Image source: generated by Unbounded AI

"90% of customer service problems can be left to AI." Talking about the role of large models in the company's express delivery business The first implementation attempt - AI customer service, said Li Chaoming, head of the Express 100 Production and Research Center.

The express delivery industry, which has been tortured by falling profits, started the first battle with "anti-involution", and large models have become a key part of it.

10000000000, this is the tens of billions of business volume achieved by each of the "three links and one extension" in the first half of 2024. The express delivery industry driven by "buy, buy, buy" has driven the order volume growth of major express delivery companies.

However, while the order volume in the express delivery industry has expanded significantly, the "explosion of orders" has not made the express delivery companies make a lot of money, but has lost their trump card. The financial report shows that in November this year, the single-ticket income of Yunda, STO and YTO were 2.03 yuan/piece, 2.08 yuan/piece and 2.29 yuan/piece respectively, down 14.71%, 5.45% and 6.96% respectively year-on-year.

“The express delivery industry has entered a stage from low prices to high quality services.” Chen Dengkun, general manager of Express 100, gave his judgment.

In 2023, the express delivery industry will take aim at AI. From the early days of waiting and watching, testing the waters, by 2024 the large model will begin to produce preliminary results in the express delivery industry. Although the application capabilities are still limited, the large model first covers a number of the most basic and effective businesses in the express delivery industry that can reduce costs and increase efficiency: AI customer service, AI marketing, and AI assistants.

In 2024, in addition to regular operations, large models are penetrating into the hinterland of the express delivery industry.

Starting from simple shipping and checking, to building a "knowledge base" for couriers, to helping to summarize and organize business information, and even to intelligent control of the supply chain, large models are used in express delivery The capabilities of the industry are gradually being released.

Express delivery companies that choose privatized deployment models and self-developed large models all believe in one thing: large models are worthwhile long-term investments, and there is still a vast space for its application in the express delivery industry to be explored.

With large orders and fierce competition, the express delivery industry embraces AI to increase efficiency

< p>On August 8, 2023, Li Chaoming clearly remembered that this was the first day when Express 100 decided to access the large model. At that time, Kingdee Group, which had incubated Express 100, had just released "Sky GPT" at its 30th anniversary conference. The group's strategic resources began to be invested in large models, prompting Express 100 to begin trying to apply AI capabilities to its own business.

When the large model capabilities as a base are broken through, the practical application of AI+ express delivery also opens up more scenarios.

Li Chaoming recalled that in October 2023, after Baidu released the Wenxinyiyan 4.0 large model, during the test of cooperation, Express 100 discovered, some scenes that could not be completed by AI in the past have achieved breakthroughs in capabilities with the support of new model capabilities.

Within two years, many companies have submitted their own answers: SF Express’s self-developed large-scale logistics decision-making model “Fengzhi” and the large-scale “Fengyu” model focusing on the vertical field of the logistics industry, Yunda launched AI Express Assistant and Express 100 chose the "Baidi Cloud GPT" model that mixes public cloud and private cloud models.

However, from the perspective of several core links in the express delivery business (collection, sorting, transportation, and delivery), the application of large model capabilities is still limited. Li Chaoming pointed out that at present, the implementation of the express delivery industry is still limited to the two links of "one end and one end", that is, collection and delivery.

The first products to be implemented in the express delivery industry were AI customer service products. It focuses on the after-sales link, helping to save manpower while solving a large number of repetitive and simple problems. Talking about the benefits brought by AI customer service, Li Chaoming said that although the express delivery business maintains rapid growth, after applying AI capabilities to the customer service link, Express 100 can effectively control the rate of personnel growth.

Li Chaoming revealed that Express 100 currently achieves 90% of customer complaints being handled by large models, and the one-time solution rate is as high as 99.4%.

AI assistant products are more used to help workers in the express delivery industry, including delivery boys and internal users of enterprises, who are all beneficiaries of the large model. In the package pickup process, AI assistants can help service personnel in the express delivery industry solve problems, such as prohibited item inquiries, express delivery timeliness and other needs.

Jiang Shengpei, technical director of large models of SF Technology, gave an example to Guangcone Intelligence: In the international express delivery scenario, D197 (the number of Musang King durian), Musang King, durian candy, and freeze-dried durian, they belong to Consignments of the same category? What are the shipping rules for land and air shipping to different countries? Which service is more appropriate?

In the past, these details required couriers to check on the customs websites of various countries, and they had to overcome obstacles such as multiple languages ​​and frequent changes in rules. Now they are handed over to AI assistants equipped with logistics knowledge bases. Just by taking out their mobile phones, SF Express's delivery boys can check the international shipping rules for different items on the company's internal AI assistant and handle customer needs immediately.

In fact, AI can not only empower couriers. In international business delivery services, AI assistants can also improve users' efficiency by carrying corresponding knowledge bases. The AI ​​assistant can call up customs mailing policies around the world and then synchronize to generate the latest mailing guide to help users understand changes in overseas policies.

The courier boy uses his mobile phone to check information

Starting from cost reduction and efficiency improvement, using AI customer service as the starting point, to AI applications for employees, in 2024, large models will gradually enter express delivery The hinterland of the business has been embedded in the entire core express service process of “sending, managing and checking”.

Full-process service system for placing orders, receiving orders, door-to-door pick-up, billing, payment, transportation, and signing., AI can provide capability support for every link. Take Express 100's "automatic redispatch" function as an example. In the return scenario, through the system's preset rules, AI can determine the order status in real time and allocate transportation resources as needed.

For example, if the courier times out after the user places an order, AI will immediately push a notification to the user asking whether to accept the rescheduling. After confirmation, the system will allocate transportation resources to take over; during door-to-door pickup, if If the courier fails to arrive on time, AI will also rearrange the pickup plan for the user.

AI’s explosive changes to the express delivery industry are not very drastic in terms of physical sensation, whether from the courier or the user, but they are just right to provide many active and natural language interactive experiences, just like spring rain moisturizing things. Silently. Although it looks calm on the surface, the underlying technology layer is undergoing earth-shaking changes.

Big models are in-depth, a difficult year in the industrial hinterland

Why in the past two years, there are still "a few" express delivery companies that can apply big models in-depth group"?

The express delivery industry involves a large number of people, a large number of express items, and complex links, but the efficiency requirements are extremely high. Therefore, when large models want to deeply transform such a traditional industry, the first step is to create a large model base belonging to the express delivery industry.

At present, companies in the express delivery industry mostly choose two ways to access large models: one is to serve enterprises based entirely on self-developed large models, and the other is to jointly use public cloud models + private deployment models. used together.

Companies such as SF Express and Yunda choose to develop large models themselves. Talking about why he chose the path of self-research, Jiang Shengpei said that in the process of using the open source model for privatization deployment, because it did not have relevant knowledge in the logistics industry, its actual implementation effect was not ideal. SF Express has the data and computing power. , Self-development of large-scale industry models is the only way to go.

“For example, if the word ‘little brother’ is given to a general model, its understanding will definitely be different from that of a trained vertical model.” Jiang Shengpei gave an example.

Out of consideration for the balance between effect and cost, Express 100 chooses public cloud + private cloud deployment, and calls it according to the focused needs of different scenarios.

Talking about why they chose to abandon self-developed large models, Express 100’s idea is to deploy multiple large models that are constantly evolving through deployment capabilities, plus large models trained with its own massive data and deployed in private clouds. , which can not only take advantage of the capabilities of various models, but also use large models deployed privately to control costs.

Li Chaoming said that each model has its own strengths. For example, Wenxinyiyan 4.0 performs better in intent understanding and content generation and extraction. In terms of multi-modal understanding and recognition, Tongyi Qian The VL model is better, and the Wisdom Spectrum model is ideal in identifying customer intentions. The combination of multiple models can better utilize their respective strengths.

Selecting the deployment method is the first step in the long march. If you want to train a large model in the field of vertical categories, for specificTraining with data is a big problem.

The Logistics Technology and Equipment Committee of China Communications Association pointed out that in the process of training and optimizing large models, high-quality data is the core production factor. If you want to train an industry large model with extremely high accuracy, what is needed is not Simple public data is industry-specific scene data, while the real interaction data required by the logistics industry has small samples and uneven distribution, making it extremely difficult to obtain.

Jiang Shengpei gave an example, “To build an industry model, you first need to collect corresponding data for various specific scenarios. However, a lot of knowledge and information in the logistics industry is highly fragmented, and a lot of data may be in It is stored in a certain computer or accounting book, so it is necessary to use a large model to complete the comprehensive digitalization process, allowing the large model to process rich media information, and through knowledge A series of operations such as extraction and cleaning are transformed into corpus that can be trained.”

Secondly, in order to combine the capabilities of the large model with the express delivery business, how to make the large model accurately understand all the elements in the problem. , such as "address", "time" and other information, and accurately call the corresponding timeliness, price and other information according to demand, which is still a challenge for large models.

Li Chaoming said that although the industry is currently discussing more about how to use AI to drive business, when applied to real business processes, there are still many problems with large models.

In Express 100’s business, it can be boiled down to two points:

First, in the process of interacting with AI, it must be ensured that it is in the business process throughout the process. For example, "send express delivery", no matter how the user talks, AI must ultimately guide the user to send express delivery.

“Can something be sent? How should it be packaged? If you don’t have a business process set up, the big model will keep explaining these issues to the user, and it will not guide the user to continue sending it. Express." Li Chaoming said, "The big model should have guided users to put their contact information and "Second, when dealing with user needs, the large model needs to use the business knowledge accumulated within the company. For example, when a user comes to send a courier, the AI ​​needs to know what is being sent. Only by having knowledge about the unit price of each express delivery company and accurately calling it can we accurately estimate the price of an express delivery order.

However, during the calling process, large models often get confused. Either they don’t know what parameters should be called, or they call the wrong information, resulting in wrong results.

In addition, in order to train a large model for commercial use in the express delivery industry, the large model needs to be able to take into account the understanding ability of general models, and it must also fully understand the professionalism that models in vertical fields should have. Knowledge, complete the call to content.

For the express delivery industry with more precise needs, it is necessary to provide products that can be guaranteed to be commercially available before they can be truly applied in business scenarios.

In this regard, Express 100’s choice is to first arrange a set of business processes for the large model and control the calling requirements of the large model in the process.

Going back to "sending express delivery", when the business process is well arranged, the large model will continuously "Push" the user to provide the required shipping information. In addition, it can also query the shipping prices of different express delivery brands based on demand, and regenerate a reply after sorting it out to help users find cheaper and more cost-effective shipping options.

In this year’s process of implementing large model capabilities in the express delivery industry, as large model capabilities are gradually improved, companies are also beginning to try to make AI the leader.

By cooperating with large-scale intelligent agent platforms, express delivery companies have also begun to launch various intelligent agents. In response to the demand for express delivery inquiries, Express 100 launched the "Check and Consignment" intelligent application of the same name on Baidu, Tencent, and MiniMax, giving the dominance to the big model.

Through this cooperation method, the agent can not only remember the context of interaction with the user, but also solve some problems in the business process through large model capabilities. For example, when querying express delivery, you do not need to know the specific information to find the corresponding express delivery company platform. You only need to give a tracking number to find the information; the agent can also "complete" based on the input address information, automatically identify the incomplete information and correct it so that it can be The correct address identified by the courier.

Intelligent agents help users check express delivery

As the capabilities of large models are further unlocked, the express delivery industry will be able to hand over more complex tasks to AI.

In order to reduce costs but also increase growth, AI is revolutionizing the express delivery industry

When all parties in the express delivery industry are applying large model capabilities, enterprises The service experience of express delivery for individual users has "qualitatively changed".

Compared with the past delivery of express delivery, people often need to fill in the form in person and complete the information step by step; now, users only need to use voice, text or pictures to provide relevant information to the large model, Li Chaoming expresses 100 AI Taking express delivery as an example, "Tell the big model 'I need to send a courier to Zhang San', and then take a picture to tell it what item to send, and you can realize the real "send courier in one sentence/picture"."< /p>

In addition, large-scale products with high demand for passing and shipping When it comes to business docking, companies currently have two main pain points when sending mails: First, employees’ mailings are not registered in the system, leading to chaotic management; second, when companies need to reconcile, the cost of mailing and express delivery needs to be calculated one by one. Finance comes with a huge workload.

In this regard, Li Chaoming said that Express 100 is currently trying to combine large model capabilities to develop a "Ctrip Business Travel"-style express delivery business service for enterprises, and implant it into the enterprise's internal office platform to help Enterprises improve management efficiency.

After two years of exploration, the advantages of large-scale model implementation in the express delivery industry are now emerging, and one of the most direct results is cost savings.

In addition to the AI ​​customer service mentioned above that can reduce the service costs of the express delivery industry, AI can also help corporate customers who send express delivery reduce administrative costs.

“ManyEnterprises all have demands for cost reduction," Li Chaoming said. "For example, it costs 10-20 yuan for employees to send a piece of paper. However, if we can use our 'express delivery' aggregation service, we can intelligently choose different express delivery brands according to different items. , companies may save several times express delivery costs. ”

For express delivery companies that urgently need to improve revenue and service quality, they not only need AI to help reduce costs and increase efficiency internally, but also It is even more necessary to further improve the efficiency and quality of express delivery services to achieve revenue growth.

G7 Yiliu has developed an intelligent assistant that can intelligently receive orders, shortening the order receiving time from 2 hours to 30 minutes, improving efficiency. As high as 75%; Lalamove uses AI capabilities to launch a "contraband identification" function for drivers, which can identify prohibited items in as fast as 1 second; the AI ​​assistant developed by SF Express for couriers can save an average of 3 minutes per conversation.

Effective After the efficiency increases, large models can become a "sharp tool" for express delivery companies to increase revenue.

With the assistance of AI capabilities, when the efficiency of checking and shipping items increases, the ability of express delivery to run 100 orders will also increase. Upgrade: from 2023 From 300,000 orders per day at the beginning of the year, to 500,000 orders per day by 2024, Li Chaoming predicts that with the assistance of large model capabilities, large models can achieve a breakthrough of 1 million orders per day in the next year.

< p>Send multiple couriers The company's service access platform, Express 100, has become an express ecological entrance that integrates "checking and custody" services:

Based on the Baidu Cloud domain knowledge base, users can not only check express delivery, but also compare it in one sentence price to help enterprises and individual users find more cost-effective prices; AI can also automatically help users complete address information to avoid problems in express delivery operations. Chen Dengkun shared that through the MiniMax large model, Express 100 has solved 98% of the problems. Address completion problem.

At present, the capabilities of large models have not been fully released. For example, on the supply chain side, there is still plenty of room for large models to be used. In the express transportation link, large models can now participate in fewer parts,” Li Chaoming told Guangcone Intelligence. “But in the entire express logistics industry, what needs to be optimized most is the in-depth integration of the industrial chain. Therefore, large models play an important role in express delivery. There is still a lot of room for industrial applications. ”

Until large model capabilities are fully applied to every corner of the express delivery industry, companies in the express delivery industry still need to continue to replenish ammunition for this long battle.