Mandatory address transformation: getting through the address mess with AI

Anyone who has ever sent a letter abroad knows that address formats can sometimes differ significantly. While in Germany an address usually follows the scheme

Street – House number – Postal code – Place of residence,

in France, the house number is usually placed in front of the street name. Things get really complicated when, for example, instead of a neighbouring country, the letter is sent to Asia, where completely different structures are used. Or take the USA: the United States Postal Service requires more than 200 pages to describe how US addresses may look – which formats and abbreviations are permissible and which are not.

Financial institutions also address countless parties around the world in payment transactions every day. The reason for this is that for payments outside the European Economic Area (EEA), the payer's address must be given and the recipient should also be indicated for smooth processing. This serves to enable the necessary checks, for example with regard to money laundering and fraud prevention.

The enormous diversity of address formats worldwide was not a major factor up until now. Because addresses are specified in an unstructured manner. Within the payment file simple fields in which the address is supplied as free text, the address lines (AdrLine), are available for this purpose. Only the name must be specified separately.

SEPA 2.0 puts an end to this. Because in the future, address data must be delivered in a structured form – for all SEPA payment formats. The changes will come into force gradually as of November 2023. From November 2025 at the latest, address data for SEPA credit transfers may only be delivered in structured form. And the challenges are not purely European: Swift and other market infrastructures have the same deadline. For payments within the EEA, the provision of address data remains voluntary. However, if banks decide to provide it, this must also be done in a structured way.

This means: in future, every component of an address must be included in the designated field. The Payment Markets Practice Group lists a total of 14 characteristics that can be assigned to a postal address.

The example shown in the figure is simple. Because everyone in Germany knows that 9 is the house number and Wiesenweg is the street name. Converting this data into the new format takes only a few seconds – provided that the application has the corresponding option.

But even then, the transformation would be a massive undertaking. Because financial institutions are sitting on millions of address data records that have to be transformed. And such simple addresses are the exception. If one estimates the required activities, a simple calculation quickly yields an effort of up to 250,000 working hours for an average financial institution with 500,000 corporate customers. In addition, there is the expense of training to equip staff with the necessary expert knowledge of the worldwide address formats.

In view of the scope, efficient approaches to solutions are therefore required. Regular expressions are out of the question in this case. As shown above with the example of the USA, the possibilities of address data even within one country are manifold and do not follow a regular structure. In addition, countless test data would be necessary.

Another option is address data services, for example from Google. However, those are not only expensive, but also questionable from a data protection point of view. Moreover, such services are often limited to certain regions or even countries.

An application based on artificial intelligence (AI) can provide a remedy. This allows data to be automatically transferred into the necessary structure. The AI is able to recognise structures on the basis of predefined training data and to transfer these structures to further cases.

We at PPI are happy to help financial institutions prepare and implement the transformation of address data. This includes the selection and adaptation of the appropriate AI application as well as the choice of the necessary training and test data.

In the end, the institutions receive a powerful and reliable solution from which not only they themselves benefit. Because corporate customers will also have to supply address data in a structured form in the future. Financial institutions that take the necessary transformation off companies' hands can gain a tangible competitive advantage.

Author: Eng.D. Thomas Stuht, Product Manager at PPI