4Paradigm lowering bar for AI products to push large-scale adoption
COVID-19 and China
Shanghai World Financial Center by Denys Nevozhai from Unsplash

Machine learning and AI algorithms have brought significant business values to Netflix and Spotify in the United States as well as to Baidu and ByteDance in China. The moat of those tech giants comes from their AI-powered personalization and recommendation systems that are difficult for their competitors to replicate so far.

Founded in 2015 by DAI Wenyuan (戴文渊), a former employee of Baidu and IT architect who created its deep learning search and platform, 4Paradigm appears to be convinced that if such AI technologies are only mastered by a few scientists and machine learning experts, it cannot become mainstream, making it hard to unleash its full potential.

Hence, the company has strived to lower the bar for creation of AI applications, potentially putting this prized technology in the hands of not just experts but essentially everyone. It has introduced an end-to-end machine learning PaaS platform, called "Prophet 3.0", to enable companies to independently and quickly develop their AI applications.

Before the platform was unveiled, DAI, who is also the CEO of 4Paradigm, let his non-IT employees use “Prophet 3.0”, and after a month of training, more than 70% of employees from the marketing team, human resources and other functions obtained a modeling score of 0.8/1.0 in the AUC test, a strong result comparable to the level of experienced data scientists. In other words, staff members with non-technical backgrounds were able to leverage such a low threshold application to build a newsfeed recommendation system similar to ByteDance's, or design AI applications that are tailored to their work.

Even though making everyone an AI expert is a distant goal in the short term, the interpretable algorithms prove to be an easier sell to client companies who are skeptical of machine learning or AI. From our perspective, this will become 4Paradigm's competitive advantage, as the company continues to roll out new versions of Prophet.

Leaving the job of developing industry-specific products to clients

According to Deloitte Insights, two major avenues for companies to access AI are1) enterprise software with AI embedded; 2) cloud-based development platforms. While Sensor Data (神策数据) and Mininglamp (明略数据) focus on delivering the first type of applications, 4Paradigm believes there is a bigger market in constructing AI infrastructure for clients. At the end of the day, companies that hope to gain a competitive advantage through AI will need to develop their own solutions. Indeed, the global AI infrastructure market is huge: spendings on AI and machine learning platforms is forecasted to grow from USD 12 billion in 2017 to USD 58 billion by 2021, according to IDC.

DAI pointed out that he had been approached by clients asking him to build AI applications, but he found that to be less cost-effective: While it would take a considerable amount of time to collect data sets and gain related-industry experience, such software may not apply to other use cases due to limited customization. DAI thus figured that 4Paradigm would better serve the needs of a bigger clientele by offering something upon which to develop their own solutions. From a supplier's point of view, since different companies have different operation procedures that they want to optimize, the AI-powered SaaS solutions developed by one service provider may only serve a niche market rather than cover bigger needs out there. Although some emerging AI vendors have released standardized products particularly in the public security industry in China, the third-party AI-infused SaaS solutions have added little value to other sectors thus far. This is the primary reason that 4Paradigm chose to take a different route from other well-known AI startups and to establish the AI PaaS platforms. The platforms do not focus on industry-specific solutions, but the solutions developed by clients based on 4Paradigm's platforms will be customized to the industries they belong to.

Strategic investments from major Chinese state-owned banks

The adoption of AI in financial services is still in its infancy; per our Research Intelligence team, China's fintech markets will be valued at RMB 24.5 billion in 2020, 31% of which will come from AI-driven risk management applications. Citigroup estimates that expenses incurred by the risk management and compliance divisions account for 10% of the operating costs or more at large-sized banks. Those divisions are expecting to integrate AI into the overall risk management framework and their current workflows as soon as possible, since reducing costs is no longer about hiring cheap labor but automating the processes.

In 2018, 90% of 4Paradigm's revenue came from its financial clients. With the financial sector as its primary focus, 4Paradigm has announced strategic investments from five major Chinese state-owned banks, namely, Industrial and Commercial Bank of China (ICBC), Bank of China (BOC), China Construction Bank (CCB), Agricultural Bank of China (ABC) and Bank of Communications, giving it a valuation of over USD 1 billion. It is a milestone for the company because banks of such scales are generally reluctant to delegate their core businesses to a startup.

4Paradigm has been helping bank clients build customized AI applications for their businesses, in areas ranging from anti-fraud applications (ICBC) to anti-money laundering (China Merchants Bank), from targeted marketing (Shanghai Bank) to bill collection reminding (China Everbright Bank). We believe that involving clients in the development phase will help close the potential trust and skill gaps. 

We learned from one of 4Paradigm's clients in the financial service industry that 4Paradigm's AI platforms have saved them a significant amount of costs and time on creating internal risk-control models. For example, the company in question used to have around 1,000 rules for identifying fraudulent activities; 4Paradigm's automated machine learning technologies have multiplied the number by over a million times to 2.5 billion and the accuracy of prediction also saw a sharp rise. While building, evaluating and revising models used to be labor-intensive for the client, automated machine learning helped it identify the most robust model and allowed the company to "test and learn" iteratively over time. As a result, it took two hours for its employees to build the risk analytics models compared to two months previously, with an increase of 3% in the accuracy rate.

Besides servicing state-owned banks, other use cases in the financial services industry include insurance fraud detection. 4Paradigm is serving PICC, China's largest property insurer, to identify fraudulent claims.

Expanding the business to healthcare

Despite the fact that client companies outside the financial service industry are contributing lower than 10% to the company’s revenue, 4Paradigm strives to lead major breakthroughs in healthcare, education and technology. Particularly, with an aging population and understaffed hospitals, the healthcare industry stands to benefit enormously from AI in areas including pre-emptive warning systems, early disease detection, personalized treatment, surgical simulation and so on.

In this sector, 4Paradigm has entered into strategic cooperation with Shanghai Ruijin Hospital, a leading hospital with the highest rating in China’s medical system, and they have been working on the auxiliary treatment and health management.

For example, "Runing Knows Sugar", the machine learning system built by 4Paragidm, collected the hospital's data from 2010 to 2013, including gender, height, weight, blood sugar levels, drinking and smoking history in order to predict whether patients would develop diabetes within the next three years. It also issues risk forecasts for the next 9 to 15 years of a patient’s life as a reference. The average accuracy rate has reached 88%, exceeding professional doctors, according to South China Morning Post.

Latest Updates:

See Also

Communicate Directly with the Author!

Ask the author questions about the copied text

Research Reports
Editor's Picks