When we do promotion (especially search engine bidding, etc.), will we encounter such a situation: the promotion account has been optimized for a period of time, and the data, quality, and account structure are relatively stable.

The leaders hope that our account will go further, but in the face of a mature account, apart from basic maintenance such as adding words, updating creative ideas, and checking rankings, what else can we do? Let's take the decoration industry as an example to introduce some simple data segmentation methods, which can help us adjust our accounts.

First we need to obtain the account data. The more detailed the better, but it must be as accurate as possible so that there will be no errors when we segment the data. Of course, if you can use the Link Tag function of Google Analytics and happen to have this monitoring installed on your website, then your data will be much more accurate than others.

Of course it is better if you are using an expensive tool like Omniture. At least in terms of data, you are already a long way ahead of your peers. However, it doesn’t matter even if you are using basic data tools such as Baidu Statistics and 53 Customer Service. Although the accuracy of the data will be different, the idea of ??analysis is the same. So assuming we only obtain data on clicks, consultations, and consumption, what should we do next?

Step one: Segment the platform

There will be different platforms for online promotion, and the traffic quality of the platform will directly affect our final benefits, but companies often regard the Internet as A single channel for evaluation leads us to have no purpose in selecting and focusing on platforms.

Even if it is the same platform, take Baidu as an example, the traffic quality brought by its PC\WAP\network alliance\mobile DSP is also very different, so the more we are used to unified delivery Strategy, at this time, it is even more necessary to make a detailed assessment.

Through the background data of simple consulting software, we can sort out the following table:

This is the data of a simple decoration website. We use consulting keywords and consulting sources 2 A perspective table has been made for each attribute, and it can be clearly seen that Baidu's consultation accounted for a very large proportion. So, is it at this point that we start adjusting keywords according to this table?

The answer is NO! Because the meaning of the platform is not revealed. Platforms with large traffic naturally attract more inquiries, but how cost-effective is it? We can't see it from here, so next we need more data support to compare different platforms.

By adding input data, we get a table like this:

The click here refers to effective consultation, so we can clearly see that 360 and Sogou have low traffic , but the price is also cheap. Baidu has large traffic but the price is very high. Especially for Baidu’s mobile version, the price is already quite high.

Then should we control the mobile data immediately? Lower the price? Check ranking? Correct keywords? If you don't want the effect of the boat capsizing at any time, you need to be more patient in digging.

Let’s look at another set of data. There are 951 valid consultations from 53 total ***, of which 668 can be counted on the platform, accounting for 72%. The remaining consultations whose sources cannot be counted , 136 of them came from the mobile terminal. Fortunately, the mobile traffic of this decoration website is only sent to Baidu mobile (natural traffic such as SEO is ignored here). Then we will correct the data. Get the following table:

As you can see, in fact, Baidu's mobile terminal is very cost-effective, with large traffic and low price. In fact, we should increase investment in this area, which is why data accuracy is important as mentioned before. He may give you a completely different judgment result.

Of course, if every platform of the website has a mobile version, don’t worry if the data is inaccurate. You can allocate the unrecognized traffic based on the proportion of PC traffic. Of course, this will have certain errors. .

The next question is, except for Baidu’s PC version, other platforms have better price/performance ratio than this one. Which one should we focus on? It's just that these data may not allow us to make an immediate decision. In this case, we still use the old method and add new parameters for comparison.

This time, the data chosen is traffic. How about adding the traffic flow of each platform to the consultation meeting:

Is this dependence very obvious? 360’s market share is greater than Sogou, but the traffic is lower and the consultation rate is higher. The consultation rates of Sogou and mobile Baidu are already very tight. So what we have to do next is obvious: increase investment in the 360 ??part and optimize the traffic of mobile Baidu and Sogou parts.

(PS: You should have noticed that we seem to have deliberately avoided the data on Baidu PC side. This area has the largest traffic. Why are there no adjustments for this area? In fact, the reason is very simple. Consumption The highest place must be the place with the highest density when doing SEM. It requires multiple data comparisons and analysis to find the optimization direction, so we will conduct a separate data analysis to determine the direction below.

)

Adjustment strategies for different platforms have been formulated, but what about the most important part of Baidu? Still the same, do another data segmentation.

Step 2: Segmentation of traffic

First, we need to obtain a series of raw data such as traffic reports, consultation visit reports, etc., and then organize them. The purpose is very simple. If possible, I hope to be able to pinpoint each keyword, how much it consumes, how many clicks it brings, how many consultations it brings, and how cost-effective it is. Of course, for most promotions, this is impossible. The data is too large and will only make our management costs too high.

What we can obtain is dead data, but the customers we face are living. It is difficult for us to track every customer and know their thoughts, so what we can do is to transparently Use these imprecise and imperfect data to analyze customer behavior. What we want to do is to make suggestions for adjusting the direction, not to demonstrate the value of each keyword.

So, let’s start with the bricks:

The data is very straightforward, from the keyword consumption table compiled by Baidu’s backend, the visit scale and consultation scale compiled by the consultation backend, Do some more piecing together. Putting aside consumption and looking at consultation, the consultation rate for the words in the renderings is 0.9%, and the consultation rate for the region + decoration company is 10.9%. This is the direct reason why we sell the company's words at a price that is more than 10 times that of the words in the renderings. Very intuitive, right? If at this time, you are busy taking these highly consulted words and adjusting the rankings one by one, then we are really doing something wrong.

Since the bidding keywords have matching patterns, you can see whether there is a way to directly merge the consumer words in the left table and the consultation words in the right table. So how to connect this fault?

As we said before, we only need a direction, not every word. Therefore, we classify and merge the keyword data. The principle and simplicity of merging are based on business lines, customer purchase stages, and search intent. You can go to the website to look at the product categories. If it is an e-commerce website, the homepage has already divided it into major categories for you. Looking back at this decoration website, we can classify it according to the customer's search intention, such as price, design, and effect. Pictures, institutions, etc.

If none of these methods satisfy you, then divide them according to the customer's purchasing stages: noticing the product, becoming interested, searching for relevant information, comparing products and prices, and consulting for purchase. Different stages must correspond to different keywords, classify the keywords, and then organize the data of different categories:

(Note: Due to space constraints, the final general category can actually be subdivided. (more)

Let’s talk about the part with the largest traffic - the general class. The general category refers to the most commonly used vocabulary when we think of this product. Taking decoration as an example, it means words such as decoration, home decoration, and decoration.

It is difficult to adjust these words because it involves multiple purchase stages, product exposure, market coverage and other issues. If the unit price is not particularly high and the budget is not tight, it is recommended to maintain the status quo. The same situation applies to brands and companies, so no further details will be given.

Putting these aside, we can see that the biggest problems are style and price, while renderings are the most cost-effective words. Although its consultation rate is only a pitiful 0.9%, the advantage is that the unit price is cheap. .

Then we can get a preliminary conclusion, expand the keyword traffic of the rendering category; limit the words of price and style, and extract the parts with high consumption and low consultation for these two types of words for optimization. ;Platform types (decoration websites, decoration forums) have low traffic and large room for expansion, and should be focused on improvement.

Does the idea of ??adjusting the account become clear at once? A few months after the adjustment, make the same data table for comparison, and then you can further optimize and adjust the account.

Summary

The idea of ??data segmentation is actually very simple: integrate existing data and use experience to compare and adjust. If the current data cannot draw conclusions, add new data dimensions and make adjustments. By constantly adding new data and subdividing dimensions, the original "whole" is divided into multiple pieces for analysis and adjustment.

Extended reading:

It is simple to have clear and logical ideas, but the key is how to get your own data through various simple tools! Just this data is enough to take a few hours to sort and analyze. If it is your first time to do such an analysis, you will have to spend a day first to sort out all kinds of data and eliminate all kinds of noise.

Previously, the display data, click-through rate, rankings, etc. were sorted out, but they were all discarded in the end. Therefore, being able to grasp key data for analysis requires a lot of practice and effort.

So, are the above efforts done? To paraphrase Uncle Yunma’s famous saying: Business in the world is difficult to do!

This is only the first step in optimization, and it only provides everyone with an idea of ????data segmentation, paving the way for us to adjust our accounts.

Before real adjustments, there are still many things we need to do:

Data combination for effective consultation:

From consultation to order completion, there is actually a very long process, as long as If customers don't pay, anything can happen, so collecting data on effective customers becomes an arduous task. What would happen if we added keyword effectiveness to the table above? What a difference.

For example, after communicating with the sales manager, I found that there are many consultations on renderings, but the effectiveness is very poor. That is, the consultation cost may be very low, but the effective consultation cost is very high. Therefore, in the end As a result, I gave up the expansion of this part of the word.

Ignored guide words:

Every time an order is placed, it consists of multiple searches and consultations. Except for some impulsive customers, most customers are in Before a transaction is completed, there is a long selection and purchase cycle, and the more expensive the item, the more so.

So during this cycle, how many words have been searched but not consulted, and how many words have assisted our transactions? This is also one of the main reasons why we gave up adjusting general categories and institutional categories before. This is also why we mentioned the Link Tag function of Google Analytics (a cross-domain tracking method) at the beginning. So, before adjusting, how many words did we have Is there any way to directly measure it using data?

The success and failure of ROI:

ROI is very important, but it is not the only important index. Experience tells us that if our investment increases, then ROI will inevitably decrease, and at the same time we To capture more market share, it seems that market share and ROI have become opposites. Just like Dangdang, a strategy of crazy pursuit of ROI, it slowly faded from people's sight. JD.com, a strategy of pursuing the market at a loss, allowed it to firmly occupy a corner of the e-commerce field.

What matters is the height at which you look at this problem, but for an enterprise, ROI is like a bird. If you hold it too tightly, it will die. If you hold it too tightly, it will die. Loose and it flies away. Maybe only time will tell the answer, but I am Fat C, not Pang Mailang.

The road to success is hidden in correct data analysis and judgment, please pay attention! is correct data analysis.

Related recommended reading (link): Why does Inke occupy the top 1 spot in the Apple market?

Some reflections on cold start of user operations

APP dry goods store official website: www.appganhuo.com