How can middle and back-offices make use of artificial intelligence?
Matthew Davey shares his views on how middle and back-offices can make use of artificial intelligence.
Artificial intelligence (AI) has been around for a while now but industry interest has only recently picked up, why is that?
There is a focus on AI now because three elements have come together and created an inflexion point: computing power, sophisticated algorithms and vast amounts of data. But there is a danger that AI is a solution looking for a problem. Financial institutions need to focus on the business requirements of clients.
Chatbots, imaging and language processing AI developments are mostly geared towards the consumer market. Institutional Financial Services firms are not typically about image processing – what we are often looking at is a spaghetti of operational processes running at high volume on legacy systems. These operational processes do not always lend themselves to AI solutions, but when applied they can extend their longevity. The downside is that they can have the effect of fixing legacy systems in aspic and making further changes very difficult. At some point, you need a transformation strategy to build new systems. A key question facing large Financial Services firms is to what extent do you keep on extending legacy platforms versus building a new one?
So where do you see AI being most useful in your industry?
AI sits at the top of a curve that encapsulates all of data science, such as big data and machine learning. Many of the developments related to AI in financial services are focused on machine learning.
There are areas where AI is particularly useful, such as fraud detection. For example, AI can identify patterns in vast amounts of data and recognise potential fraudulent transactions more quickly and accurately than human staff. AI can also be applied in customer support, synthesising data across multiple systems internally and producing cross-selling opportunities or client sentiment analysis.
A new trend in machine learning is model interpretability; a model is good and very accurate, but when it makes a decision, we want it to explain itself. Many applications in banking which do not have large data sets can build graphical reports about the factors involved in decision making. For example, a customer was approved because their income was above x, the underlying volatility was below y. The next step from this is combining machine learning with natural language processing to automate report generation. A compliance dashboard, for example, can spot an item and ‘write’ to the compliance officer, raising an alert about a potential mis-selling, and listing the issues raised. AI is about complex decision-making processes, where it is making the decisions for humans to review in context.
Robotic process automation (RPA) has been disappointing to some extent, in that it has not yet delivered all that it seemed to promise. Setting parameters for RPA systems can be very time consuming and is highly dependent on any changes in the ecosystem. Ideally, RPA would be chosen for simple, high-volume use cases. It is possible that RPA on its own is not enough and it should be combined with AI and other technologies such as optical character recognition, natural language processing, et cetera.
What is the biggest challenge to the integration of AI in your industry?
Many of the IT techniques within financial institutions are inherited from legacy systems and this can be an issue when it comes to AI. A key challenge is how to plug AI into the existing legacy infrastructure. Initially, legacy systems may need to be simplified. At Societe Generale, transformation of legacy systems will be a step by step process based on a digital transformation strategy using web services. Outside of this, any innovative ideas from the business side will be met with machine learning technology when required. Legacy platforms will be transformed into agile ones. While we must push forward with new technologies such as AI and machine learning, we also must understand that we cannot transform everything. Agility will come from the use of web services everywhere and the transparency of the systems, whereby everything is catalogued. Replacing the legacy systems will take a long time and we will have to modify these systems to meet client needs and regulations.
Financial services firms are typically running complex operations on legacy platforms. AI tools can help extend their longevity, but will only take us so far; at some point you have to make the leap and invest in a new system.
Managing Director, Global Head of Business Solutions
Can you talk to us about any AI-driven solutions currently deployed at Societe Generale?
Within Societe Generale, we prefer to talk about machine learning rather than AI. The bank employs many data scientists to train algorithms that deal with various business cases. Our SG Markets platform is an example where machine learning techniques have been used to create a recommendation engine. Based on the history of trades with Societe Generale, clients are advised of the most relevant research papers to read. We have also used machine learning to analyse thousands of contracts to detect missing clauses or recommend new clauses in legal documentation for our clients. A ChatParser application for fixed interest and equities trading desks captures information from various chat systems as well as email, translating them into requests for quotes. This saves precious seconds for traders and improves our response time to clients.