Dialogue on DingTalk: How to build a super AI application?

The key to super applications is to be able to integrate and replace multiple applications, which is naturally consistent with the characteristics of large models.

How to gradually unify the experience and catch the big model express is a huge and bloated product problem faced by DingTalk. In the past year or so, DingTalk has made many choices, deletions, and reconstructions to improve its product architecture. Nowadays, in the topic of intelligence, DingTalk seems to have become sexy again

The fundamentals of DingTalk are ToB, but it also requires user experience. “Customers are ToB, and users are ToC. DingTalk naturally has both ToB and ToC attributes. When B-side employees use DingTalk, they also have demands for ToC experience and convenience. DingTalk needs to break through from its original single point. Systematic upgrade." said Qi Junsheng, chief product officer of DingTalk.

Whether it was the burden of the past or the pressure of commercialization, DingTalk had to make a major product innovation. What’s interesting is that when DingTalk basically completed product reconstruction, it ushered in a large model. Ding CEO Ye Jun also proposed the goal of comprehensive intelligence, and product reconstruction has become a natural topic in comprehensive intelligence.

In the past, DingTalk was already a super application. Next, DingTalk wants to become a super AI application.

Before proceeding with intelligence, top-level design must first be carried out

In the early years, DingTalk’s internal team adopted a PO (project leader) system. Each person in charge led a team to jointly expand the market, with the clear goal of increasing the number of users. The team's combat effectiveness is also very strong, which has laid a solid foundation for DingTalk - with hundreds of millions of users, a large amount of traffic and application scenarios

Under the conditions at the time, the PO system helped DingTalk complete its primitive accumulation, which had positive strategic significance. However, things changed at any time, and DingTalk's scale reached a certain level. The top-level design and experience consistency of DingTalk are greatly challenged, which also explains why DingTalk was very complicated in the past, because each team was building it themselves and then stuffing it into DingTalk. After rewriting: Under the circumstances at that time, the adoption of the PO system to help DingTalk carry out primitive accumulation had positive strategic significance. However, as time went by and DingTalk reached a certain scale, the PO system operated in an independent manner, which caused great challenges to the top-level design and user experience consistency of DingTalk's entire product. This also explains why DingTalk in the past was very complicated, because each team was developing independently and then adding it to DingTalk

Ye Jun, the president of DingTalk, is also a product manager. In early 2022, DingTalk decided to focus on value. He believes that DingTalk's product design should return to the perspective of customers and users. After all, DingTalk currently has more than 600 million users and tens of millions of customers. If you still use a point-by-point approach to solve problems, the rate of return will be very low

Qi Junsheng also joined DingTalk under this background. He first needs to solve two big propositions. On the one hand, it is the consistent design of DingTalk product links. From a customer perspective, although DingTalk integrated many functions in the past, customers did not have a good experience using it. Taking the personnel scenario as an example, it integrated many types of third-party partners including training, OKR, performance, etc. There is little difference between opening these functions on DingTalk and opening them on different APPs. DingTalk wants to solve the problem of opening up different functional modules on the same interface. In particular, DingTalk has many ISV partners, making it extremely cumbersome and complicated to implement different functions into a consistent experience.

On the other hand, DingTalk is gradually starting the commercialization process, which is more testing the top-level design. Only with product sustainability can there be long-term profit margins. “DingTalk has been expanding its subscription-plus-usage commercial product matrix. On the application side, PaaS, SaaS or the product architecture of the two combined, when customers use DingTalk, whether it is secondary development, combined with ISV capabilities, or using our Our products should have a clear and consistent experience, and the cost of getting started should be significantly reduced," Qi Junsheng said.

Take the product suite as an example. To achieve a consistent experience between products provided by third-party manufacturers and DingTalk’s own products, data must first be connected through DingTalk and all the same fields can be reused instead of uploading or re-uploading them back and forth. Through the DingTalk call interface, in terms of product experience, it is ensured that users only have one experience when using it, rather than an obvious sense of fragmentation.

"Only DingTalk can do this. Vertical SaaS is difficult to do, because verticality also means that it can only solve problems point-to-point. It is truly oriented to the many needs of customers. Only DingTalk can meet a large number of scenarios. This is also The value of DingTalk lies in it." Qi Junsheng said, "I think SaaS plus collaboration is the absolute direction to go in the future."

Before shouting the slogan of comprehensive intelligence, DingTalk had already made a clear plan for the top-level design of the product and has been working hard in this direction. Under the guidance of the top-level design, including the underlying PaaS, data platform, connectors, and connections to application scenarios, DingTalk considered factors such as the building's foundation, room configuration, and even soft decoration in advance

The strong wind blew towards DingTalk

Just when DingTalk was trying its best to solve the complexity of functions and scenes, the arrival of large models provided an imaginative way.

In the past, application software interaction was mainly dominated by graphical interfaces (GUI), and the model of clicking menus step by step became the standard for actual experience. The biggest impact of large models on the interaction layer is the language interaction interface (LUI). The hybrid interaction form of GUI and LUI can solve DingTalk's interaction and product complexity problems. Of course, DingTalk still needs to continue to reconstruct and transform its products

"Don't think of interaction as simple," Qi Junsheng said with emotion. LUI relies on the ability to connect with the underlying system. Users use language to issue an instruction, and the underlying system must have the ability to accept instructions and make changes. , is a challenge for many existing applications. In DingTalk’s view, intelligence is a huge opportunity to solve DingTalk’s current problems.

Rewritten as: DingTalk naturally has rich enterprise business scenarios. By connecting with the intelligent underlying capabilities and application scenarios of large models, it can realize the continuous interaction between enterprise scenarios and large models, thereby fully mobilizing various resources such as data. Make the value of the application more abundant and concise, forming a virtuous cycle of product value. Therefore, the core of DingTalk’s overall intelligent design also revolves around activating the digital assets of corporate customers and solving actual corporate business problems

In addition to rich application scenarios, DingTalk has also accumulated a large amount of user and customer enterprise data. The power of large models is not only to analyze existing data, but its advantage lies in its ability to integrate data from different fields for reasoning. It is no longer just data analysis of a single dimension, a single scenario or a single system. This also echoes DingTalk’s open strategy. More functions meet more scenarios. Data from different scenarios are connected and integrated to provide a better large model experience. Based on DingTalk’s intelligent product architecture and design, it can significantly Reduce the threshold and cost for enterprise customers to apply intelligence.

Enterprise digital assets and numerous business application scenarios can be found on DingTalk. When enterprise customers encounter situations where large models are difficult to access, DingTalk provides a solution. From conversational chatting to creating applications and to-do items, and even various collaborative office documents, they have all been integrated into the DingTalk platform. Even if customers require more in-depth training or fine-tuning, DingTalk’s open architecture supports the introduction of industry expertise

“In the era of intelligence, DingTalk provides better connection capabilities, base capabilities, and application scenarios, which is what DingTalk is very good at. From the DingTalk side, you can see and promote it more clearly, and intelligence is from the depth From the testing stage to the in-depth value creation stage, we have a very simple goal, and the 'magic wand' must solve the most basic and key issues that customers have the strongest demands for," Qi Junsheng said.

Enterprises can now use the magic wand on DingTalk’s homepage to evoke nearly 20 AI skills in five products including chat AI, Yidai AI, document AI, intelligent Q&A, and consulting AI. At the same time, you can also click the magic wand button on 17 product interfaces such as Yida, Documents, and Meetings to use the AI ​​skills of the corresponding products

The simple needs of enterprises are nothing more than operation, efficiency improvement, cost reduction, etc. Wen Shengwen and Wen Sheng diagram are closer to the logic of ToC and are the stage of early market education. DingTalk must focus on actual work scenarios and generate commercial value.

Super AI Application

In the "Important Strategic Technology Trends for 2023" released by Gartner, super applications have received special attention. A super application is an application that integrates application, platform and ecosystem functions. It not only has its own set of functions , and also provides a platform for third parties to develop and publish their own micro-applications. Gartner predicts that by 2027, more than 50% of the world's population will be daily active users of multiple super applications.

Alipay and WeChat are typical super applications. Super applications are spreading from China to Western countries and being copied and imitated. Musk previously expressed his desire to create an everything app after acquiring Twitter. The key to super applications is the ability to integrate and replace multiple applications used by customers or employees, which is a natural fit with the characteristics of large models.

Recently, Kai-fu Lee, the founder and CEO of Zero One Thousand Things, specifically mentioned super applications when he released a large model. He mentioned that starting from simple applications, and then continuously iterating through lean entrepreneurship methods, just like Dou The first version of Yinhe WeChat is not a super application, but captures the needs of users, and uses the technical essence of the new platform to make a simple application that everyone likes, and then continuously adjusts it based on user feedback, and finally iterates into a super application. Application, this is Zero One Thing’s methodology for making super applications.

Kaifu Lee’s super application started iteration from scratch in the To C market, while DingTalk focuses on the To B market and already has ready-made scenarios and users. Although the two are different, there is no obvious difference between them.

It is understood that as of the end of October, more than 500,000 companies have joined the DingTalk AI Magic Wand invitation test. Currently, 17 products including DingTalk Chat, Documents, Knowledge Base, Brain Maps, Flash Notes, and Teambition are connected to AIGC. Online, fully open to users for testing. In the latest entrance "Magic Wand" on the DingTalk client and APP homepage, you can also use chat AI, document AI, Yidai AI and other functions for natural language conversations.

TMTpost learned that during the beta test period, more than 6% of large enterprises with more than 5,000 employees frequently used AI, and more than 9% of medium-to-large enterprises with more than 2,000 employees frequently used AI. In terms of the frequency of AI usage, users who frequently use AI use it more than 15 times a day on average, and some companies use AI more than 300 times a day on average per person. Among frequently used products, Document AI has generated 4.025 million content for users, Flash AI has generated 1.557 million summary content for users, and Yida AI has assisted users in generating 9,800 office and business applications

This is China’s first fully open national-level intelligent application, and also the first fully open AI work application. DingTalk has begun to upgrade into a super AI application

DingTalk is far from reaching the mature state of super AI applications, and Qi Junsheng still has a lot of work to do. "You still need to remain in awe and don't think of things simply. On the one hand, to do ToB, you need to have a deep understanding of various roles such as business operations, and at the same time, you need to have a clear insight into the minds and needs of ToC users." He said.

After making products for many years, Qi Junsheng realized that there are two types of products that are the most difficult to make. The first category is open products because they involve the role of ecological partners, which exponentially increases the complexity of the product. The second category is commercial products. You must be very cautious about commercial products. You cannot take them offline just because a certain link does not pass. You must take into account the many needs of different customers. Especially for paid products, you need to pay special attention to the consistency of the products.

The early preparations can be regarded as the automation stage of DingTalk, and now it has entered the intelligent stage of DingTalk. All goals are to promote product upgrades and solve the simple needs of ToB customers. "We should open our eyes and be in awe of customers' simple problems, because behind these problems there are often many problems that can be drawn from one example and draw inferences about other cases, which is closer to the essence of the problem." Qi Junsheng said

For example, logistics companies are basically in a state of low profit and are extremely sensitive to costs. For ToB managers, timeliness control is much higher than in other industries. If there are fewer shipments on a given day, managers want to make some immediate adjustments, such as removing two vehicles. In the past, traditional BI reporting methods could not meet the timeliness requirements, but with the large model, DingTalk can combine various dimensional data such as attendance and operations to estimate throughput and timeliness, thereby directly increasing gross profit

By lowering the threshold of large-scale models that represent advanced productivity, DingTalk makes it available and affordable to more people on demand. This enables enterprise organizations to have stronger predictability and decisiveness than ever before, and can optimize gross profit, etc. Various business indicators. This is the ideal change that DingTalk hopes to bring to enterprises, turning it into a super artificial intelligence application

DingTalk itself is becoming an evolutionary flywheel, attracting a large number of users through super AI applications. These users attracted the attention of partners, who provided DingTalk with more rich scenarios and capabilities. This in turn attracts more new users and existing users for further consumption. The introduction of large models expanded the breadth and depth of DingTalk in the To B scenario, and also opened up the collapsed user value

When rewriting content, the original text needs to be rewritten into Chinese without retaining the original sentence

The above is the detailed content of Dialogue on DingTalk: How to build a super AI application?. For more information, please follow other related articles on the PHP Chinese website!