Smoobu, the Berlin-based all-in-one platform for vacation rental hosts and part of the HomeToGo group, worked with Gemma Analytics to predict which trial users become paying customers. Gemma replaced a handcrafted segmentation model with a machine learning model that scores every single trial user and cut the forecast error in half.
The starting point
Smoobu helps hosts of every size run their short-term rentals, with bookings and calendars that stay in sync across channels. In a subscription business like this, growth depends on turning trial users into paying customers, so an accurate forecast of trial-to-paid conversions is essential. That forecast came from a handcrafted segmentation model. It sorted hosts into broad segments and gave everyone in a segment the same number. The model was slow, rigid and could not predict anything for an individual user. Sales and marketing could not tell which trials deserved attention, and conversions suffered.
Gemma had worked with Smoobu as its external data team since late 2022, from event tracking with Snowplow to the first data model and dashboards. The forecast could therefore build on data that was already tracked and modelled.
What we built
Gemma developed ML-based forecasting that learns from historical data and from how users behave during their trial. Instead of fixed rules, the model finds the patterns that separate future customers from other trials on its own, including signals the handcrafted approach missed. It relies far less on manual segmentation and needs no extensive interaction history before it can make a prediction.
The output is a conversion probability for every single trial user. These scores make targeted outreach possible: sales can focus on the trials with the best chance of converting, and marketing can use personalised incentives where they make a difference, which makes customer acquisition more efficient.
The result
Compared with the handcrafted segmentation model, the machine learning model reduced the forecast error by 50 percent. Instead of one number per segment, Smoobu had a conversion score for each trial user and could decide which trials to prioritise.
Tools and setup
The setup built on the data platform already in place within the HomeToGo group. Snowplow captures behavioural events on the web and in the iOS and Android apps, Snowflake is the central data warehouse, and dbt turns the raw data into a clean data model. Apache Superset provides the dashboards, and Hightouch handles reverse ETL, bringing data from the warehouse back into the tools that sales and marketing work with.