About
Crédit Mutuel Arkéa is using BERT neural networks – the latest advance in NLP – to develop an in-house tool to sniff out controversies. This step will firmly embed this issue in business processes, make work easier for employees and help us meet our CSR commitments.
Innovation presentation
Corporate Social Responsibility (CSR) issues have an important role to play in our society. The Crédit Mutuel Arkéa group is committed to CSR, not only to maximise compliance with regulatory requirements but also because it seeks to have a positive impact in the regions where it operates. For instance, the group launched its first social bond in 2019. The terms of the bond committed the group to starting a project on detecting CSR controversies in entities to which it provides finance.
To embed this new issue into business processes, the Innovation & AI Department and the CSR Department have developed a “search and alert” engine for controversies, which uses the latest advances in natural language processing (NLP) – namely BERT, a neural network developed by Google researchers the end of 2019 – to analyse social media.
Based on APIs, the search engine has been designed so that it can be built into our core business tools, provide a unique user experience for employees and translate the group’s CSR aims into concrete action.
Each time the group assesses a new business partner (a service provider, a supplier, a customer, a portfolio company, etc.), users are thus alerted in real time to the potential controversies that affect it, enabling them to make an informed decision.
The search and alert engine is flexible: employees can define which social media sources are relevant in their particular case. It can also be customised to meet the needs of the different business lines, each of which may have its own definition of a “controversy”. The CSR Department therefore participated in the overall configuration of the tool, while the business lines will help to refine and train it more precisely later on.
The tool thus fulfils several aims, namely:
- Helping us fulfil our CSR commitments and comply with regulations, while reducing the risk to the Arkéa group’s image, brand and value.
- Providing customised analysis
- Making that analysis more reliable
- Speeding up this time-consuming work, while ensuring that humans continue to take the decisions
More and more, CSR is a universal issue that applies to every business line in the Group. The design and architecture of the tool reflect this cross-functional relevance, making it adaptable to all of our entities.
Depending on its success internally, it could also be made available to our partners at a later stage.
Beyond the significant time saving in reputational research (a direct ROI), the reduction in the risk to our image, although difficult to measure, also represents a very significant indirect ROI.
Uniqueness of the project
In this project, both the issue and the implementation are unique :
While many innovations focus on environment aspect of CSR, very few deal with the Social part, and specifically controversy seeking.
Several solutions on the market allow e-reputation analysis, but those do not correctly meet our needs:
- they are not specific to CSR controversies
- they do not cover local media, which are very important for the Crédit Mutuel Arkéa Group, which is anchored in the regions (our head office is not in Paris)
- they are very focused on English, and little on French languages
- they do not allow fine customization, including the definition of controversy itself.
From a technological point of view
- The production use of these cutting-edge AI technologies is innovative in banking. BERT was chosen because it is one of the most recent, and the most efficient models of representation of textual data, pre-trained in French.
