How HomeToGo Powers Real-Time Personalization and Improved Search Results
HomeToGo relies on Snowplow for comprehensive event tracking, capturing user interactions and behaviors across its platform.
Background
HomeToGo, a SaaS-enabled marketplace with the world’s largest selection of vacation rentals, connects millions of users with over 20 million vacation rental offers. Its ML-driven search and ranking system is central to the marketplace, swiftly guiding users toward optimal vacation choices. Recognizing the critical value of immediate user behavior in enhancing search results, HomeToGo sought to integrate real-time user interactions into its ML infrastructure without extensive engineering overhead.
Challenges
HomeToGo initially relied on batch data processed through data storage in Snowflake, offering robust but delayed insights. The ML engineering team identified critical limitations:
“At some point, you will reach a certain performance plateau in terms of what you can optimize if you have batch features. We were thinking, is there value for us if we get these near real-time signals into the search? That's why we started looking into solutions that we can use that don't require us to rebuild the whole infrastructure for ourselves. Because we are aware that this is quite cumbersome. And it takes quite some engineering effort to get this into a production stable state.”
STEPHAN CLAUS | DIRECTOR OF DATA ANALYTICS, HOMETOGO
The team faced significant challenges, including:
- Collecting and transforming timely event-level data into actionable ML features.
- Quickly accessing diverse data sources, such as in-app activity and third-party signals.
- Reducing latency from event occurrence to ML inference, a process complicated by multiple delays (data ingestion, processing, and state management delays).
These limitations impeded their ability to leverage real-time user behavior fully, crucial for enhancing predictive accuracy and user experience.
Solution
HomeToGo's ML team chose not to build custom infrastructure from scratch. Instead, they augmented their existing Snowflake setup by integrating Snowplow, a customer data infrastructure platform, to collect comprehensive user behavior data in real-time. Snowplow provided an event streaming pipeline with built-in schema validation, ensuring high-quality data ready for immediate consumption.
Next, the engineering team needed a feature platform to compute and serve these features.
To meet this need, HomeToGo adopted Tecton, a feature platform specializing in real-time ML feature computation and serving. Powered by its Rift managed compute engine, Tecton seamlessly handled complex data infrastructure tasks, including streaming feature calculation, online feature serving, and real-time monitoring.
This integrated architecture enabled HomeToGo to:
- Collect rich, real-time customer interactions via Snowplow.
- Instantly stream validated events to Tecton using Snowplow’s Event Forwarding capability (Snowbridge) for rapid feature computation.
- Maintain a unified, declarative framework for feature definition, simplifying both real-time and historical feature management.
The solution significantly reduced feature freshness to sub-second latency, enabling ML predictions based on truly real-time user behavior.
Results
The integration of Snowplow and Tecton significantly enhanced HomeToGo's ML capabilities:
- Feature Freshness: Achieved sub-second latency, enabling real-time predictions based on immediate user signals.
- Rapid Deployment: The new system was deployed in just a few weeks by a small team of two ML engineers.
- Consistency & Reliability: Unified feature definitions ensured consistent calculations across both streaming and historical data.
- Performance: Tecton’s online serving infrastructure reliably handled 100,000 requests per second with latencies well below 100ms.
Most importantly, the team can now efficiently experiment with new real-time features, quickly assessing their impact on model performance.
Looking Forward
HomeToGo plans to continue leveraging its advanced real-time infrastructure to further optimize search and recommendation systems. The team remains focused on:
- Further Latency Optimization: Refining every millisecond in the ML pipeline to enhance user experience.
- Dynamic Scalability: Adapting infrastructure dynamically to handle fluctuating traffic patterns while maintaining seamless performance.
- Innovation & Experimentation: Expanding ML use cases, benefiting from rapid deployment and minimal engineering overhead enabled by their current architecture.
HomeToGo’s approach proves that achieving fresh, real-time features is possible without massive engineering resources—giving them a strong competitive edge in the fast-moving vacation rental marketplace.
This content was repurposed from our original joint blog with Tecton and HomeToGo here.
Customer Case Studies
Explore real-life success stories from companies using Snowplow.
.webp)