HSE: Predictive marketing analytics platform for hospitality industry

Main article: Definition of Business Intelligence

2023: Building a predictive marketing analytics platform

Scientists of the Center artificial intelligence HSE , together with New Service Bureau JSC, have developed and for July 2023 are testing for 6 hotels located Leningrad Region in and (Karelia Igora resort, Dacha Winter park hotel, Tochka on Map hotels in Priozersk, Sortavala, Vidlitsa, Lodeynoye Pole) a platform for predictive marketing analytics the hospitality industry. The HSE announced this on July 24, 2023.

The predictive marketing analytics platform processes data on the number of bookings, their attributes (number of days of accommodation, number of days before the expected check-in, period from the date of booking, amount of prepayment made, sales channel, etc.), visitors to sites, conversion of clicks on the site, marketing activities, media plan, weather dynamics in order to predict the demand for hotel and other services based on an intelligent algorithm for selecting signs and automatically building a model.

The platform provides the following functions: functions of connecting data of different nature, engineering of factors, the procedure for selecting significant factors, machine learning of the developed model for predicting the number of bookings, functions for assessing the probability of cancellation of reservations. The platform automatically combines data streams from external sources and builds them into a database, solving the problem of the different nature of information.

The forecast of various target variables, such as the number of bookings, the number of room-nights, etc., is implemented using algorithmic machine learning, taking into account the trend and seasonal decomposition of time series. During the development of the platform, various algorithms for building models were tested.

The best set of models was determined using the rolling cross-validation procedure by time and calculation of accuracy metrics on historical data. The logic of the models allows you to form a forecast in the context of each combination of hotel and room class.

For statistical processing of data, an automated service for launching a model script according to a schedule has been developed, which further increases the accuracy of each next forecast. The use of machine learning technologies has significantly increased the speed of processing calculations: the demand forecast is calculated up to one hour.

In total, 756 machine learning models were trained during the development process at 7 different time periods, more than 2600 factors were constructed and studied, 2268 experiments were carried out with measuring the accuracy of models at various prediction horizons. The accuracy of the demand forecast for the horizon is up to 94%.

The development of the frontend, collection service and access to the database of the predictive analytics platform was carried out by the IT company from St. Petersburg "Activika."

Developed software will allow hotels Russia to more efficiently plan and implement marketing activities, optimize the settings of ongoing advertizing campaigns against the background of reducing booking depth and volatile demand.