Vidyasoft wins the sella data challenge with HF- SCA
In the Sella Data Challenge, a project carried out for Banca Sella by Fintech District, open innovation won and the projects of the participating fintechs demonstrated how concrete and beneficial it can be. The best one was Vidyasoft’s, which won with the jury and will now be able to continue collaborating with the bank. Roberto Vergallo, Chief Innovation Officer of Vidyasoft, tells us what it is all about and what the next steps will be
Can you tell us about the project that won with the jury? (description and who it serves and the role of the data provided by sella)
Hands Free SCA (HF-SCA) is a tool that enables the assessment in real time of the risk of an online transaction on the basis of the cardholder’s previous spending habits. If the purchase is ‘plausible’ for the specific user, the bank can avoid Strong Customer Authentication, i.e. the phase of an online payment in which the customer is asked to authorise the transaction by typing in a code received on their smartphone. It may seem strange, but in April 2021 the SCA recorded a 22% drop in online purchases compared to the same period the year before, due to difficulties in the process (codes that are too long, expired time, use of different devices and apps, etc.) or even technical problems (e.g. codes that don’t arrive). Thanks to the data provided by Sella and Fabrick during the challenge we were able to train a deep neural network that understands when it is necessary to use the SCA and when it can be safely bypassed. To demonstrate the high UX of our solution, we also developed an application for the Alexa Echo Frames – Amazon’s smart glasses – which we called Personal Shopper and which allows you to shop completely hands-free while driving, for example.
How did the idea for your project come about?
The genesis of HF-SCA is particular and deserves to be told. The idea for the HF-SCA stems from a concrete need linked to an application that Vidyasoft had already created for Sella: voice banking. Through Google’s smart speakers, it is possible to find out the balance and movements of one’s current account, as well as making some small payments. Unfortunately, the need to use the SCA and therefore to have to physically take the mobile phone to authorise transactions thwarts the voice experience. Hence the idea of using artificial intelligence to understand whether a certain request is genuine or not, based on the user’s previous usage habits and data from the Internet of Things. Exactly 2 years ago, Luigi Manco and I – CTO and member of the Vidyasoft team that took part in the challenge – asked our SellaLab Lecce representative, Angelo Tafuro, if it were possible to get the transaction data. Unfortunately, the time was not ripe, not least because of the difficulties concerning privacy issues. So we kept HF-SCA in the drawer. Until a few months ago when we read about the challenge, which provided the very data we needed. So we reopened the drawer and now here we are: winners of the challenge, ready to develop our idea!
How will it continue now? Next steps?
Having won the challenge, a new phase awaits us in which we can develop and test a full version of HF-SCA together with the people at Sella. The data made available during the challenge was anonymised and partial, for obvious privacy reasons. Now that we have access to the full dataset we will be able to validate the solution in the field and publish our work on the Fabrick platform. Throughout this process we will update the Artificial Intelligence algorithm used by HF-SCA as in the meantime we have evolved it with an even better performing and state of the art model. This is thanks to the incredible work of our collaborator Benny Taccardi, data engineer and member of the team that took part in the challenge, who will soon graduate with a degree in HF-SCA-related subjects.
What were the most challenging parts of its development? How did you deal with them?
The most challenging part was definitely the performance evaluation of the algorithm. The dataset provided was partial and did not allow us to carry out a real validation, i.e. basically, we could not understand how “good” our algorithm was at distinguishing “typical” transactions from “anomalous” ones. Sella could not provide us with any more data, as this would have meant giving us an advantage over the other competing start-ups. In the end, we managed to calculate the score using a ruse, and unfortunately the result was not very exciting. So overnight, we completely changed the business model.
What did you like about the sella data challenge?
The weekly touchpoints. Every Monday we had 15 minutes to tell the Sella SWAT team about the developmental progress. This forced us to run marathons – even overnight at times – which made the solution mature quickly. In full co-creation perspective, first the SWAT team and then in particular our business coach Massimiliano Usuelli accompanied us, assisted and advised us every week for 2 and a half months, both for what concerns the UX and for what concerns the business side of our solution. Delving into this last aspect in particular was surprising for us too. Luigi Manco and I – CTO and member of the Vidyasoft team that participated in the challenge – conceived HF-SCA in a very early stage: about 2 years ago, focusing on UX issues for accessing financial services through smart speakers. The SWAT team of the challenge made us realise that HF-SCA not only improves the user experience, but also provides a strong increase in revenue due to more transactions being authorised in frictionless mode.
Keep up to date with news from FD Community and the Italian fintech world