Self-built AI becomes a To B blue ocean, low threshold, large-scale, and computing power optimization become the three core capabilities
There has never been an era full of optimism about the future of artificial intelligence like today.
In a recent report, Gartner predicts that by 2022, the average number of enterprise applications of AI will increase by 9 times compared to 2019, and by 2022, the commercial value of AI will reach 3.9 trillion US dollars; Forrester is more optimistic, and its comprehensive After a market research analysis, it is believed that by 2025, all companies will use AI.
It can be said that the Al project will flourish in the near future.
However, this is a judgment aimed at the macro trend of the AI industry. For companies that really need AI, what method and what service provider to choose to obtain AI capabilities is still a question that must be considered. Among them, for many large-scale enterprises, self-built AI rather than purchasing off-the-shelf AI services has become the primary choice, opening up a huge business space for AI technology service providers engaged in related fields, and because of the digital transformation and upgrading involved Strategic actions affect the whole body, and these companies will be more cautious in choosing service providers.
At this time, the value of some authoritative industry reports is reflected.
Not long ago, Forrester released "The Forrester Wave™: Predictive Analytics And Machine Learning In China, Q4 2020" report, which conducted an annual evaluation of Chinese PAML (predictive analytics and machine learning) vendors. AI is mainly composed of machine learning (ML) models. This report can be seen as an industry-level inventory, and it is also an authoritative reference for companies that want to choose the correct PAML solution to build AI and improve AI productivity.
From this report, we can see the huge changes in the PAML market industry pattern under the giants' competition, and the unique challenges behind the huge business opportunities of self-built AI.
Self-built AI PAML has become the general trend. Low threshold, scale, and computing power optimization have become the three major capabilities
From the perspective of player distribution, in 2020, PAML serving self-built AI will present a market pattern led by three major technology giants + a unicorn company , with Alibaba Cloud , Huawei Cloud , Fourth Paradigm , and Baidu Smart Cloud in the Leaders quadrant , Showing that they have taken the lead in helping companies build their own AI:
It is worth mentioning that the acquisition of this quadrant chart is based on Forrester's mature indicator system.
For example, the vertical axis represents the advantages of current products, including data, modeling, collaboration, model evaluation, model operations, methods and algorithms, and platform infrastructure; the horizontal axis represents future-oriented strategic advantages, including execution capabilities, solution routes Figure, implementation, partners, price strategy, community, etc.; the size of the circle indicates market position, and the result is derived from three dimensions: customer acceptance, product revenue evaluation, and market awareness.
It is precisely because of the complex and rigorous indicator library that Forrester's various analysis reports are widely recognized by the market.
At the same time, the report pointed out that China's digital economy is booming, and companies choosing the right PAML products can help companies build AI applications quickly and on a large scale, and improve corporate productivity.
Therefore, the Forrester report also emphatically summarizes the three major capabilities that PAML products should possess:
1. Simplify model development for different teams
With the continuous development of enterprise business, AI application scenarios will also expand from a few to thousands. To this end, PAML products should have model development capabilities suitable for different teams and roles. PAML needs a friendly visual interface to develop AI models; code-focused data science teams need a complete and integrated independent development environment that can cover the entire model development life cycle; business users who do not have deep ML knowledge need full-featured automatic machine learning (AutoML) ability to improve ML production efficiency.
2. Machine learning models can be deployed quickly and on a large scale
Building an ML model is just the starting point. In order to achieve business benefits, the company must deploy the model to production applications and supervise and manage it. PAML needs to have the ability to deploy models from development systems to production systems, supervise the performance of ML models in a business-friendly manner, manage ML models and ensure collaboration between departments, and use new data to retrain online ML models to prevent performance degradation.
3. Distributed and hybrid architecture can be used to accelerate training and inference
In the process of model training, a large number of parameter calculations are involved, which increases the burden of computing infrastructure. PAML should help companies effectively allocate training workloads to a distributed architecture to reduce waiting time for developers. In addition, model reasoning directly determines customer experience. In order to meet reasoning requirements and comply with privacy regulations, PAML should provide a hybrid architecture to facilitate the deployment of models across clouds, data centers, and edges.
To seize the blue ocean of enterprise self-built AI, PAML players still face three major challenges
The waves of the blue ocean do not come from competition among peers. The needs of customers determine whether players can go more stable and farther. The current self-built AI market is also the same.
From the perspective of customer needs, combined with some insights from the Forrester report, we can get the three major challenges faced by players on the PAML track today. These challenges are issues that latecomers must think about. On the other hand, it is the effective response to these challenges that Alibaba Cloud, Huawei Cloud, Fourth Paradigm, and Baidu Smart Cloud can be ranked in the Leaders quadrant. Or, from another perspective, the unicorn of Fourth Paradigm can compete with giants. We sit together.
1. Technically, the ability is infinitely high, but the threshold is infinitely low
There is definitely no upper limit for companies to claim the value of AI technology. AI modeling must be of high quality. Therefore, self-built AI is a matter of "technical content". If you put aside the external service platform, they will do it themselves. Generally, it needs to be completed by AI experts. In the context of the shortage of AI talents, this is a very expensive task and has a high threshold for resource input.
Therefore, as an external technical service provider, PAML players need to provide sufficient technical capabilities to enter the field, and they cannot make self-built AI become a high threshold. According to Forrester’s statement in the report, companies with advantages must “simplify model development for different teams” and “empower data engineers, scientists, business professionals, and application developers with more capabilities”-this is also an AI giant. The scope is a necessity and necessity for the popularization of enterprises.
In this regard, players such as Alibaba Cloud in the head quadrant have provided customers with a systematic tool platform. How can business personnel achieve the ability of AI experts while ensuring technical depth while lowering the threshold? Forrester mentioned in the report that automated machine learning (AutoML) should be the way to break the game. In simple terms, AutoML is a process of automating AI modeling, thus greatly reducing the threshold of AI application. Currently, major vendors have added AutoML capabilities to PAML vendors. However, AutoML in the fourth paradigm is favored by Forrester and fourth paradigm customers because of its technical and functional simplifications such as shortening the data preparation cycle, improving model performance through ultra-high-dimensional algorithms, and continuously optimizing the model. In the report, the interviewed customers of the fourth paradigm said that AutoML can be as good as a data scientist in some scenarios, and they are also satisfied with the project management and security features of the fourth paradigm ML.
Using AI to build an AI expert, or one of the ways to build your own AI in the next step, but its own demand for technology is another level higher.
2. At the application level, the landing of the in-depth scenario can make the value of self-built AI established
The ultimate goal of an enterprise's self-built AI is to improve business performance. No matter what advanced technology or refined modeling, it needs to be implemented in the scenario plan.
A high-quality PAML manufacturer that is widely recognized by the market must have a large number of scene practices in hand. These landings are the only final measure of their technical service capabilities.
Alibaba Cloud, Huawei Cloud, and Baidu Cloud rely on the market performance of cloud computing and the trend of AI to go to the cloud. They have natural practical advantages in the field of self-built AI. It is not surprising that they can immediately be recognized by customers once they are launched on PAML. The fourth paradigm, an AI platform and technical service provider that was founded as early as 2014 and is rarely heard by ordinary people, is recognized in the field of PAML, which is related to the promotion of enterprise self-built AI to a wide range of scenarios. Direct relationship.
The intelligent protection of transactions enjoyed by 90% of cardholders in China has the shadow of the fourth paradigm service. The total assets of financial institutions served by it exceed 50 trillion; in addition, the intelligent computing (intelligent calculation) behind each KFC order Recommendations, etc.) are also derived from the services of the fourth paradigm; during the epidemic, the AI anti-epidemic program of the CDC and the Ministry of Industry and Information Technology was also the company's backing support.
It is normal for the mass market to be unfamiliar with To B technology service providers. Alibaba Cloud, Huawei Cloud, Fourth Paradigm, Baidu Smart Cloud, etc. already have a wide-ranging layout in serving enterprise AI capabilities, which is self-built. AI’s competitive barriers. According to public information, even the “less-known” fourth paradigm has successfully implemented tens of thousands of AI applications in the fields of finance, retail, manufacturing, healthcare, energy, and the Internet. Including Industrial and Commercial Bank of China, Bank of Communications, China Merchants Bank, PetroChina, Huayou Energy , Yum China , Yonghui Supermarket, Budweiser, Laiyifen , Mei Su Jiaer, People's Daily, Ruijin Hospital, etc.
3. At the level of integrated promotion, self-built AI is not only technical services but also systematic improvement
In the field of To B services, a trend is becoming more and more obvious, that is, the service provider is no longer limited to providing technical services for a certain module, but is turning to the overall improvement of the business capabilities of the client enterprise. In addition to the consideration of business opportunities, the core service content must rely on the improvement of other supporting capabilities of the enterprise to achieve a better landing.
The same is true for self-built AI services. Even if PAML providers provide solutions with high technical standards, low thresholds, and practical support for a wide range of scenarios, they also need client companies to adapt to business lines, data logic, system integration, and talent training. Only by matching, can self-built AI better generate applications and create scene landing value.
According to the Forrester report, this is a large-scale deployment level "model deployment pipeline from development system to production system, supervising the performance of ML models in a business-friendly way, managing ML models and ensuring inter-departmental AI teams can collaborate."
Finally, looking back, although several players in the head quadrant have done a good job in dealing with these challenges, the challenges are not over. They will be a continuous process. Who can finally gain more recognition from customers and gain the market? The favor, it needs more time to verify. .
Self-built AI is improving, but market disputes are still full of unknowns
In another Forrester report in related fields, it can be found that more and more companies are beginning to attach importance to self-built AI capabilities, and the willingness ratio of interviewed companies has increased from 25% to 42%:
Correspondingly, some core AI technologies that can help companies build their own AI capabilities, such as platform-based AI technologies that have an advantage in versatility, and automated machine learning (AutoML), which can lower the threshold of self-built AI, have become companies' priority investment Technical points:
This trend is also embodied in Chinese companies. The fourth paradigm revealed that its important client, a leading financial institution, plans to increase its AI machine learning projects by 5 times in the next two years.
The larger the market, the fiercer the competition.
It can be found that in the 2020 Forrester quadrant chart, a giant's performance does not seem to be good- Tencent Cloud "ranked" in the second quadrant , and the market performance is relatively average.
Is Tencent Cloud really bad? In fact, Tencent Cloud has just launched its PAML and not long ago. Before the deadline for the report and research, the market has not given this fast-growing cloud computing giant a sufficient opportunity to showcase it.
This reminds us that since the PAML market structure can undergo such a big change from 2018 to 2020, the market structure in 2020 will certainly not be "steady", it is only a section of the rapid development process.
The future, such as Tencent cloud players will go in, who do not know, this is self-built AI field have charged performance points viable market - is an uncertain future faced by each player.
However, in this uncertainty, one thing is certain. Players who can better deal with the challenges of technology and application to the integration and promotion of the three dimensions will have an advantage.
And for those non-head PAML players, how should they gain a foothold in the future competition?
From 2018 to 2020, the rigorous indicator system used in Forrester's report has also undergone some adaptive adjustments. This adjustment to a certain extent represents some refined trend changes in self-built AI.
For example, it emphasizes inference optimization, edge computing support, etc., showing the improvement of IoT edge computing in self-built AI capabilities;
Added the evaluation of creating applications based on the PAML model, using the PAML model to create business workflows, etc., to express the increased demand for the implementation of self-built AI;
Suggest data relationships and automatic data statistical analysis indicators to reflect that "decision-making AI" has become an important trend in self-built AI;
Strengthening model verification and interpretation evaluation shows that many companies are beginning to pay attention to the interpretability of models in the process of self-built AI. This may mainly occur in the financial sector’s demand for compliance...
In the face of more obvious wishes, the demand for self-built AI has begun to go deeper. In addition to driving subversive changes in the market structure, it has also spawned a series of specific refinement trends, and these "small trends" may become customers in the future. An important factor in choosing a certain PAML platform.
No comments:
Post a Comment