domeet webmaster

Xiao Jing: Four major pain points in the financial industry, how does Ping An solve it?

The four types of scenarios in the financial industry mainly have the following pain points: risk control (miscellaneous risks + more fraud), customer acquisition (low frequency + weak conversion), service (heavy model + poor experience), and operation (low efficiency) + High cost)

Risk control scenarios are characterized by mixed risks and fraud. For example, a single transaction seems to be fine, but multiple transactions combined together have a risk of fraud. An individual looks normal, and small groups of multiple people may be at risk. Expansion into an enterprise, or focusing on a certain industry, or on the macro market, there may be potential risks in all aspects. If it relies on manual monitoring and control, misjudgments or omissions are likely to occur.

Acquisition scenario, traditional financial marketing methods include physical outlets, phone text messages, ground push salons and other methods to push standardized products to all customers. This method has high customer acquisition costs,The frequency of channels is low and the user experience needs to be improved. Most financial services have low interaction frequency and limited understanding of customers. Under these conditions, how to better understand customer needs, improve customer viscosity, loyalty, and the success rate of cross-selling and up-selling is a realistic problem.

In the service scenario, the traditional service model is heavier, and consumer behavior and demand are constantly changing, making traditional financial services face reconstruction on each chain in each scenario. At the same time, in the context of the gradual disappearance of the demographic dividend, traditional artificial customer service has the characteristics of high training costs, high mobility, and uneven service effects, which affect service quality and user experience. How to make the customer service model lighter and improve the quality of services at the same time is also a pain point that needs to be resolved in the current fierce market competition.

Operational scenarios. There are a lot of manual operations in the business operations of the financial industry, which are often simple and repetitive. There is an urgent need to reduce operating costs and improve management and operational efficiency. Taking Ping An as an example, if we increase the efficiency by 1%, we can increase more than 10 billion yuan in profits every year. Therefore, achieving cost reduction and efficiency improvement of business operation management is also an extremely important pain point.

How to use technology to solve the pain points of the financial industry

Really improve the quality and production efficiency of products and services, and solve the business pain points mentioned above. The key lies in intelligence. The use of artificial intelligence technology to restructure the production process to achieve what we call “Three lifts and two drops”-improve efficiency, improve results, improve user experience, reduce risk, and reduce costs. However, it is much more difficult to restructure production than to go online. It requires both strong technical capabilities and familiarity with traditional business processes, so that technology and business processes can be deeply integrated to achieve the goal of production restructuring. . Therefore, we must not only complete the information and data infrastructure in the IT system, have the algorithm technology and computing capabilities of intelligent construction, but also continue to iterate in actual business scenarios under the guidance of in-depth cooperation of industry experts, and finally form a complete Intelligent business solutions to effectively achieve the goal of intelligent operations.

Intelligent operation is an important strategic direction of Ping An. The first step is to build a big data platform at the bottom. We spent nearly a year completely opening up the data islands of more than 17,000 sales offices, integrating the data into a unified platform, and establishing automatic cleaning, integration, update, quality management, Standardization, desensitization, security and other mechanisms, and establish strict management regulations such as rights management and privacy protection. Statistical analysis was performed on the big data platform in accordance with compliance requirements to obtain desensitized portrait tags. Then build basic artificial intelligence capabilities, including the ability to read, listen, speak and read, such as face recognition, micro-expression recognition, medical image analysis, voice recognition, voiceprint recognition and other technologies. Then build a professional knowledge map , including automotive, corporate, medical, education, agriculture and other fields, which is the largest barrier of traditional industries compared to the Internet and high-tech industries . Finally, we will focus on business needs and integrate technologies deeply to build a complete intelligent solution. We will continue to enrich and improve the intelligent business solutions that can be applied on a large scale. , Comprehensive coverage of core business areas such as finance, healthcare, smart cities. All these work together constitute our Ping An brain intelligence engine, promote Ping An to quickly and effectively realize the intelligence of various business links, and continue to consolidate key technologies and business barriers.

Specific AI application cases

Identity authentication : Multi-modality, including face, voiceprint, lip, etc., applied to the recording and video recording of bank insurance in the financial field, to prevent business staff from misleading and identifying customers It’s me.

Microcredit: It turns out that we have more than 840 stores, and we have to judge the risk of default at the stores. Face recognition and micro-expression recognition are now available. Three-minute loans can be realized online. These big data design and other risk control methods have also reduced the default rate.

Corporate investment and credit : Based on the underlying data, static food packages, dynamic public opinion, and investment relationships between enterprises, the three types of data form a knowledge map, establishing a bond default model, and an investment risk control model. Most companies can predict the risk dimension in 6-9 months in advance, and a detailed introduction helps the business to quickly locate and judge whether it is a false positive.

Automatic loss determination for image recognition : Ping An is now the only company that realizes large-scale identification and identification of damage. Through a good car owner app, taking photos and uploading, it can determine and repair the damage within a few minutes. The process can be completed quickly under a few thousand pieces. Improve efficiency and solve problems. It is required to have a complete knowledge map in the background, different models, different losses. At the same time, auto insurance is a small profit area, and there is often fraud. After using the anti-fraud engine, billions of operating expenses have been saved.

In addition, in the field of operations, such as large-scale lawsuits, artificial intelligence models determine whether the evidence is sufficient; in the field of services, we have 125,000 customer service customers. Robot assistants can automatically help solve some customer problems.

Min Wanli: Re-dismantling AI, the effect behind scale collaboration

I graduated from the University of Science and Technology of China to an IBM postgraduate, then went to Google and Alibaba to do industry investment last year. My career path is simple: write mathematical formulas into the industry to create value.

Reassembling AI: Less artificial, more intelligence. A stands for Actionable, Accessible, Affordable

Case 1:

AI is not just image recognition, it is often a fusion of multiple technologies, especially in the industry, there are many uncertainties. For example, in this scene, there is a left turn and a forbidden card. How can this referee deal with a conflicting signal? These are not available in the laboratory, they must have a common sense, and now there are other references that can help me judge in real time. So when we go into the industry, the AI ​​judge is not even you, but a third-party market. Even if you make a good product, but the market does not pay, you will not have any profit. . AI is actually an application of integrated technology and intelligence in the context of uncertainty.

Case 2: Smart City

Taking urban traffic as an example, the reason why the camera “feels” rather than “moves” is because there is no calculation or algorithm.

Cities need to learn to think, learn to think globally, and learn to think online. The camera cannot be a “myopic” type of conditioned reflection. It must be a collaborative wisdom with full time and no blind spots.

Through the speed of all traffic at the intersections, we can accurately quantify the traffic attributes at each intersection. Based on the traffic volume, we can determine the current pressure on each node in the city, and then know the pressure of each node on all networks, and can effectively adjust the green time of the signal light. Real-time and effective matching of supply and demand. What can be done in the end? Able to get through the blocking points and reduce the congestion time. For example, due to traffic congestion in developing countries, ambulances cannot pass through and patients cannot be treated in a timely manner. In response to this problem, we can clear the cars in front of ambulances by adjusting the traffic signal.

In this case, we made a very simple use case, which is to solve the signal problem of the entire city, so that each signal light can hear the whistle of the ambulance, and let the signal light open the green lane for the ambulance in advance. . What is the effect? The travel time of the ambulance in Kuala Lumpur, Malaysia can save 233 seconds from the patient’s home, which may be the difference between life and death for the patient. When the signals of this city can learn to think, in the end it is the people who benefit.

In addition, what happens when a city thinks globally? If you only think at a single point, what you may get is a very simple part, until all are linked together, you will see every pulse of this city’s pulse, what happens at every intersection and node, and it will spread Where to go. And detect in real time what each event is like, spread from a single point to a local or even global, and then quickly generate a response plan based on this deduction. Through video and mobile Internet data, you will never get off work and never get tired, because as long as the data stream comes, this computer is always calculating, so what happens at any moment in this city will be automatically calculated and deduced Transmission to traffic police, urban management, etc., 24 hours of continuous use.

With a global perspective, it will also lead to local improvements. For example in China todayAny medium-sized city can see a lot of equipment at the intersection, the ground is a buried coil, the sky is a camera, and a control switch, which detects the traffic situation at the intersection in real time. But there is a fundamental contradiction. It is a “myopia”. It only sees local information. It cannot see upstream information. There is no linkage, but it only passively responds to the current intersection. Appears the condition, then does the deadly conditioning conditional reflex. Suppose we have the ability to think globally today, just like seeing all the intersections in the city in one eye, and knowing how many cars are coming at each congested intersection, it is also possible to make preventive measures. Instructions. There must be a global view and active intervention. This is how we use a bird’s-eye view to link all the data together globally. In real-time calculation, it is possible to form a cross-network cross-region linkage, so it is not Small cleverness at a single point, but the wisdom of forming a global synergy on the entire chain of all points together.

On September 15, 2016, a system we launched in Guangzhou has already reduced the congestion index by about 19%, and it is using the data traffic signal of the Internet APP on the most congested roads, plus The semaphore control algorithm achieves the effect after docking in real time. So this example once again tells us a phenomenon that smart cities have been doing for so many years, with so much hardware investment, and accumulated huge amounts of data, if we are still getting more congested, the speed of driving is getting more Slow, there must be something wrong. The data is used as an excuse after the fact. Today what we want is to convert those real-time data into actionable inside real time, so it is called AI to solve the current problem, and even the future may appear. Congestion resolution. It’s not for us to find out what happened afterwards, but to let those bad things happen at all.

Crossing industry barriers

This picture lists quite a few cases, and each case is solved by the same solution.

On a production line, if viewed from a digital point of view, it flows a lot of “numbers”, process parameters, and process parameters.Number, equipment status, data environment variables, etc. We pull out all production records, look at the causal relationship between the amount of each batch in history and the process, find some rules, empirical characteristics, there are It is possible to add a machine neural network, use industrial artificial intelligence, and finally find the causal relationship between those process parameter control parameters and the final product mass, not an association relationship. With causality, it is possible to regulate, and to start with every factor, it is possible to make the results better.

In a case in Hangzhou, Zhejiang, this company produces a very traditional type of circulating fluidized bed. I digitally present it inside and use a three-dimensional perspective to look at the distribution of high temperature. The blowing surface and coal powder are dynamically controlled Later, we will take out all the historical data of pulverized coal supply and air blowing, see the input and output between heat and consumption, and finally find the best control parameters and logic, and apply it to practice. The parameters were adjusted in the last two minutes, and the coal combustion efficiency was increased by 2.6%, saving 16 million annually. There is no hardware investment in this, some are data accumulation, plus AI algorithms. So the value in the industrial production line may be four or two pounds, and it can be done without buying a sensor.

We monitor in real time which part of each radio is abnormal through dozens or hundreds of radios. In the past, it was actually signal processing and monitoring signals. Later we did a very simple and rude thing, made an energy spectrum, obtained the average value of the energy spectrum, and then compared. Therefore, if the problem is discovered earlier, the maintenance cost can be reduced from 300,000 to 500,000 to 20,000 to 30,000.

These cases actually point us to a very simple formula, value = f (data, science and technology, industry knowledge), in which the CEO must find the value according to the industry track, value closed loop, business structure; CTO is responsible for Technology selection, path milestones,Scale & Speed ​​to set f; scientists build science and technology based on feasibility and universality. Cost savings are limited, while value creation is unlimited. When cloud computing is really used in the scene, it will make a lot of effort to create greater value.

In this process, there must be more and more data. To release and capture the value of this data, we need strong science and technology, which is cloud computing. Today’s massive data is beyond the limits of the human brain. It cannot be done without a computer, but it is a pity that in the past our imagination of cloud computing shrank into a simple IT operation and maintenance hosting, and then saves your IT costs. However, the biggest power of cloud computing is not the cost savings of IT, but a laser effect of coordinated scale. It can bring 100,000 machines together in one second. The law behind it. Laser is a bit like Archimedes reflecting thousands of mirrors at the same time, burning the enemy’s sailing boat with the same reason. After seeing through the law, then calculate it immediately and dissolve it. Therefore, the effect of the coordinated laser perspective on this scale can help us explore the new value behind the data, and this is also the biggest power of cloud computing, and the cost savings must be achieved. Even more exciting. When we combine the power of cloud computing with data into the scene, we can find unprecedented new value, and it may be in a very soft way.

How companies use AI: choose a scene, first vertical then horizontal

When you want to use AI in any company, find the pain point and choose the technology anchor point.

The transformation and resolution of the entire pain point must be a continuous management and must be sufficiently wise. Vertical and horizontal, choose a single point, the value is obvious, you canTransparently transmitted to the final business report. Write a question thoroughly and win trust, it is possible to do better, especially early entrepreneurs, do not talk about the platform prematurely, let alone talk about being an ecology. Ecology is a big Mac. of. If you talk about the platform prematurely, just like when you first walked the rivers and lakes, you have to be the leader of the martial arts league. At most, it becomes a joke for others. Methodology and tools are indispensable. The value of all products has been proven in industry and practice. Pick the right problem, disassemble the problem, give priority to the most valuable problem, the fastest problem, win the trust, and then do the second and third problems.

First, When we are doing AI applications in the industry, we must not be superficial. We need to go deep into the bottom of this industry, go to the workshop to the field, and go to the command center. See its pain points. Second, you can’t make it happen. Open source software comes up with a CNN tuning and says that it is AI. Today’s AI is no longer vision and speech. We need science and we need to improve technology. In the end, whoever can be a time friend must have created deep value, and can bring the industry at this time node.

Dare to challenge uncertainty, dare to squeeze out the oil in the stone, can create new incremental value

If you want to go in the industry on this picture, there must be uncertainty when you want to go deeper. It may be lucky. The one that comes out will come out soon, but there is another situation that you may I ran into a stone, and it wasn’t petroleum, but it was found to be a gem after careful carving, so it is also valuable. Therefore, I think that redefining oil is that we have to dare to squeeze the oil out of the stones. Especially in traditional industries, when everyone feels that this industry is of little value,Do you have the courage to use the latest technology to find new value space in this industry. Then the answer is very clear. We have done so many traditional industries and found 100% value. We have already told us this. AI can create new value increments in traditional industries. Its power is far beyond what we can think of in the laboratory. When we have a technical problem and have this willingness, we can find the direction of the industry to achieve it, advanced technology for social, creating social value, the value of inclusive.

Q & A session

Xu Siqing: For entrepreneurs, there is neither computing power nor good algorithms. How should artificial intelligence start?

Xiao Jing: Large enterprises such as financial institutions and technology companies are already setting up their own R & D teams. Large enterprises hope that they can have the corresponding capabilities to Respond to more future needs without using outsourcing. It is not realistic for SMEs to establish teams of the same size. It is necessary to deepen the key areas first, establish and develop their own teams, and absorb external technologies. It’s not enough to just do technology, it is difficult to solve business problems. Crossing the river by feeling the stones is in the absence of a bridge, and some standardized problems can be introduced into the solution.

Min Wanli : Entrepreneurs relying on a single point of AI technology to support valuation is becoming more and more difficult, because the valuation refers more to the value created in the industry. In this regard, small companies have advantages that large companies do not have. Large companies pay attention to economies of scale. Small companies can “take a shot to the end” and take root in an industry to deepen, solve the pain points of the industry and quickly copy it to avoid large company competition.

Xu Siqing: Will big platforms pay for startups?

Min Wanli: The buyer will not be a big platform like Ali, but an enterprise in the industry that needs comprehensive transformation. For example, the steel industry solves problems in the steelmaking process and finally realizes digital steelmaking.

Xu Siqing: Opportunities are between traditional companies, and big Internet giants are no longer needed.

Cui Kaiyun, chairman of the Silicon Valley Alumni Association of the University of Science and Technology of China, asked: There have been several major ups and downs in the history of artificial intelligence development. What are the main differences this time to make it develop continuously?

Xiao Jing: The biggest difference is that it is essentially produced