Mature industry always faces decline phase after going through the struggling emerging phase and accelerated growth stage. According to the industry life cycle model, an industry has stages of embryonic, growth, shakeout, mature, eventually decline and facing death just as the life of a human being. This 'natural law' put pressure to the companies in a middle-age industry who is encountering a flattening curve, with performance indicators and financial numbers not improving.
Those in the mature industry who do not keep pace with the times would finally be driven out due to technological development, societal change and globalization. Though technology shock is a dominant factor to the death of an industry, it is, in turn, a technology with tremendous power to save companies from dying with the declined industry. Artificial intelligence is such a transformative technology that has the potential to revolutionize how we live, work, communicate and think.
Chinese AI company 4Paradigm perceives the pain point and demand of mature industry giants and positions itself as an enabler of corporate AI transformation. As companies getting more awareness of AI technology, 4Paradigm realizes the opportunities proliferating in using AI across firms' value chains and aims to reshape business with artificial intelligence.
As discussing the business model of 4Paradigm, this article goes through the why, how, and who questions of corporate AI transformation and the future trend for AI practitioners.
Requirement for AI technology in 5G era
Classic business indicators include measures of assets, revenue growth rate, cash flow performance, and return of equity for different types of business. The founder of 4Paradigm, Dr. Dai Wenyuan said the speed of innovation is the new criterion to measure the success of a business in the world of today and tomorrow. Innovation is the key to competitive advantage of a business. Business owners should always think about innovation related questions: How many new user needs are identified each day? How many new approaches to reducing cost are invented? How many new ideas to improve operational efficiency are created?
Measures of speed also changes. 5G enables ultra-quick massive downloads and uploads that transfers more data, achieving a speed of 50G/s. To plan for an AIoT future, instant decision-making and high dimensional machine learning techniques are required due to the speed of data access and the amount of highly complex data.
In this circumstance, traditional business intelligence (BI) eclipses. There are four major limitations in traditional BI: 1) BI uses historical lagged data, which couldn't satisfy real-time analysis. 2) statistical methods lower data dimension and therefore causes information loss. 3) BI data management causes inconsistency between online and offline data. 4) Missing data labels fails to train machine learning algorithms.
AI, on the other hand, is the way to solve these problems to better the business decision-making process.
"1+N" strategy to scale up AI transformation
As early as in 2015, 4Paradigm launched AI PaaS Prophet with the mission to make "AI for everyone". In late 2017, 4Paradigm upgraded it to the 3.0 version allowing enterprises to build customized AI applications. In 2018, as some Chinese AI companies went bankrupt, Chinese startups accelerate their commercialization plans. 4Paradigm, among the survivors, made a move to finance industry, where data are relatively rich and structured so barriers to entry are relatively low. Starting from providing products and services to China's top state-owned commercial banks, 4Paradigm discovered more monetization possibilities.
In 2019, 4Paradigm announced "1+N" strategy and unveiled its corporate-level AI integration system SageOne. From the past experience working with 7,617 clients in 12,648, 4Paradigm summarizes "1+N" methodology. The strategy divides enterprise business lines into two categories: the 1 scenario and the N scenarios.
"To make the most of AI in the 1 scenario" is to employ as much as possible time and resources to achieve even one percent improvement in a few core businesses. For companies in a mature industry, the one-point achievement causes billions-of-yuan effect in increasing sales revenue, decreasing operational cost or attracting customers, where a company build a solid moat.
"To apply AI in the N scenarios," is to adopt standardized machine learning functions in tens of thousands of detailed yet not crucial business lines, achieving AI transformation at scale. Every possible scenario in a firm makes up a tiny fraction of the whole business, but the power of scalability is significant to the company. A slight enhancement in each of the N scenarios together makes a huge difference, helping companies to gain competitive advantages.
Fulfilling "1" and "N", a company completes an AI transformation.
From PaaS to software/hardware integration —computing capacity is the key
After identifying companies' desires in AI transformation and summarizing the "1+N" strategy, 4Paradigm's next move is to match its technology with its own business objectives. It provides businesses with platforms so that clients design and build their own customized AI applications.
As founded by a group of renowned machine learning scientists, 4Paradigm is most confident about its technical background. The founding team focuses on transfer learning and autoML.
Transfer learning, a deep learning technique allowing programmers to use existing knowledge of a model to the model building in another area, is especially helpful when data are insufficient in a new domain.
AutoML makes machine learning more accessible to people with programming backgrounds by automating the model building and parameter tuning process. Despite some accuracy loss, autoML achieves near-optimal results with high efficiency.
Strong tech ability not only serves as a foundation to the product but also a key to unlock new possibilities. Both of the technologies, transfer learning and autoML, have a long way to go in terms of engineering solutions in the real world, but the vision of 4Paradigm is appreciated by industry giants.
Back to the "1+N" strategy and platform position discussion. In doing so, achieving low technical barriers, high reliability and low TCO (total cost of ownership) are important for clients, and therefore 4Paradigm works hard on developing high-dimensional, real-time, automated algorithms, as well as ways to boost computing power.
High dimensional (VC dimension) machine learning algorithm suits complex high dimensional data in a world where data explode. Real-time machine learning continuously accesses and analyzes data to help generate instant business decisions, matching the 5G speed.
With numerous, complex data, overfitting (learning too many noises that only fits specific conditions) is less a problem, and meanwhile, detailed differences are extremely important to advance the "1" business scenario. High-dimensional machine learning and real-time analytics capture trillions of micro-rules, which could hardly be achieved by employing expert opinions to build models in the old way.
As for applying AI at scale in the "N" scenarios, 4Paradigm implements AutoML and will develop other technology such as AutoPTL, AutoSSL, AutoPU, AutoKGE, etc. Embedding auto ML and other automation features on its platform, 4Paradigm intends to make it easier for developers and professionals to leverage machine learning in solving identifiable business problems.
However, none of the functions mentioned is feasible without sufficient computing power.
4Paradigm realized the importance of computing capacity. For many companies, it's not the algorithm that hinders business application but the limited computing infrastructure. In order for companies to land AI at scale, the number of computers and the total cost of ownership (TCO) grows exponentially. Flexibly providing compute power at ever-lower costs is a breakthrough point.
Companies to apply AI solutions have to take TCO into serious consideration. AI could help save labor-hours, reduce operation costs, and increase sales in multiple business scenarios. But the cost of obtaining compute power, storage devices, maintenance fees are often left unclear. Besides, the time spent to restructure processes, revise systems, integrate AI applications with corresponding systems and hardware.
Traditional computing power mismatches the one AI needs, and therefore running AI on a traditional computer causes waste of energy consumption and computer space. 4Paradigm believes that AI software not only is enabled by hardware but more importantly, defines hardware.
From such perspective of lowering TCO, 4Paradigm brought SageOne to life on June 20, 2019, a corporate level AI software/hardware integration system. Combining AI applications, platform and software-defined computing hardware, SageOne would improve data management, algorithm development and computing power collectively, making corporate AI transformation affordable.
When capability of data management, algorithm and computing power meet enormous industry-specific company data, corporate AI transformation happens.
Who are ready to pay for AI transformation?
"We are not a company in any specific industry vertical," says the founder Dr. Dai Wenyuan. 4Paradigm targets those AI-ready companies – those who already invested a lot in digitalization with sufficient data collection, those who have the pressure of transformation but not aware of how to utilize their accumulation of experiences, those top companies with the largest database and the most complex business lines in different industries.
Taking energy industry as an example, Chinese economic underpinnings that support utilities are losing the momentum. Companies are forced to produce energy at ever-lower costs without compromising safety.
During the past several years, Chinese energy companies such as China National Petroleum Corporation (中国石油天然气集团有限公司) have made progress in digitalization and currently is looking for the next generation transformation incorporated with AI technology.
Pipeline security is the "1" scenario for such energy giants. Currently, The total length of existing oil and gas pipelines in China is 133,100 kilometers. It is estimated that by 2020, the length of China's oil and gas long-distance pipelines will reach 169,000 kilometers, according to China Petroleum & Gas Pipeline Telecom & Electricity Engineering Corporation (中国石油天然气通信电力工程有限公司). It is definite that damage to the pipelines not only causes gigantic financial losses but also severely jeopardize the security of the citizens. Therefore, energy companies spent tremendously to pipeline security.
The situation could be greatly improved by AI. Environment conditions differ greatly along the hundreds of thousands of pipelines. High dimensional, real-time AI technology helps detect and predict possible security accidents based on tens of thousands of rules generated by modeling the highly complex underground and upper-ground conditions. Without AI, companies can only count on the patrolling and maintenance personnel, where accuracy is less-controllable and costs high.
Energy sector, including coal, oil, nuclear, gas, hydroelectric, and renewables, has a market size of trillions of Chinese yuan. The national government is also encouraging capital investment into factory and infrastructure upgrade. We could expect an explosion of AI companies to closely cooperate with the energy sector in the foreseeable future.
Similar situations are emerging in other industries. Athough revenue from banks and financial companies makes up almost 90% of the total revenue, 4Paradigm have actively explored other industries such as new retail, insurance, medical services, energy, government services, and media.
It has already built relationship with top state-owned commercial banks, Yum! Brands, Ruijin Hospital, China Daily, etc. According to 4Paradigm, 50% of the total revenue will come from financial industry in 2019 and the rest comes from energy, retail, and public services.
What's next? An agile team in a great AI-ecosystem
Moving forward, 4Paradigm envisages an AI-ecosystem, in which it could confine the team as small and agile as possible to focus on the "mission of empowering".
Deeper relationship with industry giants
By working with industry giants, and the headquarters of the large corporations, 4Paradigm could dig deeper into complex scenarios and develop standardized technology to be used in countless branches of the corporation. The joint force of 4Paradigm's AI capability and giants' business experiences makes it possible to discover new monetization possibilities in other verticals.
Closer collaboration with tech companies
4Paradigm said to EqualOcean when asked about a list of Chinese tech startups, "most of them are our future partners to be." Especially in robotics and IoT area, there will be increasing partnerships between 4Paradigm and other companies.
AI consulting and talent training
Certainly, there is a shortage of talents with both technical skills and industry understandings. 4Paradigm launched the talent training programs in 2017, providing online courses and AI labs to help business people quickly excel AI tools. By the end of 2018, 4Paradigm, with Terminus (特斯联), Iflyteck (科大讯飞), Cambricon (寒武纪)，jointly signed the Memoranda of AI & Industry Talent Fostering Collaboration (人工智能产业人才培育标准合作备忘录), with officials attending from Ministry of Industry and Information Technology.
In addition, when working with companies with little AI exposure, it comes to the problems that business practitioners do not ask the most effective questions, or fail to explain the actual demand. AI consulting is a new service line 4Paradigm plans to monetize.
In the end
4Paradigm has a world-changing vision, strong technolgy backgroud, and good relationship with industries. But dancing with elephants can be dangerous. Especially, as the only startup with a joint investment of all the "five major banks" in China, it won't be a surprise that 4Paradigm ends up being acquired as many AI startups were.