What is Artificial Intelligence doing today for the financial sector?

The Official Chamber of Commerce of Seville, in collaboration with the Spanish Institute of Financial Analysts (IEAF), offered last March 16, 2022 a conference under the title "What does Artificial Intelligence do today for the financial sector?"

The presentation was given by Fernando PavónGamco's CEO, who in a didactic and entertaining way, gave a brief overview of the world of AI and its application today in the financial sector. 

Here is a summary of the day's events:

Uses of Art Intelligenceificial in the financial sector

In the field of portfolio management, Carlos Jaureguízar, for example, has applied to investment analysis filtering of correlation matrices of sample covariances using the deep learninga model that mimics the neural networks possessed by human beings. 

Another example in the financial sector can be found in the company GVC Gaesco, which has patented a model in which AI based on deep learning manages to find portfolio management solutions applied to behavioral finance. This has been called Behavioral Finance. To preserve capital, what it does is to discriminate situations in which a piece of information is missing or is not decisive to know whether an asset is going to go up or down, so that these neural networks are able to know where the breakpoints are.

To contextualize how AI works in the financial sector, we could go back to the figure of Santiago Ramón y Cajal, Spanish researcher and Nobel Prize winner, who first identified and analyzed between 1920 and 1930 how nervous systems functioned in living beings, revealing this perfectly identifiable structure, formed by cells called neurons. Since then, this structure of neuronal networks has been emulated by means of AI. In short, its main objective is to develop artificial neural networks.

Artificial Intelligence Classes

At a conceptual level and depending on the level of intelligence achieved, we can identify three classes of Artificial Intelligence:

1. The "real" AI. It focuses on a single task exclusively. Thus, if an AI were to learn to play chess, it would not be able to drive a car.
2. General AI (AGI - Artificial General Intelligence). It has the ability to think, so it does not present a fully robotic driving. This allows it to adapt to different scenarios and solve problems in a way that is completely comparable to how a human mind would.
3. Artificial superintelligence (ASI - Artificial Super Intelligence). It is the one usually shown in movies, where AI has achieved the ability to be autonomous and superior to human intelligence. AI not only replicates human knowledge, but can also think better, faster and more skillfully than a person.

However, in AI focused on the financial sector, and in general to any sector, we can find three principles:

1. Data. They are our raw material and allow us to extract useful knowledge to make predictions by means of recognized behavioral patterns. In this way, we will be able to deduce which are the most important variables of our client.
2. The facts. From the data, AI infers relevant knowledge that usually takes the form of: predictions, estimates of the occurrence of a given event (non-payment, purchase, abandonment of services by customers, etc.) and segmentations of customers and their behaviors in order to know them better, predict behaviors and be able to carry out simulations: answer questions that in English would be of what if?
3. Action. The knowledge acquired and the predictions must be put into action: use them in the day-to-day running of the companies. This action is carried out by means of software resources that allow us to take information to the right person at the right time, because if we manage to predict behavior but do not take action, we will not really be improving anything. This action should optimize business objectives while improving the services and products given to customers.

In short AI creates Attendees. They allow a company's human talent to be freed from those tasks that can be automated. In this way, a worker will have more time to perform the tasks for which he/she is better qualified and that today the machine cannot do, for example, having a coffee with a customer to learn more about his/her needs and motivations. 

For AI to learn about a particular problem, it needs to work with data. For example, predicting demand to buy stock requires time series sales data. 

However, if we have a very broad portfolio, the machine will have a greater amount of data to process and resolve. In most cases, it is not necessary to go to the massive data (Big Data), but it is better to go to the "small data" (the data that the company already has) to obtain more fluidity in the analysis and even better results, because otherwise it could take several months to analyze the data of a company. Therefore, it is advisable to start by asking only for the data that companies usually have, such as turnover data. CRM and ERP. 

Main problems of AI

The 3 main problems to which AI is applied are prediction, classification and segmentation.

Let's take a look at a practical example from the financial sector:

If what I am looking for is to predict income to manage my liquidity, it is essential to look for temporal data in which the time factor is very important. That is to say, it is necessary to determine at what time to predict income, and how far in advance to predict it and, when the time for which it was predicted arrives, to evaluate the prediction with reality.

The objective is to analyze how good or bad it has been, and for the learning algorithms to readjust according to the goodness of the predictions. For example, for a bank it is necessary to predict the cash needs and to know what cash will be available in the near future in its branches, ATMs, conveyors, central bank, etc... This could be applied to any company to know what cash will be available.

The same applies in the area of classification, but it already refers to the payment or non-payment of an invoice or the amortization and interest obligations of a loan. The objective here is to classify future behavior, for example, the typical scoring credit admission (good or bad). 

In terms of segmentation, the important thing would be to determine what types of customers I have and what are the most important variables. The level of complexity here is higher, because you have to have more knowledge of the business. In this case, what AI does is to analyze the information and present it, for example in the tools we have developed, through behavioral maps, so that we can implement methods to extract behavioral patterns with typical variables and provide much more information, such as the characteristics of the company and the most relevant individuals for the business.

The most important thing is to collect the variables on which you tend to move within each segment. In the example below, each cell corresponds to a certain behavior, composed of its own characteristics, customers who exhibit this behavior and differences with neighboring segments:

Here are some examples of applications in the financial sector:

• Credit ➜ Granting decisions
• Risk ➜ Management of the risk admitted by the business with its customers.
• Trading ➜ Automate trading.
• Advisors ➜ Client-to-client customization of recommendations, monitoring and anticipation of trends.
• Fraud ➜ Security against errors and misuse.

We have to be aware that AI is not only intended for large enterprises. Because science and technology are quite developed in this area, we can also to put it into practice in small and medium-sized companies. 

The implementation of AI solutions in companies through cloud technology will facilitate their deployment, control by the business and interconnection between company departments.

An example of the application of AI to small and medium-sized companies can be the prediction of defaults by currently healthy customers (who have not given default problems). This is a development that we have been working on for years in large banks for asset portfolio management, and which is extremely useful for optimizing customer payment management in SMEs. 

In many SMEs we can find a reactive part in the system where the treasury, debts, debt recovery, VAT reclaim of unpaid invoices... are managed.

But there is also a proactive part where the problem that a customer may have must be foreseen. At this point, it is convenient to keep in mind that, although there are customers who have always paid, there are certain behaviors (reflected in the values of the variables) and indicators that serve AI-based systems to give alerts and warn us that these customers may not pay.
All this can be solved thanks to the ARM SaaS software of Gamco for the default prediction. ARM SaaS allows you to detect those customers who present a high risk of non-payment. In this way, you will be able to minimize your business losses, as well as reduce the time for decision making.