Dynamic Pricing
“What rate maximizes revenue for this room tonight?”
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 rate maximizes revenue for this room tonight?
Hotels leave 8-15% of potential revenue on the table through suboptimal pricing. Revenue managers update rates daily based on occupancy forecasts and competitor rates, but miss the demand signals embedded in search patterns, event calendars, weather, and competitor sellout cascades. For a hotel chain with 50,000 rooms at $150 ADR, a 5% RevPAR improvement generates $137M in additional annual revenue.
Quick answer
Hotel dynamic pricing AI optimizes room rates in real time by modeling the full demand network: competitor availability, event calendars, weather forecasts, and booking velocity patterns. Traditional revenue management updates rates daily based on occupancy forecasts, missing the cascading demand effects when competitors sell out, events overlap, and weather shifts booking behavior. Graph-based pricing models improve RevPAR by 5%+, generating $137M in additional annual revenue for a chain with 50,000 rooms at $150 ADR.
Approaches compared
4 ways to solve this problem
1. Manual Revenue Management
Revenue managers set rates based on experience, occupancy forecasts, and competitor rate shops. Rates are updated once or twice daily. The traditional approach at most hotels.
Best for
Small independent hotels where the revenue manager has deep local market knowledge and a manageable number of room types.
Watch out for
Human revenue managers cannot process all demand signals simultaneously. A competitor selling out at 2 PM creates a 4-6 hour window of unoptimized pricing before the next rate review. Event combinations, weather shifts, and booking velocity changes all happen faster than daily rate reviews can capture.
2. Rule-Based Revenue Management Systems (RMS)
Automated systems that adjust rates based on occupancy thresholds and booking pace rules (e.g., if occupancy >80% for a date, increase rates 10%). Most major hotel chains use some version of this.
Best for
Chains with consistent demand patterns and well-understood seasonality where rules can capture most pricing decisions.
Watch out for
Rules are static. They cannot adapt to novel demand situations (a conference plus a concert on the same weekend), competitor-specific dynamics (Hotel Alpha sold out, shifting demand to you), or weather-driven demand changes. Rule-based systems also treat rate changes as independent: raising the Standard rate without considering the impact on Deluxe upsell probability.
3. Statistical Demand Forecasting + Optimization
Forecast demand per room type per date using statistical models, then optimize rates to maximize revenue given the demand curve. The current best practice for sophisticated hotel chains.
Best for
Hotels with rich historical data and stable demand patterns where statistical models can learn reliable demand curves.
Watch out for
Treats the hotel as an isolated demand system. Cannot model how competitor pricing changes affect your demand, how event combinations create non-linear demand spikes, or how weather forecasts shift the leisure vs. business travel mix. The demand forecast and rate optimization are separate steps, missing feedback loops.
4. Graph Neural Networks (Kumo's Approach)
Connect rooms, bookings, competitor rates, events, and weather into a hospitality demand graph. GNNs learn pricing dynamics from the full market network, including competitor sellout cascades and event combination effects.
Best for
Hotels in competitive markets with significant event-driven demand, where competitor dynamics and demand cascades drive optimal pricing.
Watch out for
Requires competitor rate data (available from rate shopping services), event calendars, and weather data. Best value for hotels with 100+ rooms in markets with 3+ direct competitors.
Key metric: Graph-based dynamic pricing improves RevPAR 3-8% by capturing competitor sellout cascades, event combinations, and weather-driven booking velocity. For a 50,000-room chain at $150 ADR, a 5% improvement generates $137M annually.
Why relational data changes the answer
Hotel demand is a market network, not an isolated system. When Hotel Alpha sells out its Standard rooms at $169, guests searching for Standard rooms in the area are redirected to your hotel and Hotel Beta. Your demand just spiked, but your rule-based RMS does not know why. Meanwhile, a Tech Conference with 8,000 attendees coincides with sunny 78F weather, creating a business-plus-leisure demand combination that neither signal alone would predict. These are relational effects: your demand depends on your competitors' status, the event calendar, and the weather forecast, all interacting simultaneously.
Flat pricing models forecast demand based on your hotel's historical patterns. Graph-based models see the full market network: competitor availability, event combinations, weather impacts, and booking velocity trends. SAP's SALT benchmark shows graph models at 91% accuracy vs 63% for gradient-boosted trees on relational prediction tasks. RelBench confirms at 76.71 vs 62.44. In hotel pricing, better demand prediction translates directly to RevPAR. A 5% RevPAR improvement for a chain with 50,000 rooms at $150 ADR generates $137M in additional annual revenue. The revenue was always available. The question is whether your pricing captures it.
Pricing hotel rooms without market network data is like setting gas station prices without knowing what the station across the street charges. When the competitor runs out of regular unleaded, your demand surges and you should raise prices. When a marathon reroutes traffic past your station, demand spikes for a few hours. Static pricing misses these dynamics. Graph-based hotel pricing works like a smart gas station that monitors every competitor, every event, and every traffic pattern in real time to set the optimal price at every moment.
How KumoRFM solves this
Graph-powered intelligence for travel and hospitality
Kumo connects rooms, bookings, competitor rates, events, and weather into a hospitality demand graph. The GNN learns pricing dynamics from the full market network: how a competitor selling out at a lower rate creates demand spillover, how event combinations (conference + concert) create non-linear demand spikes, and how weather forecasts shift booking velocity by room type. PQL predicts revenue-maximizing rates per room type per night, updating as demand signals change.
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
ROOMS
| room_type | inventory | base_rate | floor | view |
|---|---|---|---|---|
| King Standard | 120 | $159 | 2-8 | City |
| King Deluxe | 60 | $219 | 9-15 | Ocean |
| Suite | 20 | $389 | 16-20 | Ocean |
BOOKINGS
| booking_id | room_type | check_in | rate_paid | channel |
|---|---|---|---|---|
| BK5001 | King Standard | 2025-03-07 | $179 | Direct |
| BK5002 | King Deluxe | 2025-03-07 | $245 | OTA |
| BK5003 | Suite | 2025-03-07 | $420 | Direct |
COMPETITORS
| hotel | room_type | rate | availability | date |
|---|---|---|---|---|
| Hotel Alpha | Standard | $169 | Sold out | 2025-03-07 |
| Hotel Beta | Standard | $189 | 4 left | 2025-03-07 |
| Hotel Gamma | Deluxe | $255 | Available | 2025-03-07 |
EVENTS
| event_id | name | type | attendees | date |
|---|---|---|---|---|
| EVT101 | Tech Conference | Conference | 8,000 | 2025-03-07 |
| EVT102 | Beach Music Festival | Concert | 15,000 | 2025-03-08 |
WEATHER
| date | condition | high_temp_f | rain_prob |
|---|---|---|---|
| 2025-03-07 | Sunny | 78 | 5% |
| 2025-03-08 | Sunny | 82 | 10% |
Write your PQL query
Describe what to predict in 2–3 lines — Kumo handles the rest
PREDICT SUM(BOOKINGS.rate_paid, 0, 1, days) FOR EACH ROOMS.room_type
Prediction output
Every entity gets a score, updated continuously
| ROOM_TYPE | CURRENT_RATE | OPTIMAL_RATE | OCCUPANCY_PRED | REVPAR_DELTA |
|---|---|---|---|---|
| King Standard | $159 | $199 | 98% | +$38 |
| King Deluxe | $219 | $269 | 95% | +$44 |
| Suite | $389 | $449 | 90% | +$48 |
Understand why
Every prediction includes feature attributions — no black boxes
King Standard rooms -- March 7, 2025
Predicted: Optimal rate: $199 (98% predicted occupancy, +$38 RevPAR)
Top contributing features
Competitor Hotel Alpha sold out
Spillover demand
30% attribution
Tech Conference (8K attendees)
Business travel surge
26% attribution
Current booking velocity
+45% vs same DOW
19% attribution
Weather (sunny, 78F = leisure demand)
Favorable
14% attribution
Days until check-in (short window)
6 days
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 dynamic pricing
How much can AI improve hotel RevPAR?
Graph-based dynamic pricing typically improves RevPAR 3-8%, depending on market competitiveness and current optimization maturity. Hotels using manual pricing see the largest gains (6-8%). Hotels already using sophisticated RMS see incremental gains of 3-5% from competitor cascade effects and event combination modeling. For a chain with 50,000 rooms at $150 ADR and 70% occupancy, a 5% RevPAR improvement generates $137M in additional annual revenue.
How quickly should hotel rates adjust to market changes?
The optimal rate update frequency depends on market dynamics. In high-demand markets with active competitors, rates should adjust every 1-4 hours. In stable markets, daily updates are sufficient. The critical moments are: competitor sellout (demand shifts within minutes), event announcement or cancellation, and weather forecast changes. Graph-based models detect these triggers and recommend rate adjustments in real time.
Does dynamic pricing hurt direct booking relationships?
Not when implemented correctly. The best practice is to maintain rate parity across channels while optimizing the overall rate level. Loyal guests should see consistent pricing with loyalty discounts preserved. Dynamic pricing optimizes when to be at $199 vs $159, not whether to show different prices to different guests. Transparency about 'today's rate' based on demand conditions is well-accepted by travelers, similar to airline pricing.
How does dynamic pricing interact with OTA (Online Travel Agency) distribution?
Rate parity clauses in OTA contracts mean your rate changes must be consistent across channels. Graph-based pricing optimizes the rate level that maximizes total revenue across all channels, then distributes consistently. The model also predicts channel mix at different price points: lower rates shift bookings toward OTAs (higher distribution cost), while higher rates shift toward direct bookings (lower cost). The optimal rate factors in channel economics, not just occupancy.
Can dynamic pricing work for small independent hotels?
Yes, but the competitive dynamics differ from chains. Small hotels compete in a local market with 5-15 direct competitors, making competitor monitoring essential. Graph-based models add particular value for independents because they lack the proprietary demand data that chains have across their network. The model compensates by pulling stronger signals from competitor pricing, local events, and market-level booking patterns.
Bottom line: A hotel chain with 50,000 rooms generates $137M in additional annual revenue through 5% RevPAR improvement. Kumo's demand graph captures competitor sellout cascades, event combinations, and weather-driven booking velocity that daily rate updates miss.
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.
