Build and own a Guest Personalization Specialist Model.
What room, amenity, offer, or experience should be recommended or proactively arranged for this specific guest?
Overview
Increases ancillary revenue through targeted upsell and amenity offers.
Who owns this decision
Chief Guest Experience Officer, VP Revenue Management
Why owned AI may fit
Guest preference patterns are property- and brand-specific, and depend on a hotel's own loyalty and booking data rather than general knowledge.
What this Specialist Model actually does
Decision: What room, amenity, offer, or experience should be recommended or proactively arranged for this specific guest?
This is a larger system with 2 specialist model candidates and 2 supporting or external components.
Who uses this Specialist Model
- Chief Guest Experience Officer
- VP Revenue Management
- Director of Guest Services
- Head of Loyalty and Marketing
Where the specialist model sits
This is a larger system with 2 specialist model candidates and 2 supporting or external components.
Recommendation model trained on historical guest preference and offer-acceptance data
Specialist Model
Specialist LLM
Specialist Model
Specialist LLM for generating personalized guest-facing messaging and staff briefing notes
Integration layer connecting property management system, loyalty platform, and guest digital touchpoints
Supporting component
How Oumi connects to your stack →
Real-time inference service
Supporting component
Real-time inference service for check-in and in-stay recommendation delivery
How to evaluate and improve it
Evaluation
- Incremental ancillary revenue per stay
- Guest satisfaction score lift vs. baseline
- Recommendation adoption rate by staff
Feedback signals
- Offer acceptance or decline
- Post-stay satisfaction survey and review scores
- Staff override of recommended assignment or offer
- Loyalty tier change following stay
- Repeat booking within a defined window
- Complaint or service-recovery incident following a recommendation
Model candidates to evaluate
- qwen3.5-9b
- modernbert-large
These are candidates, not performance claims. Run them against your evaluation set.


