Decision Frame
This is an explainable policy simulation. The model asks:
Given a service failure, guest relationship, operational constraints, and source-data confidence, what recovery gesture best protects the guest relationship without creating unnecessary cost or room-rate erosion?
The target is intelligent generosity, not comp minimization.
Evidence Classes
| Evidence | Use |
|---|---|
| Observed public booking data | Shapes synthetic booking and stay context |
| Official Santa Monica Proper public facts | Establishes property-fit and guest-facing value anchors |
| Sample-seed public context | Stress-tests pricing, review-risk, and demand logic |
| Synthetic operating systems | Demonstrates reconciliation, decisioning, and audit workflow |
| Policy assumptions | Defines weights, thresholds, cost ranges, and option scoring |
| Internal unavailable data | Prevents unsupported claims about margins, outcomes, inventory, or policy |
The current policy is smp-public-context-v1.0.0 in config/policy.v1.json.
Input Validation
One scenario contract validates the batch pipeline, CLI, HTML interface, and JSON endpoint. Probabilities must be between zero and one, severity must be a whole number from one through five, monetary values must be nonnegative, and unknown categories are rejected.
Derived fields such as guest-value score are calculated consistently. Missing or inferred source values remain visible through data-quality flags.
Policy Logic
- Calculate recovery need from severity, hotel responsibility, reputation risk, guest value, sentiment, delay, recovery timing, and issue baseline. Losing the in-stay recovery window raises timing risk; it does not reduce the measured need.
- Convert recovery need into a five-tier policy band.
- Score eligible gestures by issue fit, perceived guest value, estimated cost range, property fit, room availability, public-rate pressure, and repeat-comp pattern review. Room upgrades and late checkout are unavailable after checkout.
- Return the top gesture, two alternatives, manager-review decision, and policy version.
- Remove important context signals one at a time. Report a context reason only when the selected gesture changes.
- Rescore under ±20% perturbations to fit, cost, occupancy, external context, the overall recovery-need scale, and each individual recovery-need weight. Report the share preserving the selected gesture as stability.
Cost And Value
Guest-facing values can be anchored to published fees, credits, and wellness values. Internal marginal cost cannot be inferred from public price.
Each gesture therefore carries low, midpoint, and high internal-cost assumptions. These ranges support sensitivity analysis; they are not Proper Hotels accounting estimates.
The modeled recovery-value field is also policy-simulated. It should not be interpreted as causal revenue protected or projected return on comp spend.
Explainability
Direct reasons identify policy factors such as high severity or clear hotel responsibility. Context reasons require a changed counterfactual. For example:
Operational availability changed the recommendation:
without this signal, the model would prefer a room upgrade.High stability means the decision survives tested parameter changes. It does not mean the decision is empirically optimal.
Simulated Policy Audit
The historical comp ledger is synthetic. Audit classes such as under-recovered, over-comped, aligned, manager review, and data-quality hold demonstrate how a real historical policy could be reviewed. Their counts and dollar values are not observed findings.
Research Anchors
- Hart, Heskett, and Sasser, The Profitable Art of Service Recovery.
- de Matos, Henrique, and Rossi, Service Recovery Paradox: A Meta-Analysis.
- Tax, Brown, and Chandrashekaran, Customer Evaluations of Service Complaint Experiences.
- Vargas-Calderon et al., Review-Based Quality-of-Service Framework.
These sources motivate disciplined recovery, fairness, timing, and review signals. They do not provide property-specific policy weights.
Production Requirements
- Historical actions, approvals, policy versions, and manager overrides.
- Post-recovery satisfaction, review, repeat-stay, and cancellation outcomes.
- Marginal-cost ranges by gesture.
- Live occupancy, inventory, room type, outlet, staffing, and timing constraints.
- Jointly reviewed severity, responsibility, and escalation definitions.
- Prospective monitoring before any automated decisioning.