How important is classification awareness to product managers?

You should have bought something on Taobao, we say the most common shopping process:

From an Entry class page (or Broad to product), after a certain shopping guide process, to the page that requires you to make a transaction decision, then enter the transaction process.

The entrance, the most typical is Taobao homepage or an activity page, of course, for some senior Taobao buyers, his entrance may also be "my Taobao", a "vertical market Home", a collection of a "shop home", or even some SNS properties of the page, such as Taobao inside the forum, gangs, or outside the Mushroom street, Beauty said.

Then, from this entrance to a specific product Details page, also called the process of commodity detail page, can be regarded as a broad shopping guide process. Taobao can be based on whether the buyer has a clear shopping needs, with different products to guide.

Therefore, when designing a guide path, there is a clear shopping goal (for example, to buy a pair of Nike shoes, the buyer of the item number, provide search, search results page; For buyers who only have vague shopping goals (like buying a few T-shirts, a router), they provide category navigation, various channel pager; For those who do not have a shopping goal (very much like the girls who shop under the line, look around and see the good buys), they can provide information about various promotional activities.

Shopping guide of the Process page, is often a list, so also called the list page.

As shopping demands more and more diverse, in recent years Taobao began to try to shopping through the psychological appeal to classify the guide, for example, novelty, seek cheap (the list page according to the price sort is very typical of the corresponding function), for the benefit, quality, conformity (according to the Order of sales), chasing stars, etc., or the scene of the shopping guide, such as the decoration, can be in accordance with the hydropower stage, mud tile stage, carpentry stage, paint stage, soft loading stage and other guide buyers. This is a lot of learning, relying on the industry and the user's understanding.

Behind this link, there are also some buyers are not too able to feel the product, also play an indispensable role, such as marketing system, membership system, advertising system, etc., sometimes appear in front of your baby, may be based on your membership information, marketing or advertising system through a powerful algorithm to calculate out.

Next, you go to the Product detail page, browse various information on the page, make purchase decision, can not buy, can "buy immediately", also can consider, put the baby into "favorites" or "Add shopping cart", also can communicate through Ali and seller.

Then enter the transaction process, from the payment to the seller shipments to the buyer to confirm receipt, evaluation. Of course, there may be reverse processes, such as refunds, returns.

Product Manager must be a classified control

From 2003 to 2013, Taobao started from scratch, the development is very fast, to achieve the scale of tens of millions of billion a year. 2012-year Double 11, Taobao's peak turnover is 19.1 billion, the days of the cat did 13.2 billion, the market did 5.9 billion.

Technology is of course important to support these needs, but from a product standpoint, this is a system that, according to the changing requirements, is becoming more complex from a simple reality case. Next, we will Taobao on the most core of an object-commodity, also known as the baby for example, to say that the system changes.

I think that the product manager must be classified control, I would like to put all kinds of things in the home into the corresponding box, and then the box is labeled, and then Box box ... The core of commodity management, or the basis is "classification."

May 2003, Taobao just on the line, only a few commodities, you can imagine, very easy to manage, do not need to classify, nothing more than a few products spread out to the user to choose. As the product increases, the classification is involved. Because we have a lot of merchandise, we need to manage, first need to comb clear, clear to pipe, can not 1 billion commodity one tube, no meaning, "classification" method of change is the whole commodity system changes in the core.

Just start classification is very simple, called category, category is a pure classification, such as clothing, electrical appliances, mobile phones and so on. We take clothes for example, when all the merchandise to more than 100 times, clothing has 20, do not need to be divided again, when 1000, the clothing has 400, can also, but has not been very good election, if 40 goods one page, must turn 10 pages, the user page-turning patience up to 3 pages, 3rd page After the merchandise, click Very Low.

When you arrive at 10,000 items, you are silly, it is a waste. Because we found that when the quantity of goods is less, the total turnover is strongly correlated with the quantity of the goods, and the quantity of the goods here is an effective quantity that can be seen by the buyers. So, when the number of goods is more and more, to 100,000, 1 million, 10 million, we need to divide the class into finer, but, it is impossible to all tile. Imagine, a sudden show to the buyer 1000 categories to let him screen, is meaningless. Therefore, the concept of multi-level class of purpose, is a tree-like.

But the problem is: more and more commodities, the tree will be more and more deep, this is the nature of the problem, a full 4 years of trouble Taobao time.

Why? Normal user shopping path, from clothing to menswear to T-shirts, each click, users will have a certain proportion of the loss, about 50%, so you can calculate, 0.5x0.5x0.5, so that the next few users come in. But on the other hand, there are too many goods, you have to have such a point, there are many brands, such as Jack Jones, Mattel, Nike, do not separate out, it is hard to find on the T-shirt.

Moreover, this division, our small two will be more tired, very messy. We found that in addition to men's clothing, women also have t-shirts below, T-shirts are also under the brand, the United States, Granville, Nike ... After such a tree is divided, cross and overlap inevitably become a difficult network to manage.

How to solve these two problems? Until 2006, Taobao introduced the concept of "attribute". This is a revolutionary change. Why so long? On the one hand, Taobao small two business understanding has a process. Commodity management is the essence of classification, today many small two can answer, but the original can answer the few, on the other hand, the quantity of goods is gradually increasing, the problem until 2006 become a problem of the dead. For product managers, the problem is always too much, resources are always too small, what problems will be dead, the first solution. In 2004, 2005, with the category of the head down, not dead, by 2006, the number of goods has been broken tens of millions, means that this can not be divided into the way.

This time, Taobao has a legendary product manager-a lamp, put forward the concept of attributes, attributes well resolved the previous paragraph mentioned two issues. Compared with the class, the attributes are more discrete and more flexible, similar to tag.

Brand is a property, gender, color, size, material is also attributes. Attributes are valued below, such as attribute values, such as Jack Jones, Nike, Nokia, and Apple under the brand, so the nature of the attribute is to classify.

In fact, the first attribute of the class is food. Taobao to sell food, food regulatory department said food some standardized information must have, such as approval, production date, health font. So Taobao to the food category to do more than 10 attribute definitions, each product has different attributes, different property values. The earliest attribute was used to describe, not to classify. Taobao a lot of products, are published in grayscale, such as the first use of individual categories to do properties of the pilot is an example.

Back to the attribute, we talked about the class, and then there are attributes, properties below the attribute value, in fact, some property values will also become child properties. Sub-attributes are for better categorization, such as branding and sub-branding (or series). Brand is attribute, Adi, Nike, Converse is attribute value, at the same time Artie will be a child attribute, its child attribute value is "configured series (three stripes), Originals series (clover) and Style series (ball type logo)". Without the concept of a child attribute, there will be no subclass of the same situation, too many attribute values tied, resulting in the inability to filter.

Brand and series, in the reality is also a subordinate relationship, more suitable for the series hanging under the brand, such a brand has a corresponding series, another brand has a corresponding series, in essence, in order to better classify. So, as with the class, the attribute is also a tree--the attribute tree, and it has the control of the series.

Let's take a look at the management of attribute values. Have you found that property values and attribute values are actually different? Some are simpler, the attribute values are discrete, such as sex only two types-male, female, plus "neutral". Some slightly more complex, such as the brand under the mobile phone has Apple, Nokia, Samsung, clothing brands have Jack Jones, Nike, but small two difficult to complete, because this thing is constantly changing. There are more complex, such as weight, it is a continuous value, there is no way to list all. Use a table to simply represent the above three attribute values, as shown in the following table.

The value of the property is objectively different, resulting in some differences in the product, these three kinds from top to bottom, there are two points worth talking about.

First, the scale of control is becoming more and more loose, which is easier to understand. Enumeration is not input-------------sex, it's all human, we'll direct it. Enumeration can be entered--the brand, we imagine, if the brand can not be entered by the seller, the seller released the goods may not find their own brand. What happens now? It's easy to think of a solution--adding "other". But this will make us never know some brands, so you can input the role of a collection of information, because the normal property value of small two is not predictable in advance. The system through the user input behavior of a sort, more important to jump out, small two add it to the optional range, so that the property value is more perfect. But not to enumerate, is to let the sellers play their own.

Second, the classification function is getting smaller. Things that are generally not enumerable are not categorized, which means that the value is descriptive. For example, size, size, weight, this is not enumerable, because this kind of thing directly to the value of the classification is meaningless, but sometimes it may be classified by interval. For example, color, it is the same as the price, because the color change a little bit, you can separate out, but who can say more than 20 colors? It's hard, if you describe tens of thousands of colors, that doesn't make sense to users. So the product manager is doing the simplification of the action, we classify the good, such as red, blue, gray, and then a little more, sky-blue, Sky Cyan, and then add a "other", so as to divide, an enumerable attribute into an enumerable interval.

With these attributes, how to hook up with the class?

First, properties can only be hung under the leaf category. The non-leaf category also has a need for attributes, but that is done in a different way, when it comes to public properties.

Secondly, the class and attribute are related to the relationship. For example, for T-shirts This category, brand this attribute must be required, the material is also necessary, sex does not need to assume that the father of the T shirt is menswear, color is required, the size is also needed ... We found that some attributes are suitable for this class, and some attributes are not suitable for this class, so we have established a very important relationship, that is, the class-head attribute. This is Taobao's proprietary vocabulary, the category and attributes are related to the relationship, each category has its own unique attributes.

So, how are the categories and attributes related to the commodity? In fact, it is very simple, the seller released the goods, first from the root directory to the leaf category, and then select the leaf category corresponding to the class of properties of the property values.

At this point, our description of the entire product is very complete, but also very perfect, with attributes, we can easily handle 100 million of goods. But the next emerging problem is more serious.

Product managers need to solve problems rather than optimize products

We continue to use the branding attribute as an example. As Taobao's small two, everyone can say the brand is very limited, generally only dozens of, can say hundreds of are already industry experts, but the whole China and even the world collection of brands at least tens of thousands of, how to do? How to choose the consumer? means that the attribute value of the brand must be further subdivided. Small two of course is also very clever, on the "Men's brand", "Women's brand", "mobile phone brand" and so on different attributes.

Then, the men's T-shirt hanging in the men's leaves category T shirt below, women's brand T-shirt hanging in the women's Leaf category T-shirt below. Different kinds of purpose operation small two because of the relationship between the division of labor, each maintain their own class purpose brand attribute value, this brings a few problems.

First, not the same kind of brand this attribute, but the database ID inconsistent.

Second, the attribute value under the brand is not unique. For example, the men's "Nike" in the database has a separate value, women's "Nike" in the database there is another value, the attribute value is not returned. This is messy, originally Nike, but men's Nike and women's wear Nike is not related to the front guide, this is very weird things.

Third, the same "Adidas", different small two definitions of the format, can be Adidas, Adi, Adidas, Adidas and so on.

If the above problems are allowed to exist, based on brand management can not be done, so we did "attribute to one" thing. This matter will be divided into two pieces, one is the history of the problem, one is the future.

First of all, history. At the time we listed some attributes that needed to be grouped, for example, "brand", through this keyword to find, and then look at all the attributes inside the "brand" meaning of attributes, these attributes merged, and under an ID, and then the following attribute values are also merged to solve the last two problems. As you can imagine, we first have to consider "which attributes need to be attributed to one, which attribute values to do one", and then the next step, this is a very large human flesh work.

Even more tragically, this can only solve the historical problem, and there is no way to solve the future.

When we put "brand" human flesh finished, the end of that moment, brand and brand value are very "clean", and in order to avoid the third problem, the brand value we also made a specification, English in front, Chinese in the back, and then separated by a backslash. However, there are tens of thousands of property values under the brand, let the small two in tens of thousands of attribute values to pick out, he will wearies, so many small two or "illegal" operation, he may directly enter a, according to their wishes, such as "nokia/Nokia" has been, he will directly enter a Nokia, Then hang the value into his own class-like attribute. That is to say, there are two more Nokia, so we have to clean it regularly.

This problem is difficult to solve with products, for many reasons, such as the complexity of the brand, there may be some brands do not have English or Chinese. So, we later used a look silly scheme--the entire Taobao brand management, allowing the brand to enter only one small two (as of the beginning of 2012). Who is it now? How many people are there? Or is there a better way? From these examples we see that in a particular scenario, a problem does not have to be solved in the traditional sense of "developing product function", and the responsibility of the product manager is fundamentally to solve the problem rather than make the product.

But, with this full-time small two, still need to clean up regularly, why? Because it is relatively possible to add a brand by the small two, but there are a lot of brands through the enumeration can enter the way in, this is to enrich our brand library. Once these sellers enter the brand below the number of goods to a certain extent, we need to put these brands into a small second recognized brand value inside, so this cleanup is a continuous action, until 2012 still doing.

Brand This property is more critical, it may also be human meat, while other attributes, a lot of the chaos is placed there. Of course, in the future we want product managers to be able to solve the problem completely in a systematic way.