Guest Personalization
“What amenities should we offer this guest?”
Book a demo and get a free trial of the full platform: research agent, fine-tune capabilities, and forward-deployed engineer support.
By submitting, you accept the Terms and Privacy Policy.
Loved by data scientists, ML engineers & CXOs at
A real-world example
What amenities should we offer this guest?
Hotels spend $2-5 per guest on amenity and upgrade offers with a 3-5% take rate on generic campaigns. Personalized offers (right amenity, right guest, right price) convert at 15-25%, generating $30-80 in incremental revenue per stay. For a chain with 10M annual stays, moving from generic to personalized amenity offers generates $200-500M in additional ancillary revenue annually.
Quick answer
Guest personalization AI predicts which amenities, upgrades, and experiences each guest will purchase by matching offers to individual preferences, trip context, and spend capacity. Generic amenity offers convert at 3-5%. Personalized offers based on graph-learned guest profiles convert at 15-25%. A hotel chain with 10M annual stays generates $200-500M in additional ancillary revenue by shifting from generic to personalized amenity offers that match the right offer to the right guest at the right price.
Approaches compared
4 ways to solve this problem
1. Generic Amenity Promotions
Offer the same amenity packages to all guests via in-room collateral, email, and check-in upsell scripts. Blanket promotions with no personalization.
Best for
Properties with a single dominant amenity (resort spa, golf course) where most guests are interested regardless of profile.
Watch out for
3-5% take rate means 95-97% of guests ignore the offer. Worse, the wrong offer can feel tone-deaf: promoting a couples dinner to a business traveler, or a kids' club to a solo guest. Generic offers waste guest attention and brand credibility.
2. Segment-Based Offers
Divide guests into segments (business, leisure, family, luxury) and create segment-specific offer packages. Better than generic but still coarse.
Best for
Chains with clear guest segments and distinct amenity sets per segment.
Watch out for
Segments are too broad. 'Business travelers' includes a first-time guest on a tight budget and a Platinum member who always books the spa. Same segment, completely different preferences. Segment-based offers improve take rates to 8-12% but leave significant revenue on the table compared to individual-level personalization.
3. Collaborative Filtering (Purchase History)
Recommend amenities based on what similar guests purchased. The Netflix-style approach applied to hospitality.
Best for
Chains with large loyalty databases and rich purchase history across properties.
Watch out for
Optimizes for popularity, not individual fit. The most popular spa package is recommended to everyone, even guests who have never shown spa interest. Also suffers from cold-start: new guests and new properties have no purchase history to learn from. Cannot incorporate trip context (business vs. leisure, solo vs. family).
4. Graph Neural Networks (Kumo's Approach)
Connect guests, stays, preferences, feedback, and loyalty into a hospitality graph. GNNs learn individual uptake patterns from the full guest network, including trip context, companion type, and spend capacity.
Best for
Chains with diverse amenity portfolios, mixed guest profiles, and enough stay data to learn individual preferences from relational context.
Watch out for
Requires integrated data across PMS, POS, loyalty, and guest feedback systems. The model adds the most value for chains with 5+ amenity categories and diverse guest profiles.
Key metric: Personalized amenity offers convert at 15-25% vs 3-5% for generic offers. For a chain with 10M annual stays, this 5x take rate improvement generates $200-500M in additional ancillary revenue from the same guest base and amenity portfolio.
Why relational data changes the answer
Amenity preferences are deeply contextual. Guest GST001 (Platinum, 45 stays, family leisure trip) has a 72% probability of purchasing a spa package at $180 because: their past feedback highlighted the spa, they listed spa and golf as activity preferences, they are on a family trip (spa while kids are at the pool is a common pattern), and 68% of similar Platinum guests purchase spa at this property. This prediction requires connecting feedback, preferences, trip context, loyalty status, and property-specific patterns in a graph.
Flat recommendation models see a row of features and predict uptake. Graph-based models see the full guest history across properties, connected to trip context, companion type, and property-specific amenity patterns. SAP's SALT benchmark shows 91% accuracy for graph models vs 63% for gradient-boosted trees on relational tasks. RelBench confirms at 76.71 vs 62.44. In guest personalization, this accuracy translates directly to take rate improvements. Moving from 3-5% (generic) to 15-25% (personalized) on amenity offers with $50-200 average value generates $200-500M in additional ancillary revenue for a chain with 10M annual stays. The amenities are already available. The revenue comes from matching the right offer to the right guest.
Generic amenity offers are like a waiter who hands every table the same wine recommendation regardless of what they ordered. Some tables are having steak (recommend the Cabernet), others are having fish (recommend the Sauvignon Blanc), and some do not drink at all. A good waiter reads the table: the entree choices, the occasion, the price sensitivity. Personalized guest offers work the same way: read the guest's history, trip context, and preferences, then recommend the amenity that fits.
How KumoRFM solves this
Graph-powered intelligence for travel and hospitality
Kumo connects guests, stays, preferences, feedback, and loyalty into a hospitality graph. The GNN learns individual preference patterns across the guest network: how stay history, trip purpose (business vs. leisure), companion type, and feedback sentiment predict amenity uptake. PQL ranks amenities per guest per stay, maximizing incremental revenue while maintaining guest satisfaction scores.
From data to predictions
See the full pipeline in action
Connect your tables, write a PQL query, and get predictions with built-in explainability — all in minutes, not months.
Your data
The relational tables Kumo learns from
GUESTS
| guest_id | loyalty_tier | total_stays | avg_spend_per_stay |
|---|---|---|---|
| GST001 | Platinum | 45 | $520 |
| GST002 | Gold | 12 | $340 |
| GST003 | None | 2 | $280 |
STAYS
| stay_id | guest_id | property | purpose | companions |
|---|---|---|---|---|
| STY501 | GST001 | Beach Resort | Leisure | Family |
| STY502 | GST002 | City Hotel | Business | Solo |
| STY503 | GST003 | Beach Resort | Leisure | Couple |
PREFERENCES
| guest_id | room_pref | dining_pref | activity_pref |
|---|---|---|---|
| GST001 | High floor, ocean view | Fine dining | Spa, Golf |
| GST002 | Quiet floor, workspace | Room service | Gym |
| GST003 | Unknown | Unknown | Unknown |
FEEDBACK
| stay_id | overall_score | highlights | complaints |
|---|---|---|---|
| STY501 | 9.2 | Spa, pool area | Slow room service |
| STY502 | 8.5 | WiFi, gym | Noise from hallway |
LOYALTY
| guest_id | points_balance | nights_ytd | spend_ytd |
|---|---|---|---|
| GST001 | 125,000 | 12 | $6,240 |
| GST002 | 28,000 | 4 | $1,360 |
| GST003 | 0 | 0 | $0 |
Write your PQL query
Describe what to predict in 2–3 lines — Kumo handles the rest
PREDICT BOOL(STAYS.amenity_purchased, 0, 3, days) FOR EACH GUESTS.guest_id, AMENITIES.amenity_id RANK TOP 3
Prediction output
Every entity gets a score, updated continuously
| GUEST_ID | AMENITY | UPTAKE_PROB | OPTIMAL_PRICE | RANK |
|---|---|---|---|---|
| GST001 | Spa Package | 0.72 | $180 | 1 |
| GST001 | Ocean-view Upgrade | 0.65 | $85 | 2 |
| GST002 | Late Checkout | 0.58 | $45 | 1 |
| GST003 | Couples Dinner | 0.41 | $120 | 1 |
Understand why
Every prediction includes feature attributions — no black boxes
Guest GST001 -- Platinum, family leisure stay at Beach Resort
Predicted: 72% uptake for Spa Package at $180 (Rank #1)
Top contributing features
Spa highlighted in past feedback
9.2 score, Spa mentioned
30% attribution
Activity preference: Spa, Golf
Known preference
25% attribution
Family trip context (spa while kids at pool)
Leisure + Family
19% attribution
Spend capacity (Platinum, $520 avg)
High
15% attribution
Similar Platinum guests' Spa uptake
68% at this property
11% attribution
Feature attributions are computed automatically for every prediction. No separate tooling required. Learn more about Kumo explainability
PQL Documentation
Learn the Predictive Query Language — SQL-like syntax for defining any prediction task in 2–3 lines.
Python SDK
Integrate Kumo predictions into your pipelines. Train, evaluate, and deploy models programmatically.
Explainability Docs
Understand feature attributions, model evaluation metrics, and how to build trust with stakeholders.
Frequently asked questions
Common questions about guest personalization
How much revenue do hotels leave on the table with generic amenity offers?
A hotel with 500 rooms at 70% occupancy serves roughly 128,000 room-nights per year. At a 4% generic take rate on $100 average amenity value, that is $512K in ancillary revenue. At a 20% personalized take rate, that is $2.56M, a 5x increase from the same guest base and the same amenity portfolio. For a chain with 50,000 rooms, the gap is $200-500M annually. The amenities already exist. The revenue is lost to poor matching.
What is the best time to present personalized offers to guests?
The highest-converting touchpoints are: pre-arrival email (48-72 hours before check-in), check-in interaction, and in-app push notifications during the stay. Pre-arrival offers work best for planned experiences (spa, dining, excursions) because guests can build them into their itinerary. In-stay offers work best for spontaneous purchases (room upgrade, late checkout, minibar). The model should recommend both the amenity and the optimal presentation timing.
Does personalization require a loyalty program?
A loyalty program helps enormously because it provides cross-stay identity and purchase history. But personalization also works for non-loyalty guests using: booking channel signals (direct vs. OTA), stay context (dates, room type, companion count), and property-level patterns (guests at this property on weekends tend to purchase X). For Guest GST003 with no loyalty history, the model predicted 41% uptake for a couples dinner based purely on the leisure/couple trip context and property-level patterns.
How does personalization affect guest satisfaction scores?
Done well, personalization increases satisfaction because guests feel understood rather than marketed to. Hotels with personalized amenity offers see 0.3-0.5 point increases in overall satisfaction scores (on a 10-point scale) because the offers feel relevant rather than intrusive. The key is relevance: a well-matched offer feels like service, while a poorly matched offer feels like spam. This is why accuracy matters more than offer frequency.
Can guest personalization work across a chain with different property types?
Yes, and this is where graph-based models excel. A Platinum guest who loves spa at the beach resort may prefer gym access and late checkout at the city business hotel. The model learns these context-dependent preferences from the graph: same guest, different property type, different trip purpose, different optimal offers. Cross-property knowledge transfer is a major advantage of graph-based approaches over property-specific models.
Bottom line: A hotel chain with 10M annual stays generates $200-500M in additional ancillary revenue by personalizing amenity offers. Kumo's guest graph matches amenities to individual preferences, trip context, and spend capacity, lifting take rates from 3-5% to 15-25%.
Related use cases
Explore more travel & hospitality use cases
Topics covered
One Platform. One Model. Infinite Predictions.
KumoRFM
Relational Foundation Model
Turn structured relational data into predictions in seconds. KumoRFM delivers zero-shot predictions that rival months of traditional data science. No training, feature engineering, or infrastructure required. Just connect your data and start predicting.
For critical use cases, fine-tune KumoRFM on your data using the Kumo platform and Research Agent for 30%+ higher accuracy than traditional models.
Book a demo and get a free trial of the full platform: research agent, fine-tune capabilities, and forward-deployed engineer support.
