Open banking and AI for household spending
Italian banking group. Household accounts aggregated into a single app, automatic expense classification and savings goals, inside a tight regulatory perimeter.
- My roleProject Manager — requirements gathering and delivery
- DurationDigital innovation programme under PSD2
- AreasPSD2 / open banking · Automatic classification · Gamification · App and web
The European PSD2 directive makes it possible to aggregate, inside a single application, current accounts held at different institutions. An Italian banking group wanted to turn that regulatory obligation into a feature customers would actually want to use.
Aggregating accounts is the easy part. The hard part is that a list of transactions across several accounts, belonging to several people in the same household, is an unreadable mass of data: nobody looks at it twice. To make it useful, expenses had to classify themselves, habits had to surface without the user filling anything in, and saving had to take a concrete shape rather than remain a good intention.
- 01
Understand
Requirements gathering with business, product, legal and compliance functions. In a banking group every requirement has at least three owners and no decision belongs to one person: much of the work is bringing together people who answer to different objectives.
- 02
Design
Definition of the features for the customer app and the website: household account aggregation, automatic classification of income and expense items, grouping of spending habits across similar user and household profiles, savings goals with gamification mechanics.
- 03
Deliver
Delivery of the features in-app and on the website, with careful verification of privacy and data-handling compliance at every step. In this sector a regulatory requirement is not an obstacle to work around: it is part of the definition of done.
- Multi-institution aggregation of a household’s accounts, available in-app and on the website.
- Automatic classification of income and expense items, with no manual input from the user.
- Grouping of spending habits across similar user and household profiles.
- Savings goals with gamification mechanics.
This applies whenever AI is used on personal data in a regulated sector. The decisive question is not “what can we get the model to do”, but “what can we show the user, on what legal basis, and who inside the organisation has to say yes”. Skip that and you build features that never leave the test environment.
The client is not named, for confidentiality. Sector and scale are real, as is everything else here.
Recognise the situation?
If something here resembles your case, send me a couple of lines. If it is not my ground, I will say so straight away.