Teach you to optimize the crowd through direct train to create dimension labels

According to industry data dimension analysis, compared with the average level of peers, when your conversion rate can maintain an average level, the platform will allocate natural traffic. If it continues to maintain, then more will be allocated. When your traffic reaches a certain threshold, and if the conversion rate can be maintained to the excellent level of the industry, then natural traffic will enter the pool, which is the node of explosion. This is how ordinary popular products come up.

So the question is, in addition to the innate conditions of the product itself, how should we operate it in the later stage to maintain the conversion rate scientifically?

In Taobao's official definition, every product, store and even user behavior will be tagged. This is a scientific rule summarized by relying on Taobao's big data and cloud computing ranking algorithms, through the personal browsing behavior of users, it can improve the accuracy of traffic and improve the conversion rate of the store.

This tag behavior will have certain differences in the initial data situation, but it can also be adjusted through the direct train to optimize the population.

Similarly, the accuracy of the crowd label will also affect the data of the express train. If the crowd is not accurate, the click rate of visitors will not be good, and the conversion rate will not be good. In this case, the keyword quality score cannot be well maintained, and the PPC will be high, and the overall account weight will be pulled down, and the search traffic will naturally be affected.

OK, after getting this point, let’s continue to talk about it if you go through the direct train to optimize the crowd.

1. The direct train test group

The functions of the direct train are now very complete. After driving, the crowd can keep up with the operation. Now there are more data for the crowd than before, so it is more operational.

If these population dimensions are combined alone, how many people tag combinations can there be? I think it's not difficult to combine all the combinations, but how many can it be calculated by looking at the express train? That must be controlled within 100.

So we need to filter these 100 combinations to meet two indicators

1. You need to have traffic after combining

Second, label it well and the data must be beautiful

Then we will combine and split, find a group of people that suits our store stage, and determine the current relatively accurate population data to provide premium ratios to maintain data indicators, gradually extend it, and re-enabling new populations, and gradually make the label accurate. Do you think the system will not search for traffic for you?

The freezing of three feet is not a day. It does not mean that data can be operated immediately by starting to optimize. You should pay attention to the accumulation of data in daily life. This will immediately reach the Double Eleven node, and the current operation is still relatively critical.

2. The direct train crowd tag

The people on the express train are divided into five major sections: baby targeted people, store targeted people, industry targeted people, basic attributes targeted people, and Damodisk people. Small components can also be subdivided under the section, such as the basic attribute population, the population attribute population, the identity attribute population, the weather attribute population, the Taobao attribute population and the holiday attribute population.

This is a change in the past month. Compared with the previous segmentation of the population, it will be more refined. This may be a bit complicated and can be divided into two categories in a general way. The first category is independent groups, which cannot be combined or split. The second category is freely grouping people. This type of population can be freely combined or split.

It can be seen that there are many types of people, so many novice businesses may listen to classes or visit forums, and everyone will say that the crowd is important. In the era of tags, there are thousands of people and thousands of faces and a series of similar words. Then, after screening the crowd alone, I found that, hey, why has my traffic still declined!

3. Various groups of people

Let’s start with the freely grouping of people. The labels of this type of people will be more detailed than before. There can be many combinations, and we define this as first level, second level, etc. So how do we combine them specifically?

It can be seen from the picture:

Level 1 population: age

Secondary population: age + gender

Level 3 people: age + gender + average monthly consumption amount

Four seasons: age + gender + average monthly consumption amount + category comparison unit price

Of course, these orders can be disrupted according to the store situation, and there is no clear regulation on which factor must be composed of people at the level. For example, the second-level group: the average monthly consumption amount + category comparison unit price. Or second-level group: gender + monthly consumption limit. And all of these can be combined irregularly. As long as you understand this definition and combine them according to the store situation.

This is just a combination of molecular populations in population attributes, such as identity attributes and weather attributes, which can be combined.

Level 1 group: occupation

Level 2 people: Occupation + Whether to have a house

Level 3 people: occupation + whether there is a house + marriage stage

Wait, I won’t list these one by one, they can be very combinatorial. The purpose of our people is to label them well, accurately traffic, and remove invalid traffic from garbage, so that the overall data indicators will be improved as a whole, and when the indicators are good, it will drive natural traffic.

4. Cluster the most accurate groups and increase the premium ratio

For example, your store is a nail art store with an average customer price of 89, aged 25-29, and is more popular among white-collar workers in Guangdong. According to your own understanding, the group of people with a relatively ideal conversion rate should be:

Female + 25-29 years old + 50-100 unit price + 1750 yuan and above + Guangdong + white collar

This is the same as the sixth-level crowd. The result of easy operation is that the label is perfect, but there is no traffic!

It is easier for everyone to understand when making pictures. The more molecular combinations of people at the hierarchical level, the less traffic, and the higher the conversion rate. What do we need to do? It is to increase the number of people covered by the fourth category of people, increase the conversion rate of people in the second category, and achieve the precise labels we often call.

In operation, we can use a premium to lower the premium for people with low conversion rates to reduce the entry of garbage traffic. Similarly, the premium ratio of people with high conversion rates is increased, and the most accurate people are gradually identified, and then the combination is gradually split and combined according to data feedback. In this way, the labels are gradually made accurate to the account, and the system will naturally give search traffic. Now, thousands of people are refined in different ways. The accuracy of the label is also an important basis for considering maintaining the homepage traffic. If you want to explode, the details must be in place.

When you increase the premium ratio for people with high conversion rates, the ranking will be high, and the traffic base will increase. Then the question is why the conversion ratio for people with low conversion rates will also be high. What is the reason?

I don’t know if you have ever encountered it when you were building a store. Some products in the store have low traffic and high conversion rates, but when the traffic explodes, the conversion rates are low?

Yes, the same is true for the express train. The current data indicators are pretty good, but with increasing the delivery, the data cannot be maintained, and the conversion rate is not as good as before.

This is because, whether it is bid or premium, after increasing the investment, the ranking position will be relatively high. The traffic displayed by the baby will increase, and some inaccurate traffic will definitely be mixed in the middle, and just want to poison your account health like a virus.

This is not difficult to understand. Just like the user's habits, customers with high-quality demands have a higher chance of placing orders for your product on the previous pages, because the demand points are enough, so there is no need to slide down. At this time, if your account meets the top ranking and the crowd label meets the users, the conversion rate will naturally not be too low.

Similarly, the conversion group of people who have been searching for several pages before placing an order is more accurate, because your premium ratio is relatively low and the ranking is low, and the population meets user needs, so that the conversion rate can be well controlled. Because the base is relatively large, the smaller the crowd level, the larger the coverage area, and the greater the garbage flow. It mainly depends on the feedback of the data, which is actually the conversion rate indicator.

5. Create a perfect label

There are too many groups of people in the express train. We can only drive up to 100 people. We definitely don’t need to drive them all. So how can we quickly find the most accurate people?

To put it simply, the initial population is a relatively simple population, and basically the second-level population starts to operate. At this time, if the conversion rate is good, there is no need to evolve into a third-level or fourth-level population. After the level-up, the number of people covered will be smaller.

Of course, if the conversion rate does not meet expectations, you can gradually increase the number of people, but do not add and delete frequently, which is not conducive to data reference. After a certain amount of data is available, the optimization and deletion of people with a large premium but the conversion is not ideal. When deleting people with a display without conversion, the gradual splitting and evolution will give high premiums to the precise group and create a label. The label is gradually accurate, but I am still afraid that I can’t sell it?

In addition to these levels of people, people who do not need to combine must also adjust and optimize. To give a simple example, for example, high-quality people in the industry targeted population can make high premiums. This group of people has a relatively high demand for purchasing, especially those in the industry who are highly keen to buy. For those who have long-term value in the store, they can be maintained for a long time. If the store has a certain foundation, they must be given a high premium. My operation is basically at a premium of about 80%. This can control the demand level of the population. These combinations need to be gradually adjusted based on their own data feedback. There is no precise crowd as soon as you come up. Moreover, different groups of people in the segment are not static. Only by adapting to the rhythm can you survive.

Okay, that’s all for today’s sharing. Everyone is welcome to comment and exchange below!