SOPHISTICATED PREPARATION OF ADDRESS DATA

The Baloise Group is more than a traditional insurance company. Its business activities focus on the changing security and service needs of society in the digital age. Over the past 30 years, the company has invested heavily in the development of data warehousing in Switzerland and has established extensive data infrastructures. Among other things, the data enables client segmentation, which forms the basis for sales management and consequently enables optimal client support. In future, the visualisation of data at Baloise will also be used to ensure greater responsiveness in the event of loss events, for example, or to inform customers proactively about possible events such as storms.

Due to the constantly growing data basis, Baloise would like to focus more on the visualisation of data in future. Existing customer data records should therefore be enriched with geodata (coordinates). Although the overall data quality was very high, there was a heterogeneity in the database - due to the long history of the data - which made comparison with fuzzy matching more demanding.

Some of the address data entered contained minor errors: for example, districts or city districts were typed in instead of the city. Due to the multilingualism, there were differences in the names, for example, the street became the avenue or cities were stored in the system with Geneva, Genève and Geneva. Also postal codes were partly incomplete or wrong. Double entries and typing errors made the data assignment even more difficult. To obtain consistent customer data, the help of data preparation specialists was therefore necessary.

HOW ONEDOT HELPED

Baloise commissioned Onedot to link the customer data with geographical coordinates for visualisation purposes. The unstructured address data was previously cleansed, normalised and standardised for internal processing purposes. Onedot assigned and merged the street names, postal codes and city names using machine learning.

Baloise agreed the desired fuzzy matching granularity with Onedot in order to determine which fault tolerance is appropriate for this matching process. The granularity proposed by Onedot was determined using statistical and probabilistic methods and proprietary artificial intelligence (AI). Thanks to fuzzy matching logic, the data sources were precisely consolidated.

THE RESULT

Machine learning was used to optimally weight the various attributes and to prepare the address data for visualization. As a result, over 96% of the customer data was quickly made available for further analysis and prepared for import into the Baloise Analytics tools. As a result, Baloise was able to link the address data with building locations and geographical coordinates in order to graphically display the customer segments.