Technology

How BNP Paribas Personal Finance’s DataLab builds a transversal data model

Better known to the general public under the name Cetelem, a European consumer credit player, BNP Paribas Personal Finance launched a DataLab in 2017. The BNP Paribas group entity has an atypical positioning although it is focused on a subject whose mastery is now recognized as fundamental for almost all organizations: data.

Three years ago, BNP Paribas Personal Finance positioned itself on this subject identified as strategic for the years to come.“, explains Jérémie Guez, DataLab manager at the company’s IT Services department. Three people are assigned to the launch of this laboratory, which now has around thirty employees, and which aims to implement a real technology strategy of large scale based on data.

A private cloud and no sharing of intellectual property of technologies

The DataLab adopts a model that draws on the projects carried out by the employees of the BNP Paribas group, ie 20,000 people in more than ten countries. “Our vision is to create value from our data and our way has been to favor internal investments rather than solicit external providers“, he continues. The structure creates our own tool for its data platform, with the aim of not using a third party.”Many tech companies in this area have been bought out by giants, such as Dataiku (which counts as a Google investor, editor’s note) or Domino Datalab (Dell)“, adds Jérémie Guez.

For example, customer data is hosted in a private and internal cloud in order to preserve the confidentiality of the data and the entity does not share the intellectual property of its technologies.

Marketing and scoring tools

Concretely, the data is used there for several service offers, marketing or fraud for example, for use cases meeting the needs of BNP Personal Finance, which is dedicated to the financing of personal projects: the detection of fragile customers and increased vigilance in a responsible credit approach, advocated by the banking subsidiary.

“We work for different professions: those of marketing, with the optimization of campaigns and client websites but also on scoring solutions to fight against fraud and assess risk and grant“, sums up the manager, who adds that this transversality contributes to this atypical positioning.”Data scientists can play with a very rich data panel, even if we must not lose sight of our role which is above all dedicated to support and expertise.“.

The emphasis is on projects fueled by artificial intelligence technologies with a priority on unstructured data, particularly text recognition and comprehension. “The use cases appeared quickly and in number, continues Jérémie Guez. This concerns the treatment of emails, chatbots but also search engines“.

Another topical solution is voice recognition. “This is a question that we ask ourselves but it was not deemed necessary to do it for the moment internally because the voice has less added value in our verticals. But it could be of value for call centers, and language processing, especially on the transition from oral to written“.

An automatic email sorting project

The DataLab is currently finalizing a project combining artificial intelligence and robotics technologies to automate the classification of emails without losing relevance, in order to relieve operational teams. Put into production in February, it aims to help the credit request reception centers. These must process a very large quantity of incoming requests that must be categorized and dispatched to the teams concerned. One person is therefore dedicated solely to sorting the approximately 1,250 electronic messages received each day. “Very time-consuming, repetitive work with little added value, and the risk of making mistakes“, we explain to DataLab.

The tool allows initially to retrieve all the emails sent to this team and then according to the specialization of the request, to sort them, for example those mentioning a missing attachment or relating to the revaluation of a contract. The robot takes the information and directs it to the correct category.

After having tested the efficiency of sorting, the second step is to be able to carry out automatic response actions. If the data categories are sufficiently precise, this generates a response. “The challenge for us is to categorize 5 or 6 good boxes, no more, announces Jérémie Guez. We want to keep human interactions. It is important not to chain categorizations, but to use this saved time to go to more and more services for our customers and partners. “

A “large AI community” within the group

If the DataLab does not share its data, it shares its algorithms to improve its technologies. Its various projects coexist with the other Labs working at BNP Paribas. “We have a large AI community at BNP Paribas which allows us to know what is going on in the different labs, explains Jérémie Guez. Thus, we were able to meet several labs including that of PACE (Partners in Action for Customer Experience, editor’s note) with whom we have been collaborating since. We also collaborate with BivwAk! (the internal incubator launched in 2018) on our projects“.

The different structures currently have two projects in common. The establishment in the medium term of a common data science platform, in order to capitalize on the respective expertise to make this “innersource” platform a “BNP Paribas strong asset“and, in the short term, the acceleration of the deployment and packaging of BNP Paribas Personal Finance’s” homemade “data science platform.”With PACE, we are collaborating on a common brick of artificial intelligence, the “Sentiment Analysis”. Initially, we worked with PACE in parallel on this brick. Then thanks to events in the Group’s AI community, we decided to discuss our best practices and our developments to make it a more efficient building block benefiting from the strengths of each entity.“.

Today, the DataLab claims more than 200 users and also works internationally. “The scalability of our projects lies in their development in other languages“, concludes Jérémie Guez. Internally qualified as a success story for the company, the model is of interest to other entities as well as the partners of Personal Finance, including large French distributors.

Aude Chardenon

@ChardenonA


Back to top button