Written by/Editor by Chen Jiying/Wan Tiannan’s cool parameter competition and brutal battle of hundreds of models, is this the end of AI large models? Microsoft, which has won a lot of money in this wave of large model trends, has done exactly the opposite. It has successively co

/Edited by Chen Jiying

/Wan Tiannan

Cool parameter competition and cruel battle of hundreds of models, is this the end of large AI models? Microsoft, which has won a lot of money in the

large model trend, has done exactly the opposite. It has successively cooperated with openai and meta's large models, and its strategic focus is on volume applications.

This coincides with Belle Fashion’s view: the value endpoint of ai is not only in technological innovation, but also in business applications.

In fact, Belle Fashion has been favoring AI for a long time before this wave of large AI models became popular. As early as 2022, in the upgraded architecture diagram established internally by Belle Fashion, "the top layer is AI."

is not just an idea, it has already been implemented.

In November 2023, in Belle Fashion's "digital workbench"-DingTalk group, AI assistants "commodity digital employees" officially "joined", "In the future, tens of thousands of employees will be equipped with an AI assistant." This is the blueprint for Belle fashion.

Large shoe and apparel companies like Belle are not the only ones to benefit from AI.

Aiwei Electronics, the leader in the chip industry, its IC intelligent customer service can answer customers' various professional questions about chips in real time 24/7; as a representative of small and medium-sized enterprises, Youtian Youshi is one step ahead in the exploration of "ai + low code" and realizes It provides automatic filling and data analysis of sales information such as "customer visit follow-up", leading to lower-cost and higher-quality growth. These explorations of

confirm a conclusion made by Ye Jun, CEO of DingTalk. In 2024, "will not be the protagonist of large models", but "c-end AI native products and b-end AI products will become the protagonists."

According to Ye Jun’s prediction, 10 million AI assistants will be created on DingTalk in the next three years.

Elephants can also dance, the king of shoes is light and "smart"

Belle Fashion's application of AI large models is both far-sighted and pragmatic.

As the king of shoes in China, Belle Fashion owns about 20 private brands and cooperative brands, with nearly 10,000 stores in more than 300 cities across the country.

For Belle Fashion, it is necessary to be "well aware" in order to dynamically manage the import, storage, export and replenishment of goods and continuously optimize the supply chain.

But in the face of such a vast number of terminals, it is not easy to track sales and inventory in real time.

In the early years, data statistics relied on the headquarters to issue emails to the regions, and then to provinces, cities, and stores, and then distributed them layer by layer, and then reported upward layer by layer. Under the long and tedious process, it will take at least 3-5 days to summarize and compile the data.

2019 is a turning point. At that time, IT personnel from all regions of Belle Fashion gathered in the DingTalk group and used the latter's table online editing function to achieve online summary of various business-related data. It only took two hours to complete thousands of data across the country. Store statistics work.

Whether it is three days or two hours, statistics are always lagging behind and it is also labor-intensive.

Today, Belle Fashion can call AI at any time to query the sales performance of more than 10,000 stores and each clerk. "The store manager can ask at any time who is the top seller in the store; which pair of shoes sells best and which pair of shoes has the fastest sales growth?" Waiting for data, AI can also report it in real time."

In Belle Fashion, AI not only “recognizes numbers”, but can also assist in advancing processes and help complete system operations efficiently and conveniently. The Belle “product digital employee” mentioned above is an example.

In order and replenishment mode, inventory flows efficiently and it is necessary to check the sending and receiving data in real time.

In the past, when there was a discrepancy between the delivery and receipt of goods (the number of goods shipped did not match the number of goods received), Belle Fashion data management personnel needed to intervene and get through the sending and receiving parties and even third-party personnel to communicate and determine responsibility.

Now, when it is necessary to take the initiative to acknowledge responsibility, the sender and receiver can directly initiate a dialogue with AI digital employees. From data verification, to proactive acknowledgment, to correction of responsibility, and completion of documents, the entire process is smooth and smooth, and AI can take over the entire process. Not only does it free up manpower, it also greatly improves business efficiency. After the above-mentioned attempts of

are successful, Belle Fashion plans to expand the implementation of AI to more scenarios such as document execution follow-up, goods adjustment and replenishment, etc.

dot-like application is just the beginning for Belle Fashion, which has determined AI as an important direction of the group's digital transformation strategy. In Belle Fashion's AI vision, can replace manpower, assist decision-making, improve organizational efficiency, etc., are all on the future list.

Currently, among the tens of thousands of Belle Fashion employees, many are front-line workers. In the past, they had to spend a lot of energy learning various digital tools. “You need to understand not only products and people, but also numbers and tools. For front-line store staff, , it's a bit difficult," a Belle insider revealed to "Financial Stories".

With AI assistance, front-line employees do not need to get lost in a sea of ​​tools. They only need to call the corresponding AI application when necessary, which can save more energy and time and focus on serving customers.

For example, when a certain shoe model in a Belle store is out of stock, the salesperson in the past may have to log in to the system, enter the product number to search for inventory, call to apply for stock transfer, etc., which is time-consuming and labor-intensive; with an AI assistant, just say With just a few words, you can ask it to help check inventory, help communicate and adjust goods, help place orders and deliver them to customers' doorsteps, etc.

For organizations, from a top-level perspective, AI technology can break through individual limitations and insufficient computing power of the human brain, support comprehensive, efficient, dynamic, and real-time analysis and decision-making of the organizational system, and can also assist terminal intelligent decision-making in line with the company's strategic will.

For large companies like Belle Fashion, with the help of AI, it is easier to achieve the ultimate goal of improving organizational operation efficiency.

Standing on the stage of digitalization and AI, elephants can also dance lightly.

Indecisive questions ai

For enterprises, it is not difficult to provide good standardized products, but it is not easy to provide good personalized services.

Take Q&A, a high-frequency interaction scenario that is spread throughout pre-sales, sales, and after-sales, there are constant shortcomings - manual customer service is either not professional enough, and automated customer service often behaves "mentally retarded".

The quality of the customer service Q&A session is directly related to the conversion and repurchase of potential users (customers).

Chip company Awinic has a deep understanding of the above pain points. Its chip products have thousands of SKUs, involving more than 40 sub-categories, covering professional knowledge in different fields. Moreover, each product has dozens of parameters. It is not easy for customer service to respond quickly, accurately and professionally.

If all customer service staff are experienced technical experts, on the one hand, recruitment and training costs may be significantly increased; on the other hand, chip products are updated and iterated quickly, and the learning ability of manual customer service may not necessarily keep up with the times. The traditional approach of customer service is to download product documents from the official website and then search for parameter answers in massive amounts of information. Naturally, it is difficult to provide an immediate and professional response.

For this reason, in 2023, Awinic Electronics and DingTalk worked together to create an Awinic exclusive model. Based on this, an "ai intelligent customer service" was built that can answer customer questions 24 hours a day, 7 days a week.

In contrast, traditional question and answer robots cannot understand the meaning of the context and can only adopt a passive mechanism. When keywords are triggered, they can give standard answers set in advance, which will inevitably lead to incorrect answers.

Awinic's AI intelligent customer service can understand contextual conversations and generate more natural and accurate answers. It does not need to rely on manual maintenance of keywords and question and answer libraries, which can not only improve user experience, but also greatly reduce costs.

When intelligent customer service identifies high-potential customers during Q&A, it can also remind product consultants to join the conversation as soon as possible. The latter can follow up on needs, lock in business opportunities, and drive conversions based on customer consultation summaries summarized by intelligent customer service.

intelligent question and answer can not only transform into an intelligent customer service, but also become a "teacher" within the enterprise to train and empower employees.

Mitsubishi Elevator's front-line personnel used to need to temporarily look up information or consult relevant experts when encountering maintenance problems on site. However, this mode is time-consuming and labor-intensive, and also affects the customer experience.

Now, when Mitsubishi maintenance personnel encounter problems, they can directly consult AI. The latter can not only answer in real time, but also mark the authoritative source of the answer. The accuracy is higher, maintenance personnel save effort and worry, and customers can also save time; not only maintenance personnel , front-line employees can also ask AI when they are in doubt. In the group @Q&Abot, they can get answers at any time.

Nowadays, intelligent question and answer, including intelligent customer service, has become one of the scenarios where AI large models are quickly implemented. According to Sullivan's forecast, the scale of intelligent customer service is expected to grow to 18.13 billion yuan by 2027, with a compound growth rate of more than 20% within five years.

ai+ low code, small and medium-sized enterprises can also play with

and Belle Fashion and other giants in the AI ​​layout, looking at ten years, foresight and foresight are different.

When exploring AI, small and medium-sized enterprises focus on small, light and accurate solutions to actual needs. The superimposition of AI and low code gives small, medium and micro enterprises the opportunity to be on the high list of AI at a lightweight and low cost.

Retail company Youlianyoushi is headquartered in Jinjiang, Fujian Province. Local IT talents are scarce, and the company's IT department only has a few technicians. To this end, there is a shortcut to the quiet world. Based on the low-code model, the digital system was built on DingTalk.

In April this year, with the launch of DingTalk’s comprehensive intelligence, and the subsequent launch of smart products such as Yida (a low-code platform of DingTalk) AI, people living near the water and food quickly followed up, further liberating through AI increased productivity.

You Ling You Shi’s entry point this time is the AI-based transformation of the sales process.

In the past, front-line sales had to enter DingTalk’s Yida platform before and after a visit, open the “Customer Visit” system and “Customer Management/Customer Visit Plan…” and other forms, and then manually fill in the relevant information.

However, sales are supposed to be on the front line, and the cumbersome and inefficient entry process has forced sales to reduce the time and energy spent visiting customers. For managers, if they want to know the progress of sales visits to customers, they can only access the company's CRM system, which has many entry steps and is inconvenient for data query.

To this end, with the help of the AI ​​assistant built by DingTalk, salespeople with snacks and food can complete the planning of customer visits, automatic filling in reports, inspection and execution, etc. through dialogue, which improves work efficiency and can also Set aside more energy for customer visits and service.

company managers also benefit from the fact that they can quickly track the progress of the sales team through dialogue.

For example, managers can directly ask, "Help me see how many S-level customers in the team have not been visited so far?" "Help me see how many customers xxx sales have visited" etc., and the AI ​​assistant will immediately use the indicators. In the form of cards or lists, you can feedback relevant data in the low-code system, allowing managers to see it clearly at a glance.

According to Gartner predictions, by the end of 2023, the global low-code market is expected to reach US$26.9 billion; by 2025, it is expected that 70% of enterprise digital applications will be constructed by low-code platforms.

Especially for small, medium and micro enterprises, "ai + low code" can be called the "best cp", ai is the "brain", and low code is the "hands and feet".

Youtianyoushi’s exploration of this is still in its infancy; in the future, the deeper combination of AI and low code will be “natural language issuing requirements, and AI completing development independently.” By then, any business personnel can Business needs can be changed, produced, and called in a few sentences and tens of seconds.

uses battle as training, what are the implications of AI application?

Judging from the aforementioned cases such as Belle Fashion, AI has begun to penetrate into every process and link of enterprise operation.

Enterprises are eagerly waiting for AI applications, which also confirms Microsoft's strategic choice.

coincides with Microsoft. DingTalk, which aspires to be "comprehensively intelligent", does not roll out self-developed large models. "We don't stock up on cards. We go back to the basics and let large models generate value."

Ye Jun is very sure, “Only when large models are truly applied in the processes of production, supply, sales, design, management and research and development can they be valuable."

From a horizontal perspective, large AI models can help achieve equal rights in digital intelligence, and large, medium, small and micro enterprises can find ways to apply them.

In the past, the process of enterprise digitalization depended to a large extent on the digital thinking of managers. , the digital literacy of corporate employees, the scale of corporate capital investment, etc. Zhang Qingyuan, president of

Kingsoft Office, once proposed a "28-20 rule". 80% of users only use 20% of the functions of digital tools, and users use the other 80% of functions. The problem of not understanding, not being able to learn, and not being able to use it well is particularly prominent in traditional enterprises and small, medium and micro enterprises.

The application of large AI models has greatly lowered the "entry threshold" for front-line personnel. Enterprises can "smart" change on demand.

DingTalk is also trying to lower the threshold for enterprises to "smart" change. At a press conference on January 9, DingTalk officially released the Startup Edition priced at 980 yuan/year, and Provide one year of free use support to 10,000 newly registered companies, reduce open platform commissions up to 100% rebate, etc.

Ye Jun explained that the above actions are “hope that everyone and every company can create AI with a low threshold. Super assistant, promoting intelligent inclusiveness. "

And from the AI ​​exploration of the aforementioned companies, we can also summarize the common methodologies.

On the one hand, in order to be on the high list of AI, the internal drive of the enterprise and the external promotion of the platform are indispensable.

Even the leading giants, With sufficient funds and a high-level talent team, it is unbearable to develop AI from 0 to 1. The cycle is too long and the cost is too high to be cost-effective. Therefore, many companies have chosen to work with DingTalk and Didi. Co-created by Pu and other external partners.

On August 22 this year, DingTalk opened the intelligent base (ai paas) to ecological partners and customers. It plans to use large models to help ecological upstream and downstream products to be fully AI-based, with multiple intelligent scenarios Solutions and intelligent industry solutions have been launched one after another, which is the starting point for DingTalk intelligence to fully enter the ecological layer.

So far, more than 700,000 companies have implemented intelligent applications on DingTalk.

Another On the one hand, the prerequisite for AI to be successful is actually the onlineization of data and business.

Taking the above-mentioned company as an example, whether it is Awinic’s intelligent customer service or Mitsubishi’s intelligent master, they all integrate the internal affairs of the enterprise with related industries. Professional knowledge base, access to DingTalk AI PaaS layer capabilities, can incubate and optimize enterprise-specific AI applications, so that AI can be upgraded from "universal" to "knowledgeable".

Belle Fashion can be a few steps ahead of others in the implementation of AI. , and it is also inseparable from long-term and continuous investment in digital infrastructure.

Therefore, companies that run faster in digitalization and have deeper accumulation are often more comfortable in AI transformation. Taking the DingTalk platform as an example, as of 2023 By the end of the year, there will be as many as 25 million companies running through digital transformation here, and they are at the forefront of AI exploration. The layout of

ai also has a strong first-mover effect. companies use war to train, and platforms use war. Through generation training, the accumulated large amounts of data and industry knowledge can feed and incubate AI, and the new data accumulated in the exploration of AI implementation can reversely promote the iterative upgrade of AI capabilities, ultimately forming a flywheel of positive growth.

Therefore, for enterprises For us, it is better to get involved as soon as possible rather than wait and see.