In the 1990s, the “wolves of Wall Street” who set off a bloodbath in the capital market with their minds may not think that today there will be a machine that can replace them, which can complete thousands of transactions per second and judge whether fraud exists in a few milliseconds. In the past half century, the financial industry has gone through three stages, from “IT+ finance” to “Internet plus finance” and now in the stage of “AI + finance”.
Finance is an industry based on data and is one of the fields most suitable for the development of AI technology. With the continuous development of AI technology, the combination of financial industry and AI is becoming closer and closer. AI has gone deep into various fields such as algorithmic trading, risk management, fraud identification, financial product recommendation, customer service and so on.
In recent years, due to the rapid development of AI technology and the rapid growth of computing data, gp-gpu (Programmable GPU) has become the most important accelerator architecture in AI era due to the acceleration of machine learning algorithm training, the harsh delay requirements at the reasoning end and the support of various programming languages on the accelerator. As the pioneer and creator of gp-gpu, NVIDIA has been paying attention to and promoting the development of AI in the financial field.
In order to gain an in-depth insight into the application and potential problems of AI in the financial industry, NVIDIA investigated financial industry experts and practitioners around the world and launched the report on the current situation of AI in financial service industry, revealing the real situation behind the prosperity of financial AI; From a professional perspective, it answers the questions of how AI develops, deploys and challenges in the financial industry.
1. Voice from practitioners: AI is accelerating its penetration into the financial industry
AI is profoundly subverting all major industries like electricity more than 100 years ago. As a data-driven industry, finance is one of the most suitable application fields of AI technology. It can be seen from this report on the current situation of AI in financial services industry that more than 80% of people in the industry agree that AI is very important to the success of the company, and more than 1 / 3 believe that AI can bring 20% or even more revenue growth to their company. AI is improving the operation capacity of enterprises in many aspects.
There is another key data in the report, which mentions that the impact of AI on financial enterprises is mainly reflected in the increase of revenue, the reduction of operating costs and the improvement of customer satisfaction. Specifically, AI can generate more accurate models for enterprises, develop new products, and improve the company’s operational efficiency. From the report and public data, AI is accelerating its penetration into the whole financial industry, and is actually strengthening the enterprise’s business ability and improving profits.
2. The financial industry focuses on the three hottest applications
The role of AI applications in financial enterprises is becoming more and more obvious. The report results show that algorithmic trading, fraud detection and investment optimization are the three AI financial applications most mentioned by fintech and investment companies.
1,Algorithmic Trading: replace traders with lower error rate
Using the deep learning algorithm can replace the trader to complete the transaction. Compared with human traders, AI algorithm has lower labor cost and lower error rate, and can bring higher profits to financial companies.
For another example, in the quantitative transaction model, the back test simulation link is very key. Financial companies need to simulate real transactions and use mengka algorithm to judge the profitability and maximum pullback rate of the transaction model in millions of data. The high throughput, low latency and parallel computing performance of GPU can play its advantages here.
In overseas financial trading markets, the huge amount of data brought by high-frequency exchanges also enables GPU to give full play to its performance advantages to the greatest extent.
2,Fraud detection: explore potential customers to a greater extent and enhance customer experience
With the development of technology, the types and discrimination difficulty of various fraud methods are no longer a simple linear development. New fraud models emerge one after another. According to the data of juniper research, an American research company, online payment fraud exceeded US $22 billion in 2019 alone, which poses a great challenge to financial companies.
In traditional coping methods, manual experts need to view the report and try to find the behavior patterns of fraudsters, and then write detection rules describing these patterns. AI can analyze the data through in-depth learning, find possible fraud actions and reduce fraud risk. In addition, the Al algorithm can redefine the AML (anti money laundering) process to identify suspicious activities.
However, when the data analysis is immature, fraud detection often presents false positive problems, that is, customers without problems are identified as possible fraudsters. This will greatly reduce customers’ trust in financial enterprises and customer experience, resulting in customer loss. Deep learning can greatly reduce false positives and accelerate the analysis speed. It only takes a few seconds to complete the whole transaction process. While ensuring accuracy, it also brings customers a better transaction experience.
3,Investment Optimization: real-time portfolio adjustment has been widely used in the world
In this field, AI can provide customers with personalized investment and financial advice according to their income objectives and risk tolerance, and adjust the investment portfolio according to the changes of their income objectives and real-time changes of market conditions through algorithms.
Financial companies can also use natural language processing (NLP) to capture global language information on the Internet to predict stocks. Compared with human investment consultants, AI has great advantages in cost, benefit and efficiency.
According to the financial advisor under the financial times, intelligent investment advisers have been widely used worldwide, and their total assets under management have exceeded US $1.4 trillion in 2020. This figure is expected to double to about US $2.5 trillion by 2023.
3. What is the difficulty of financial AI? Lack of personnel, data and experience
Although many financial companies have advantages in capital and financial data, their AI technology accumulation is still immature. From AI model research to enterprise scale production, these companies still face many potential pitfalls and challenges. 76% of respondents around the world think it is difficult to deploy AI.
From the feedback of these practitioners, “too few data scientists”, “insufficient technical infrastructure” and “lack of data” are the main challenges for the financial industry to deploy AI.
To this end, more than half of the executives interviewed mentioned that their company plans to hire more AI experts to solve the problem of talent shortage. But many financial companies cannot keep top data scientists.
From NVIDIA’s observation, these challenges are often relevant. For example, the loss of data scientists is often due to the lack of infrastructure and data, which makes them trapped in the bottom of repeated work. If financial companies collaborate with data scientists and infrastructure, the more data they have, the higher the value that algorithm models can provide, and the higher the return on investment in managing additional data, a virtuous circle will be formed.
But the reality is that many data scientists have little time to deal with data science. They often need to spend a lot of time on underlying hardware, software and web development, which is also a big pain point for the financial industry to deploy AI.
In terms of data, some financial companies have a series of problems, such as insufficient data quality, single data acquisition method, decentralized data system and so on. In addition, financial companies often lack experience in deploying AI, and the lack of infrastructure and architecture is also an obstacle to the deployment of AI by these companies. Therefore, NVIDIA has been committed to cooperating with customers and providing customized solutions for different customers on these problems.
4. Viewing AI practice from classic cases
According to the characteristics and needs of financial services, NVIDIA has been committed to working with customers and experts in the industry to break through difficulties and solve problems through AI solutions, so as to tap customer value and potential.
According to public data, American Express, a global financial and financial giant, has more than 115 million valid credit cards. It can be imagined that it is facing the pressure of fraud detection. Then, the fraud rate of American Express has maintained the lowest level in the industry for 13 consecutive years.
Manish Gupta, vice president of American Express machine learning and data science research, revealed that in cooperation with NVIDIA, it detects fraud through in-depth learning model. American Express’s fraud algorithm can monitor its transactions of more than $1.2 trillion a year in real time and make fraud decisions in a few milliseconds. This has brought great benefits.
Royal Bank of Canada (RBC) is the largest bank in Canada, with 17 million customers in 36 countries / regions. The huge data challenges the time-consuming training. After adopting NVIDIA’s solution, RBC can train thousands of statistical models in parallel, but the time required is greatly reduced compared with the past.
Ant group has a huge customer base and an amazing amount of data. NVIDIA helped improve the throughput of its overall reasoning service by 2.4 times, while the delay was reduced by 20%, which not only met the performance requirements of the business, improved the user experience, but also reduced the cost by 50%. According to the estimates of McKinsey Global Institute, AI can increase global production efficiency by about 1.2% every year and increase global economic output by 16% by 2030.
At the same time, every financial transaction requires more convenience and intelligence. The high-quality experience of customers has become one of the main development directions of financial services in the future. AI application is the potential star for financial enterprises to be intelligent, real-time and improve customer experience.