Publications

Artificial intelligence, an opportunity for financial and insurance industries

The first step in implementing artificial intelligence is data collection. The financial sector is experienced in this, data is abundant and secure.

The next step is to process the data to make it more valuable. In this way, AI makes all its sense, using big data to support workers in this sector to perform certain specific tasks, which will be outlined in the rest of the article.

A common application of AI across the three sectors is the practice of personalizing offers through advisory activities.

Among other things, AI is very effective at identifying recurring patterns and therefore similar customer profiles. By learning the types of offers that work best for a given profile, AI is able to make personalized suggestions that best suit the customer. This innovation is particularly relevant for wealth management, asset management and investment advice.

Going further, in the financial sector, personalized offers and customer services can be combined with credit scoring, i.e. the use of artificial intelligence to help lenders make credit decisions by determining the creditworthiness of a person. The work of evaluating a customer's file is carried out automatically according to a set of criteria that can be learned by ai models.

AI is also used in the three sectors to perform support tasks, i.e. to automate low value-added tasks which can facilitate the work of advisors by automating the processing of documents, filling in certain parts or requesting additional information, therefore saving valuable time for employees

AI Algorithms can also assist humans in making decisions. In fact, by being able to analyze large amounts of data, algorithms can gain knowledge about information that is of great value and can influence decisions. For example, after a series of analyses, the algorithm can make recommendations. The aim is not to replace humans, but to support them, in particular by assisting them in tedious tasks with little added value, such as sorting cheques in the banking sector (2).

A major challenge for AI in finance and insurance is fraud detection and preventing it is a real battle. Losses in the financial markets due to fraud or operations outside accepted limits have been in excess of $40 billion over the last decade (3). 

To give a rough idea, the Association of Certified Fraud Examiners (ACFE) estimated fraud at $3.5 billion in 2011, or an average of 5% of global revenues. Fighting fraud is therefore a major challenge, and with the analytical capabilities of AI, it is possible to imagine automated fraud detection solutions that will assist specialists in their prevention efforts. 

In fact, AI is effective in detecting behavior that is different from usual behavior and, above all, it allows for the simultaneous analysis of very large numbers of files. 

In the banking sector, for example, Mastercard uses machine learning algorithms (4) to "verify up to 160 million transactions in milliseconds, applying more than 1.9 million different rules to examine each transaction". In this way, artificial intelligence can detect money laundering activities, extortion attempts or indicators of bank code theft.

For the banking sector, artificial intelligence is also being used to strengthen cybersecurity capabilities, including detecting attacks against financial institutions. While hackers use AI to improve their attacks and level the playing field, AI can also be used to protect against cyber-attacks. In a sector where security is key, especially for data protection, using AI as an additional tool in the defensive arsenal is a notable advantage (5).

Another strategic use of AI is in risk management activities. AI helps to improve risk control elements, whether they are preventive, detective or corrective. AI can be used to conduct surveillance, analyze documents and extract relevant information. It can also perform trend analysis to identify specific risks, such as declining market volumes, resource shortages, impending financial crises, etc. Alerts can then be automatically triggered when alarming signals are detected. AI saves time and can reduce interpretation errors, ultimately leading to better risk management.

In the insurance sector, AI can also be used to support claims management. It is possible to automate claims analysis, compensation, prediction and even reporting; AI optimizes the entire process. In this way, insurance companies can reduce their costs by spending less time and fewer expert visits. AI also supports these experts by enabling better data analysis and increased accuracy in claims management (6).

Finally, AI can be used in the form of "robots" that automate tasks for humans. In banking and insurance, customer relationships can be partially managed by artificial intelligence, particularly through the use of chatbots. In finance, financial trading activities such as buying and selling stocks can be automated using trading programs. In both cases, the human can be monitoring and can step in at any time to take control or, conversely, let the algorithm do the job.

AI can have many applications in the financial sector, but many of these applications raise questions about the safety and trust placed in these systems.

For critical sectors where security is essential, artificial intelligence remains a technology with many advantages, but also disadvantages that must be managed (7).

First, AI algorithms must respect a number of principles, including ethical values such as non-discrimination. However, AI algorithms are often subject to bias problems, which can affect the operation of algorithms by causing them to consider irrelevant parameters. There have been cases where bias has caused AI algorithms to become discriminatory. For many, AI algorithms are then a matter of trust. Biases are often hidden in data sets, even if they contain high-quality data. 

To increase trust in AI systems, it is therefore necessary to be able to perform tests to detect these biases and to evaluate the robustness of algorithms in the face of perturbations. It is also necessary to be able to understand the basis on which the algorithm makes its decisions. Currently, a large number of AI systems behave as "black boxes", i.e. their operation cannot be explained. Explainability is therefore a dimension that AI developers must work on to ensure that algorithms make the right decisions and, above all, for the right reasons.

Robustness and explainability therefore seem essential to limit risks, especially ethical risks, and to build trust in the implementation of AI in the financial, banking and insurance sectors (8).

However, there is another difficulty in implementing AI in these sectors. This is because the decision-making processes in these environments are very complex, with a large number of parameters that can be colossal. Although AI algorithms can handle a very large amount of data, the number of parameters to be processed must remain relatively small; as the number of parameters increases, the performance of the algorithm decreases (9). 

This observation refers to the principle of data dimensionality and the curse of dimensionality, which affects all machine learning algorithms. One technique to overcome this problem is to reduce the number of parameters by selecting the most relevant ones. However, this exercise is not straightforward for the financial sector, which has a large number of parameters and requires very experienced data scientists (10).

Finally, there is one last risk associated with the use of AI in finance, banking and insurance. This concerns security issues and, more specifically, cybersecurity. This difficulty is inherent to the sector, which has historically always been vulnerable to attacks. The information held by the financial sector is critical, and the economic impact can be very significant, leading to a stock market crash. 

Protecting algorithms is therefore a cyber-defence challenge against hackers with malicious intentions to redirect algorithms or cause them to malfunction. However, a machine learning algorithm can be difficult to protect because, in addition to the code, the data also needs to be protected. This is where the robustness of the algorithms comes in again, because it can allow us to face disturbances that would be created voluntarily. The security of AI algorithms is therefore an important issue because it represents an additional risk in terms of their implementation.

The challenges posed by the implementation of AI in finance, banking and insurance are recognised by these organizations, which are already working to address them. For example, the ACPR, the Prudential Control and Resolution Authority integrated into the Banque de France, organized a hackathon in 2021 with the aim of explaining the behavior of 'black box' predictive models (11). 

Banks can also fund other initiatives, contributing to the development of the financial services market and facilitating the uptake of financial innovation. The BPI, the public investment bank, is also contributing to the growth of this sector and the resolution of these challenges by launching calls for projects, such as the Finance Innovation calls, aimed at the development of fintechs (12).

Although the challenges of using AI are recognised, the implementation of these algorithms remains a major challenge for the finance, banking and insurance sectors, despite the many benefits. These challenges are related both to the nature of AI algorithms and to the constraints inherent in these sectors in terms of functionality and security.

In conclusion, artificial intelligence is assisting employees in various support, management and protection roles for the finance, banking and insurance sectors. From automating repetitive, low-value tasks to fraud prevention or decision support, AI is multidisciplinary and can effectively transform these sectors. However, as always, AI raises issues and challenges that need to be addressed, especially in the financial sector where data security and privacy issues are even more critical. For example, the robustness and explainability of algorithms, as well as cybersecurity, need to be given particular attention.

As the implementation of AI grows rapidly, other technologies could be combined with it to shape the future of finance, banking and insurance. Blockchains and quantum technologies are among the technological investments that could also benefit these sectors. 

Blockchains highlight their ability to secure operations, with the potential to automate financial transactions. Combined with AI, it would be possible to imagine fully automated contract processes. 

For quantum technologies, the main interest would be to increase computing power and operational capabilities. These components are very important when coupled with AI and the data and processing volumes of financial services.

Therefore, the technological challenges for the finance, banking and insurance sectors are growing rapidly, but the need to secure these technologies to enable wider implementation remains high.

(1) Etude ResearchAndMarkets et Etude MordorIntelligence

(2) Intelligence artificielle assurance service satisfaction client

(3) Prévention de la fraude

(4) IA dans la finance

(5) Challenges de l'IA dans la finance

(6) IA dans le secteur de l'assurance

(7) Enjeux de l'IA dans les institutions financières

(8) Ethique et IA dans le secteur bancaire et financier

(9) La malédiction de la dimensionnalité

(10) Opportunités et challenges de l'IA dans la finance

(11) Hackathon ACPR

(12) AAP Finance Innovation BPI

(13) IA, technologies quantiques et blockchains pour la finance

Written by Baptiste Aelbrecht & Camille Jullian & Jacques Mosjilovic.