
Answer the diagnostic questions for decision quality, owner information, commitment, goal alignment and conflict using evidence IDs.
Eliminate participation modes that cannot satisfy the diagnosis and explain each elimination.
Select the feasible mode and design the ten-day sequence, participant roles, information inputs and decision owner.
Define what the simulation model may contribute, what it may not decide and when the issue escalates.