Airbnb
dynamic price suggestion to sellers (host pricing)
Airbnb computes a nightly price for each host with a three-part model (booking probability, optimal price, personalization), deployed in production and validated through A/B testing (KDD 2018).
Key points
- Dynamic price suggestion to hosts (Price Tips, Smart Pricing).
- In-house model in three parts: classifier, regression, personalization.
- Deployed in production, effectiveness demonstrated in online A/B testing.
- Evidence level B, confirmed active status.
Objective
Help each host set a nightly price that maximizes both booking probability and revenue, across a catalog where no two listings are alike.
The deployment
Price Tips, launched in 2015, suggests a nightly price to the host. Smart Pricing, launched the same year, automatically adjusts the price within the bounds the host sets. The model published at KDD 2018 combines three parts: a classifier that predicts the booking probability of each listing-night, a regression that predicts the optimal price with a custom loss function, and then a personalization layer that produces the final suggestion. The difficulty specific to Airbnb is that no two listings are identical, which prevents estimating a classic demand curve. The model is deployed in production and validated through A/B testing.
Results Proof B
Scientific paper published by Airbnb's teams at KDD 2018, describing a production system validated through A/B testing. No public revenue figure isolates the model's effect, hence level B rather than A.
How it works
Documented architectureThe stack in detail
- llm Classifieur de probabilite de reservation Airbnb in-house model that predicts the chance a listing-night is booked at a given price.
- llm Regression a fonction de perte personnalisee In-house model that computes the optimal price by balancing booking probability and revenue, published at KDD 2018.
- outil Couche de personnalisation Adjusts the final suggestion to the specific attributes of each listing.
- outil Price Tips et Smart Pricing Host tools that surface the suggestion and apply the daily automatic adjustment within the bounds set by the host.
How it runs, concretely
For ops teams-
1Predicting booking probability AI
A classifier estimates the chance that a listing-night is booked at a given price.
-
2Computing the optimal price AI
A regression with a custom loss function proposes the price that balances booking and revenue.
-
3Personalizing the suggestion AI
A personalization layer adjusts the final suggestion for the listing.
-
4Host settings customer
The host sets minimum and maximum price bounds and enables or disables the automatic adjustment.
The predicted booking probability per listing-night, crossed with local demand. Without this signal, the price suggestion loses its relevance and the host falls back on a static price.
How your customers perceive this type of use
Sourced studiesLe pricing algorithmique est le terrain le plus inflammable : 68% des consommateurs disent se sentir leses quand les marques utilisent le pricing dynamique et 80% jugent plus dignes de confiance les marques aux prix constants (Gartner, 2024). L'equite percue varie selon le secteur : le pricing dynamique n'est juge juste que par 33% a 40% des repondants selon qu'il s'agit de concerts ou de cinemas (YouGov, 17 marches). Le prix personnalise par les donnees individuelles est le plus rejete : 47% des Americains s'y opposent fermement (Consumer Reports, 2024).
Acceptance conditions
- La constance des prix comme signal de confiance : 80% jugent plus fiables les marques aux prix stables (Gartner 2024)
- Le secteur conditionne l'equite percue : le pricing dynamique est mieux tolere pour les cinemas (40% le jugent juste) que pour les concerts (33%) (YouGov 2024)
Red lines
- Le pricing dynamique percu comme abus : 68% se sentent leses (Gartner 2024)
- Le prix individualise a partir des donnees personnelles : 47% d'opposition ferme (Consumer Reports 2024)
- Les frais caches et hausses imprevues, vecus par 79% des consommateurs sur un an et associes a la perte de confiance (Gartner 2024)
Sources: Gartner 2024 · YouGov 2024 · Consumer Reports 2024
How to replicate
Inference, not sourcedData prerequisites
- booking and price history
- local demand signals (events, seasonality)
- listing attributes
Org prerequisites
- pricing data science team
- host tool to surface the suggestions
- A/B testing loop
Possible stack
- classifier and regression on tabular data
- personalization layer
- daily update pipeline
The plan, step by step
- Step 1Consolidate the booking and price history with local demand signals (seasonality, events).Deliverable: Listing-night dataset ready for training
- Step 2Train the booking probability classifier and validate it offline.Deliverable: Model evaluated with calibration curves
- Step 3Attach the price optimization (regression with a tailored loss function) and the personalization layer.Deliverable: Per-night price suggestion validated in backtest
- Step 4Surface the suggestions in the seller tool with minimum and maximum bounds controlled by the host.Deliverable: Price suggestion interface in beta
- Step 5Run an A/B test on a subset of hosts and measure occupancy and revenue.Deliverable: Controlled test result and deployment decision
First step: Build a booking probability model per listing-night before attaching price optimization to it.
The same pattern, proven in other industries
full-funnel AI ad suite (automated ROAS bidding, new-customer prioritization, budgets and targeting managed by the algorithm)
ML-optimized ad platform (automated campaigns + ranking + genAI formats)
Shopee
AI automation of seller ad campaigns
Sources
- S1 Customized Regression Model for Airbnb Dynamic Pricing (KDD 2018) Primary archive pending
- S2 Customized Regression Model for Airbnb Dynamic Pricing (ACM SIGKDD Proceedings) Primary archive pending
- S3 Airbnb Engineering and Data Science at KDD 2018 Interested party archive pending
An error, newer info, a source?
This page lives on its accuracy. If a figure has moved, if the deployment has changed, or if you have a higher-quality source, tell us. Every sourced correction is verified before publication.