Actualités

Artificial intelligence: promise, hope and probable disappointment in the financial services sector

In March 2016, the victory of AlphaGo, a computer program, over Lee Sedol, a professional Go player, highlighted one of the most important technological advances of the last 20 years: deep learning.

 

What impact will deep learning have on the financial services sector in the next 5-15 years?

 

Here, we offer some ideas to spark debate in your company and find the right path to adopting this new technology.

 

Conviction #1: the media buzz generated by artificial intelligence (AI) promises more feats like the AlphaGo win that will not materialize any time soon.

 

AI refers to a very wide range of programs in terms of sophistication and ambition. We can distinguish between “weak AI” and “strong AI.” Weak AI is not intended to copy human intelligence and reasoning but simply imitate them and reproduce human behavior as faithfully as possible. Strong AI, however, aims to not only produce intelligent behavior but also emulate its way of working. AI combines several innovative technologies, including expert systems, and self-training systems such as machine learning. Basically, the more AI integrates machine learning, the stronger it becomes. Programs based on machine learning adapt and improve over time, as they are fed new data. Deep learning is a form of machine learning, or automated learning methods that attempt to model data with a high level of abstraction using architecture based on non-linear transformation.
Deep learning occurs when a human can no longer explain the choices made by this type of AI.
The success of AlphaGo comes from deep learning. Yet the majority of AI applications currently involve weak AI, which implies the promises made by the media cannot be kept in the timescales announced.

 

Conviction n#2: despite not living up to media hype, AI technologies should continue to attract interest.

The potential applications of AI for banks and insurance companies are numerous: chatbots or conversational programs, regulatory compliance, fraud detection, client behavior analysis and market trends forecasting.
The rise of chatbots addresses the constant rise of client contact by email, which has already made a significant impact on bank managers, sometimes to the detriment of their “sales” efforts. CAI technologies can now understand written requests and reply to the most common ones without human intervention but with complete transparency for clients.
Another example is “intelligent” sales scripts that can create AI by adapting to client data to personalize dialogues and improve the performance of outgoing calls.
These AI technologies are already operational. Unlike deep learning, they do not draw on advanced modeling approaches. However, they will no doubt drive a major evolution in the client experience in the financial services sector.

 

Conviction #3: some of the most talked-about technologies can bring value unless they distract banks from deeper issues.

Journalists are paying particular attention to robot process automation (RPA), which automates low-value manual tasks without requiring a transformation in the processes or IT systems that support them. However, the benefits of these technologies are limited and they should not distract banking and insurance leaders from the real need to transform their processes and IT or fail to acquire the agility required by digital transformation.

 

Conviction #4: the most sophisticated AI technologies will have a very strong impact in the banking and insurance sector, but this will take time.

It is not particularly relevant to use AI to improve investment decisions since this involves large volumes of heterogeneous data (both structured and unstructured), near real-time analyses and recognition of complex patterns.
In a context of low interest rates, directionless stock markets and prevailing cost pressures, there is a pressing need for “alpha-generating” ideas.
AI also offers very powerful tools to help financial institutions meet regulatory constraints. For example, a machine-learning system that has been fed traders’ transaction data will develop detailed negotiating profiles for each of them and be much more accurate at identifying any suspicious behavior that deviates from their usual patterns. The suspicious transaction and order reports (STORs) imposed by EU Market Abuse Regulation constitutes one potential application.

 

Conclusion: before launching a “strong AI” project, banking and insurance companies must first identify the relevant business applications. The investment required is significant but essential to ROI.

 

First published in Point Banque