Finance AI,Artificial Intelligence in Finance – AI banking

Finance AI technology and uses,why AI banking is the future?

Artificial intelligence has become one of the buzzwords of today. True, not as pompous as pre-crisis hedge funds, today’s Bitcoin blockchains and startups. But really: everyone is talking about it, although very few understand what lies behind the meaning of this buzzword, i.e. what is artificial intelligence really.

Artificial intelligence is already being used by banks to provide services to customers and improve business processes. But the heyday of this technology may be yet to come. According to a study by the Fintech Association and Accenture, which conducted a survey of banks, AI will be used for the next generation of financial services. Survey participants noted that they need to develop competencies in the field of AI and other related technologies (machine learning, collecting and processing big data, open APIs, and so on).

Working machine learning

It can be argued that now artificial intelligence does not work yet. Well-programmed talking robots can, however, beat the Turing test. used to distinguish between human-simulating computer code and a human, but if you take the time to ask them in more detail, it will become clear whether it is a robot or a human. What really works is the technology behind the phrase “artificial intelligence” – machine learning.

If artificial intelligence is both a paradigm and an abstraction, like the term “vehicle” in the case of a car, then machine learning is a concrete, operational technology, such as an internal combustion engine. And those who are interested in artificial intelligence want to know what is this internal combustion engine that makes the vehicle move.

At the anniversary of the finance department of a large local government, we asked asked to explain whether artificial intelligence consists of regressions – a well-known analysis tool for financial data analysts – although in the case of artificial intelligence, the mentioned regressions are performed so much that the results are “smarter” than conventional analysis data.

The first answer is always wrong

This statement is not far from the truth, although there are many more details hidden behind machine learning. Machine learning algorithm based on the so-called. artificial neural network, works like a real nervous system: it consists of many neurons – perceptrons that can perform numerous mathematical operations, including regression, although individually each of them is not even able to “think” about something. Nevertheless, acting together, they can always offer at least some solution to the issues that are put before them. In other words, the answer to a random question posed to an artificial neural network is always wrong (from the point of view of a machine learning algorithm, such an answer is denoted by a special term: dummy).

In order to understand anything, an artificial neural network must first learn

The neural network needs to know which answers to similar questions used to be correct. If the set of correct answers, i.e. training material is available, training begins, an important part of which is the back propagation of a learning error, or backpropagation: an artificial neural network measures how much the answers it offers differ from the correct ones, and, in accordance with this, reconfigures itself. After that, the neural network tries again to find the answers and measures the deviation of the results of the new attempt from the correct answers.

Thus, this set of artificial neurons repeats the cycle of answer-measure-reconfigure until its answers begin to coincide more and more with the correct answers. By this point, she has trained herself and can get to work: give recommendations, make decisions for people. There are other technologies behind this, such as Gradient Boosting, Learning Rate, Activation Function, etc., but, in general, this is how it works.

A living organism behaves similarly to an artificial network: once a person gets burned a couple of times, a very stable reflex will be instilled in him for the rest of his life – to stay away from sources of fire.

Why do banks use AI?

AI in banking accelerated access to products for many customers and automated some stages of internal processes, which also affected the speed of service,another reason: cost optimization

Finance AI – Artificial intelligence is useful

To get an overview, in the early spring of last year, we robotically downloaded an overview of 327 data analysis competitions from the Kaggle data scientist platform. we then sorted these contests into well-known topics using a Wordnet-based text analysis algorithm. It turned out that the largest cash prizes for data scientists ($100,000 or more) were for algorithms that can recognize the location of mineral deposits, give answers to medical diagnostics, and understand satellite and other photographs.

More than a million were received by those data scientists who managed to predict the market price of real estate. From Kaggle, it turns out that in real estate valuation with the help of artificial intelligence, very good results have been achieved – in terms of the most successful works. We are talking about real estate appraisal without inspection of the appraised object.

A shift to using such algorithms – and they are fairly easy to develop – would jeopardize the jobs of many real estate appraisers in Estonia.

Among the recipients of millions or more prizes was also predicting the degree of danger of passengers entering the United States, which also gave very good results. However, medicine undeniably dominated Kaggle with several big prizes, of which the millionth was awarded to the data scientist team that best diagnosed lung cancer (the prognosis was not so good this time), and half a million to those who predicted the overall incidence of healthy people. people (and this attempt was not so successful). If the two big-prize competitions cited here give the impression that machine learning in medicine is not very effective, then this impression is false.

Business needs

Estonian companies are practical and willing to implement proven machine learning and statistical analysis solutions. Such, for example, are recommendation systems, i.e. recommending new products to customers based on their previous purchases and the history of other customers who have made similar purchases, and in finance, risk assessment algorithms.

A very common and highly sought-after application of machine learning is focused on understanding photographs. An example is not only the location of a car in the parking lots of large shopping centers, which is familiar to us, but also numerous government services, including those that are not talked about much. Did you know that government authorities have known for years when you travel abroad and return to your home country? Even when you go to a country with which there seems to be no border? True, there are poles near the former state border, and there are cameras on them that record the numbers of all passing cars.

Practical applications

As already mentioned, in second place after medicine in Kaggle are forecasts in the field of economics and finance. Large prizes were offered for algorithms for automatic real estate valuation, as well as for predicting financial results and the behavior of securities. We also ordered forecasting of credit risks of a loan issued to borrowers. The results were both good and average, and rather poor.

When we said that the intelligent robot did an excellent job of valuing real estate, this only applied to the markets of the Western world. Thus, when predicting real estate prices in the United States, a zero error rate was achieved. But at the competition for predicting the price of real estate in other country may differ. 

In the case of the financial sector, we cannot provide comprehensive data for Estonia, with the exception of the public sector, where the situation is mapped by the Ministry of Economics and Communications and adopted as a so-called. Kratt program for the development of artificial intelligence solutions. For many other countries, reviews are also available for the private sector.

In the UK, according to the latest data, machine learning solutions are used by almost two-thirds of the banking sector, half of the enterprises providing investment services and enterprises offering support for financial services, a quarter of non-banking lenders and absolutely all insurance companies surveyed.

The same study found that the most popular application of machine learning was risk prediction and management, followed by customer service and intelligent service offerings.

Finance AI technology AIl Use Cases

1. Customer scoring : automatic decision making on customer applications for loan products. “Previously, an application for a loan from a large business was considered for two or three weeks, and this took the time and effort of many different specialists. Now, when these applications are considered by AI, it takes no more than seven minutes from the client’s request to the receipt of money. Everything happens remotely, without the use of paper documents, and the percentage of delays has decreased to almost zero.
2. Voice assistants and chat bots : are used when a client contacts a call center or a bank chat to reduce service time and optimize the work of employees. As for the chatbot, it processes over 40% of client requests and saves the bank more than $200 million per month.
3. Anti-fraud and financial monitoring : AI is used to counter financial fraud by analyzing atypical behavior of individuals and legal entities.
4. ATM Maintenance : AI predicts ATM occupancy and reduces cash collection costs.
5. Document processing . Using AI, automatically processes and enters customer data when opening accounts and performing banking operations where identity verification is required. “Artificial intelligence recognizes more than 70 details from scans and photos of documents for each client in 2 seconds and performs about 15 automatic data checks.

New opportunities for AI in banks – Ai Banking

From customer risk assessment to service personalization and emotion analysis.: A couple of years ago, credit scoring, risk assessment, and customer support bots were the main applications of AI in a bank. Now – personalization of client experience and assessment of client emotions when servicing in a branch.

The system based on machine learning technologies recognizes behavioral patterns in the client’s transactions and his interests in the bank’s products and services in a mobile application almost in real time, explains the way to create personalized offers Bagiyan: “Based on this information, we remain in the context of the client’s life circumstances and offer really relevant product. For example, a sharp increase in spending and a credit score request can be markers that a customer is interested in a loan. And a client with free funds who viewed a story about investments Online can be offered an investment product.

With the help of recommender models, Bank can create personalized recommendations to clients: for example, it can remind you of purchases that a client usually makes at a certain time, or, seeing that a client enters a PIN code incorrectly, promptly offer to go through identification and generate a new one.

AI should become an assistant to the client and at the same time be in a channel convenient for him.Following this logic, A chatbot can be launched based on neural networks and machine learning not only in Internet and mobile banking, but also in Telegram and WhatsApp messengers.

Determining where to open branches : new location intelligence technology to manage a network of branches.This technology aggregates data on all branches and subdivisions of the bank, assesses the potential and workload, calculates the effectiveness of potential offices based on data on the activity of clients, competing banks, population, traffic on the city streets and other statistical information. As a result, the bank has a “thermal” map for each city of presence with an assessment of the potential location of the branch at the level of walking distance (100 m).

Determination of the best working hours for employees : AI for scheduling employees who are engaged in sales uses “Opening”.

AI for scheduling employees who are engaged in sales uses “Opening”.Some work better in the morning, someone, on the contrary, in the evening. The AI-driven system evaluates sales performance and plans the employee’s schedule in such a way as to increase its efficiency.Using same AI method, monitor its advertising platforms, which has greatly increased the effectiveness of advertising campaigns

Barriers to using AI

The main barrier to the development of AI technologies today is the ability to collect and exchange depersonalized data to learn solutions in compliance with all norms of the law and protect citizens’ data

Most of the banks surveyed and found the lack of qualified specialists the main barrier: data scientists (they know how to convert large amounts of data and apply them to solve specific problems), specialists in AI, data analysis and machine learning, etc.

The market now requires 6–7 times more data scientists than it required 3 years ago.

You should not treat AI as a magic wand that solves problems on its own – building high-quality models requires deep expertise of the team in understanding algorithms, tools, and banking processes. This barrier can be overcome by investing in the development of employees and training programs for beginners on the side of companies.

Another equally important barrier is the availability of a trusted infrastructure that will allow you to create and train AI solutions.

Risks of using AI

The professional community, both scientific and industrial, is beginning to raise questions related to the safety of the use of artificial intelligence, the ethics of its use, and the economic consequences associated with AI. For example, about what to do so that automation of work with the help of AI does not lead to job cuts? How to fix the fact that algorithms can be subject to bias? Artificial intelligence must be objective, fair in making a decision, controlled by a person. For example, when making a decision on issuing a loan, AI should not rely on the characteristics of a person’s gender, race and nationality.

The risks include the exit of systems out of control and causing harm to a person and society, misunderstanding and unpredictability of the actions of algorithms, insufficient stability and reliability of decision-making systems. It is often difficult to figure out why the AI ​​chose a particular solution. This can cause distrust in systems using artificial intelligence technologies.

The result of the work of AI models depends on a significant amount of data from various sources, including external ones. This increases the requirements for information security systems to compensate for the risks of intentional distortion of input data

Artificial intelligence in banks

New generation banking platform :

Chatbots and roboadvising

Modern chatbots can:

• Informing about the features of products and services
• Providing contact information
• Carrying out payment transactions
• Financial advice to the client

• show rates and exchange currency
• manage personal finances
• transfer from card to card
• send applications for merchant and Internet acquiring and check the counterparty by TIN
• answer user questions

Robo-Advisers as a promising example of AI application

Robo-advising has become an alternative to financial consultants on banking issues, specific purchases and other online money transactions.

Robo-advisers provide great advantages in the field of online trading. First of all, these are one-click applications and real-time account opening, monitoring, current news and processing of large volumes of transactions at once. The distribution of brokers in social networks makes investment knowledge more accessible and understandable, and communication with a client is simple and targeted.

Automation allows you to present information 24/7, while reducing process costs. Robo-advisers are available on the desktop or as mobile applications and carry the functions of a portfolio manager who determines risks and the optimal investment strategy.

Individual offers and loyalty increase

• Recommendations for banking products and purchases ( loyalty programs from various retailers), including using customer knowledge from social networks
• Determination of B2B relations of the client with subsequent recommendations of new contractors
• Modeling financial risks for small businesses (default, cash gap) in real time with recommendations for targeted strategies and products

IoT (Internet of Things)

• Managing and tracking the use of leasing assets
• “Smart” insurance for retail clients ( medicine , car loans)
• Smart Home + Daily Shopping: order groceries, pay utility bills, subscribe to TV content

Antifraud . External and insider threats

• Signs of using a client’s plastic card by a third party
• Signs of the so-called. “droppers” based on the nature of receipts and transactions in the Internet Bank and ATMs
• Identification of fictitious salary projects (loans, cashing out)
• Identification of unauthorized debit transactions on customer accounts and customer plastic cards
• Errors in the parameterization of bonus programs for plastic cards, which lead to “cheat” and damage
• Cash-out schemes, incl. using Internet banking and plastic cards
• Abuses in the course of conversion operations against both individuals and legal entities
• Unauthorized connection of the Internet Bank to the client’s accounts and issue of plastic cards without the client’s knowledge
• Unauthorized increase in credit card limits

Operational efficiency

• Identification and automatic correction of deviations in transactions
• Natural Language Processing algorithms for the analysis and generation of claims
• Monitoring and forecasting the failure of infrastructure (ATMs, IT resources)
• Optimization of cash turnover and balances at cash desks and ATMs. Optimization of work of collection services
• Optimization of search and recruitment of personnel (resume analysis and primary selection)
• Real-time speech analytics for call centers and departments (consultation quality management)

If you get the impression that only resource-rich people in the financial sector can afford machine learning Kratts, this is far from the case. Thanks to breakthroughs, both in the technology itself and in computing power (clouds!), the burden of implementing the technology has decreased, and in accordance with this, costs have also decreased.

• Artificial Intelligence in Housing and Communal services
• How Artificial Intelligence will Provide Value to each person?

Conclusions

All businesses with a history of economic activity (such as a customer base) can and should apply machine learning. At least to store the same customer database. Predicting customers to leave before they leave – keeping them from leaving with various appropriate incentives. The identification of debtors, also before they are indebted, is the limitation of their consumption. All this today is simple, all you need is data.

The same findings emerge from the aforementioned research paper: businesses have found that the greatest benefits of machine learning come from offering customers personalized products. Machine learning has found support in cost efficiencies and has long been widely adopted as a risk management technology (fraud cases, money laundering, regulatory compliance).