From chaos to clarity, thinking about the construction of Ali's global commodity category domain

Author: Ding Haoran Alibaba Global Business Platform Team

Commodities are one of the cores of the e-commerce product system, and categories are one of the cores of the commodity model. The basic business data provided by the category system runs through the entire e-commerce system. This article will share with you the construction and thinking of the commodity category domain in the process of globalization.

As we all know, commodities are one of the cores of e-commerce product systems, and there are many basic information that constitute commodities, among which categories are one of the cores of commodity models. In the evolution history of Ali e-commerce, the category model has also evolved from the original simple classification to the current category/attribute/attribute value model, which is actually a manifestation of refined classification. The category system provides basic business data and runs through the entire e-commerce system. Throughout the process of globalization implementation, we have successively completed capacity building such as data compliance, product and technical architecture integration, local refined operations, algorithm intelligence, and multi-country commodity interoperability. Due to the limited space, this article will introduce the changes that the following categories have experienced and the domestic arrivals in terms of product and technical architecture integration and local refined operation through the comparison path of three years ago and three years later. International differences. I hope this article can give some guidance to friends in the same industry or those who are about to enter international business.

1. Understand the category system

Before formally introducing the international category system, let’s have a general understanding of commodity categories, understand the relationship between categories and commodities, and the importance of categories in commodity circulation.

1.1 What is the category system?

To put it simply, the category system is a structured way for commodities. Through classification, it is easier to complete the standardization of commodity information composition, which facilitates the standardized interpretation between various systems in commodity circulation.

In the Alibaba e-commerce category system, it mainly includes three basic models: category/attribute/attribute value.

Category is also a classification tree. Commonly used categories are background categories and front-end categories. Background categories are used to structure products, which is convenient for sellers to fill in standardized product information and is relatively stable. The front-end category is used to display to the consumer side, and will change according to seasons, activities, etc.

Attributes are actually the refinement of classification. The category tree has a depth limit. If you need to further refine some commodities, you need discrete attributes to solve them. Most of the attributes are classified differently on e-commerce platforms, but abstractly There are mainly two categories: SKU attributes and non-SKU attributes. The SKU attribute is used to define the dimensions generated by the product SKU, such as the "color" and "memory" of the mobile phone. Non-SKU attributes are used to define commodity infrastructure, such as screen size, etc.

The attribute value is the refinement under the attribute, such as A and B in the "brand" of the mobile phone, and there are differences between different "models".

Categories/attributes/attribute values ​​are related to each other, and the combination of their relationship and basic data finally forms a category system. Of course, there are also advanced models generated based on the basic data of the categories. Due to space limitations, we will not focus on them here.

1.2 Look at the whole picture of products from category

What is the relationship between the category and the product? The author abstracts the core elements of product information and forms the following Jiugongge.

The seller can effectively complete the creation of a product and its multiple SKUs by filling in the category, attribute PV, and SKU. When these elements are determined, the basic product skeleton has been built. This part is under different categories. The commodity skeletons are different, so they are more flexible.

The title, detailed description, and main image/SKU image are basically the fixed elements of the product. This part is more for shopping guides and consumers, so this part will be used more on the shopping guide side.

Price, inventory and service marks are used more frequently in transactions, logistics, contract fulfillment and other links.

After we have a basic understanding of categories, let's use a picture to see what role categories play in commodity circulation:

It is basically complete here and we have introduced the entire category system. After we have some preliminary understanding of the category system, we will focus on the construction of the category system under the international business.

2. The difference between domestic and internationalization

2.2 Uncertainty in international business

The global market is changing rapidly and will encounter some restrictions at any time. Faced with such uncertainties, the road of globalization of Chinese enterprises is also full of uncertainties. In this battlefield without gunpowder smoke, only by constantly changing and innovating can we occupy a larger market.

For sellers, they need to provide them with a better delivery experience, help them quickly bill and sell more products. For buyers, they need to provide them with better quality and more accurate products according to their needs. Therefore, product delivery experience, one-click moving, supply and demand matching, and product power will naturally arise.

At the same time, the platform side also needs to carry out global compliance upgrades according to the legal provisions of each country . One part is the compliance upgrade of database storage, and the second part is the compliance upgrade of the data itself. For example, some special The category is not allowed to be sold in some countries in Southeast Asia, so the compliance of the category needs to be considered during the operation of the platform.

2.2 The particularity of category operation

The international category operation system comes from the empowerment of the domestic category operation system, such as Ali’s ICBU international station and AE AliExpress’s old overseas business, which is mainly cross-border business. There are many Chinese sellers. However, after the group acquired Lazada in 2016, the operational problems facing local sellers and local second parties gradually came to light.

Problem 1: International and local operators have little knowledge reserves and high mobility. Category operation requires a deep understanding of the business in order to be able to operate well, but local juniors in different regions may not have so much knowledge reserve, if there is no systematic SOP and good products to guide, they will always in a state of confusion. At the same time, due to the relatively high mobility of the operation, there are also problems with inheritance. At this time, the reliance on category operation products is even greater.

Problem 2: Operational products are difficult to use and do not conform to the habits of international juniors. International category operation products are originally derived from domestic category operation products, which are biased towards the usage habits of Chinese people. Due to cultural differences and operating habits, the old category operation products have blocked the efficiency of category operation.

Problem 3: The operation process is too long and the product operation efficiency is low. An increase in category SKU attributes and attribute values ​​will lead to multiple process initiations, and it takes a week for a single process initiation and circulation. At the same time, some operations that should be transactional, after being decomposed, also brought a lot of quality problems, which may lead to online release problems after being pushed online.

3. Globalization category three years ago

Let's review the construction of the global category domain three years ago:

Lazada, AE, and Daraz are all self-contained e-commerce systems. In their respective e-commerce systems, they all have their own category operation platforms, category data services, and category data storage. In terms of category operation system, lazada and Daraz are localized operations, and there are multiple sets of category data; AE is not strong in localized operations and focuses on cross-border business, so it always operates with one category system.

The three systems of each BU come from different development stages of domestic category products. The above "differences between domestic and internationalization" mentioned the particularity of category operation and the problems encountered. These problems were introduced three years ago exposed more clearly. At the same time, due to the burden of history and the lack of professional category technical teams, each BU has been blocked in operating product changes, model innovation, and technical capability optimization, and ultimately suffered from operations, sellers, and buyers.

At the same time, due to policy restrictions, global website building will become more frequent, but which of these three sets of category technology systems will prevail, and how to quickly and cost-effectively support website building? These are the focus of our future thinking.

4. Problem analysis

We have sorted out the problems arising from the construction of globalization, and the problems presented are mainly three parts:

• Problem 1: Multiple sets of category products and technical architectures are not unified, and the business-side development team is difficult to maintain and the maintenance cost is high.

• Problem 2: The weakening of category infrastructure has brought about confusion in category data assets, which will eventually affect other domains of the e-commerce system.

• Problem 3: The cost of building a global website and operating a business is high.



Based on these three issues, we went deep into the business team, and after further analyzing and sorting out the issues, we decomposed them into the following four levels of issues.

• Technical level: The category vertical domain has relatively high knowledge breadth and technical requirements for R&D students, and many of them belong to the long-term infrastructure part, and the progress speed is slow.

• At the operational level: the local primary school needs to maintain multiple category trees, which requires high manpower and maintenance costs. At the same time, the knowledge reserves of local primary school students are not enough and the inheritance is not good, and there is no set of methodologies and products to support the standardization of categories, which ultimately leads to unclear data assets.

• Product level: Multiple sets of category operation products cannot empower each other, there are many redundant infrastructures, and poor scalability, which is not conducive to global website building.

• Interoperability level: multi-BU category system, data and models have similarities and differences, and there is no mapping relationship between data, which affects the subsequent development of the global currency business.



Based on these four issues, the direction to solve the problem has been relatively clear.

• Category operation system upgrade: from multi-country operation to global unified operation, differentiated operation by country, reducing operating costs.

• Integration of category products and technical architecture systems: Integrate multi-BU category products and technical systems to reduce the cost of technical architecture upgrades and product capacity building costs.

• Standardized governance of category data: Standardize governance of existing category data, build high-quality data assets, and provide better supplies to other domains.

5. Implementation of the plan

5.1 Category operation system upgrade

In the initial stage of international category operation, Lazada's category operation method is to operate by country. For each site, there is a set of category trees, and there is no connection between various category trees.

In terms of localization operations, when category operators add or update categories, they need to operate at least N times (N=number of sites), and the operation cost is very high. At the same time, in the new stage of cross-market sellers, sellers also need to publish multiple times on multiple sites, and the cost of publishing is very high, which greatly hinders the efficiency of new products. Based on these two points, there is the consideration of global unified operation.

After cleaning and aligning category trees in multiple countries, we have completed global unified operations, and Lazada category operations have entered the second stage.

We regard each site as a sub-market under the unified category tree, and keep the differences of each site. Operators only need to configure the category once, and they can be automatically distributed to different markets. The change from multi-country operation to unified operation , greatly improving the efficiency of category operations.

By unifying the category tree, sellers only need to edit once, without re-selecting categories for each market, and can complete multi-country sales, which greatly improves the efficiency of listing products.

5.2 Integration of category products and technical framework systems

The products and technical architectures of multiple BUs are quite different. At the same time, because of their historical burdens, it cannot be achieved overnight, so the integration can only be completed step by step. The whole fusion is divided into four stages:

The initial stage of integration: that is, the stage of globalization three years ago, when three sets of operations and three sets of service systems coexisted.

The first stage of integration: the premise of product system integration is to integrate multiple sets of category data services, so the integration of three service systems needs to be completed in the first stage of integration. This part involves the unification of the underlying DB model, and the middle layer category packaging/push package The unification of the upper client data consumption service.

The second stage of integration: After the category data service is unified, it is necessary to unify the category products by combining the business. In the process of unification, for the local operation mode of Lazada/Daraz and the cross-border operation of AE/ICBU Therefore, it is divided into two stages to complete the unification, and the second stage of integration mainly focuses on the unification of partial local operation products

The final stage of integration: complete the unification of cross-border operation products and local operation products.

5.3 Category data standardization governance

Category data standardization governance is a relatively big proposition, and it is also something that various e-commerce platforms are committed to accomplishing. In the process of promoting the standardization of category data, we encountered two major challenges.

Challenge 1: How to establish a continuous optimization mechanism to ensure the continuity of standardization.

The standardization of category data is a long-term, long-term thing that requires many teams to cooperate. How to set goals and make everyone agree with this goal and establish a continuous optimization mechanism is extremely challenging. Therefore, the standardization of category data must be a top-down matter, which requires the approval of the bosses and their willingness to invest in this long-term value, so as to ensure the continuity of standardization.

Challenge 2: The trade-off between category data standardization and seller experience

The platform side hopes that its own data will be more standardized to facilitate the accumulation of better quality data assets. At the same time, the combination of good data and search algorithm side will be better, so this part has always been the focus of the platform.

With the enrichment of category data and the improvement of standardization, the ultimate value is still on the product, so it is still necessary to cooperate with the seller to complete the standardization of category data in the product. But when the standardization comes to product release, sellers have more release constraints, and the delivery time becomes longer, and the delivery experience will naturally decline. It is very challenging to establish a mechanism to balance standardization and seller experience.

How to standardize the category data? Before establishing the category standardization mechanism, we completed the sorting as shown in the figure below:

The standardization of categories is divided into two parts, one part is the optimization of its own structure, that is, category splitting/merging/moving. For example, the male and female attributes of clothing need to be upgraded to categories and become women’s clothing and men’s clothing categories. This Sometimes it is necessary to refine the categories of clothing, so there is a way to split the categories. The other part needs to be completed in combination with product growth, which includes product diagnosis and product management. The general logic is to use the category misplacement algorithm to complete the cleaning of the product dimension, and provide sellers with the ability to batch adjust misplaced products through product diagnosis, allowing sellers to make batch modifications.

The standardization of category attributes should be a relatively complicated matter in commodity standardization, and it is also the thing that has the greatest impact on sellers. The influence of sellers is mainly filling efficiency and filling power. According to this influence, we decompose it into three steps:

• Step 1: Through the introduction of SPU standardization, improve the attribute recommendation and automatic filling capabilities of new products, and reduce the difficulty of filling for sellers.

• Step 2: Through the intelligent evolution of attribute values, automatically combine the custom text entered by the seller with the attribute value without changing the seller's publishing efficiency.

• Step 3: Provide batch tools through commodity growth, combined with AB measurement, to accelerate the filling of category attributes in commodities.

6. Summary and Outlook

In three years, the international category system has changed from chaotic to clear, which is inseparable from the close cooperation of operations, products, and technologies. At present, we have completed the second phase of integration, achieving the integration of localized operation products and category technology integration. In the next fiscal year, we will focus on completing the integration of local/cross-border products. After the integration is completed, the construction of Ali's global category domain will also be clearer and more complete. Complete site construction at a lower cost.

In the future, global commodity interoperability will be one of the key points of globalization, and the most important part of commodity interoperability that affects the success rate and accuracy of commodity distribution is category and attribute mapping. We will complete the category through product + data + algorithm Coverage and accuracy guarantee for attribute mapping.