Forecaster

Forecaster evolved at the start of the Global Pandemic, born out of a hotel client’s need to predict what their bookings would look like in the face of an unprecedented travel climate, with an uncertain future.

Our Data Science team solved this client need by building a tool which automatically ingests up to 130+ different covid, media and travel related datasets each week, feeding into 10 bespoke machine learning models, which learn how these fast-changing covid and market behaviours are affecting hotel bookings. These models are deployed into a front-end UI which allows users to test different media channel spend scenarios and predict 2 key KPIs, up to 2 months into the future. This helped clients to inform which Markets were most suitable to invest in as they restarted media activity mid 2020, how new lockdown measures would affect future bookings and revenue, and currently allows the media team to optimise budgets at a granular channel level, as markets come out of the pandemic. The fact that we used fast learning and highly accurate XGBoost models, which made predictions at an accuracy of average 90% on unseen data, meant that the team were able to be reactive to the fast-changing covid and travel factors- such as newly introduced restrictions, mobility levels, travel news sentiments and more. Each week these models ingest fresh data using an automated airflow data pipeline, which feeds into the models, embedded into our front end flexdashboard to produce live predictions as spend levers are changed within the tool. Forecaster has enabled the client to understand how market receptiveness influences demand and how spend in upper funnel journey phases influences lower funnel efficiency and contributed to two of the most successful quarters for the client in terms of media performance and efficiency.