The traffic discussed in this section is the traffic that can be obtained by the listings we put on the shelves. Before elaborating on this logical system, we must first correct an operational misunderstanding: the absolute “traffic first” keyword optimization orientation. For example, many third-party charging software provides single-day traffic data tables. Even though many operations do not have such programming skills, cross-border e-commerce companies with deep pockets can spend tens of thousands of dollars to purchase from those third-party information companies and have no money to bother with them. The “sellers” will also work hard to compare various phrases on Google Trends or other keyword data analysis platforms to make their own search popularity ranking list (because in most cases, shopping habits or search habits are almost the same. will change), but will this have a miraculous effect? The answer is that it has no effect.
Of course, optimizing keywords from the perspective of “traffic first” can indeed increase the product’s search exposure and traffic growth rate, but it should be noted that we are talking about search exposure and traffic. Growth rate rather than exposure and traffic. Their respective relationships are as follows:
The total traffic (Y) our listing can obtain is equal to the exposure (X) multiplied by the click-through rate (C), that is, Y=X× Because C click-through rate is related to the main product image, it can be considered as a fixed value for the time being;
Among them, the exposure (X) is equal to the total search volume of the keyword (F) multiplied by the search exposure rate [P(X )], that is, Or it can be recorded as P ( n), the exposure volume in a single day is xi(1≤i≤n), the keyword search volume in a single day is fi(1≤i≤n), and the total number of products that can be searched in a single day is ni(1≤i≤n) , so the above formula can be integrated into a continuous equation.
From the above derivation, it can be found that simply starting from the flow rate is basically “useless work”, because doing so simultaneously increases the denominator ni (1≤i≤n) and the numerator fi (1≤i≤n) , and at the same time, as the score ni (1≤i≤n) increases, the number of search result pages will also increase sharply.
Take “sexylingerie” versus “bodystocking” as an example. The search results for the two under Amazon’s “Women” are 40,000+ and 3,000 respectively, and the number of search pages are 400 and 81 respectively. According to the above simulation data, in When the home page churn rate is only about 10%, there is generally no traffic after 40 pages. What’s more, in actual web page data, the traffic churn rate is as high as 50% to 70%, so once the number of search results increases, there will be an increase in natural traffic. Growth brings destruction.
How to find keywords that match your own products? This requires obtaining a balance between traffic and the number of search page results. Explain in terms of mathematical derivation.
We have obtained the mathematical relationship between traffic (y), exposure (x) and click-through rate (C), where the single-day traffic is yi (1≤i≤n), and the single-day exposure The quantity is xi(1≤i≤n), the keyword search volume in a single day is fi(1≤i≤n), and the total number of products that can be searched in a single day is ni(1≤i≤n).
Considering that the search exposure rate [P(X)1] is inversely proportional to the total number of products displayed after the keyword search (N), that is, P(X)o1/N. However, this simple inverse relationship is too simple and imprecise, so the traffic churn rate (L) needs to be introduced here, which represents the decrease in traffic due to the increase in the number of pages.
For example, when the traffic on page 1 is 10,000 and the traffic on page 2 is 4,000, the traffic loss rate is 0.6 and the traffic retention rate is 0.4. At the same time, set the number of pages of search results to P. Its specific value will change due to changes in the total number of products (N). The relationship is P=N/48,48. The source of this value is the number of home page displays on Amazon search. The values may be different under different categories.
Therefore, the relationship between the number of search pages in a single day pi (1 ≤ i ≤ n) and the total number of products that can be searched in a single day ni (1 ≤ i ≤ n) is pi (1 ≤ i ≤n))=ni(1≤i≤n)/48.
The traffic of page k (p=k) is converted into single-day data.
At the same time, we need to slightly improve the calculation of search exposure so that it is inversely proportional to the number of products on the page where the listing is located (because the more products displayed on the page, the greater the probability that users will click on a certain product. Low), that is, when it is on page 1, the search exposure rate is:
The single-day traffic is yi (1≤i≤n), the single-day exposure is xi (1≤i≤n), and the single-day key The word search volume is fi(1≤i≤n), and the total number of products that can be searched in a single day is ni(1≤i≤n). Formulas help us understand what factors affect traffic fluctuations for a single listing.
For example, suppose it is known that the number of searches for a certain keyword in a single day is 200,000, and the total number of products searched for this keyword in a single day is 500, then ni(1≤i ≤n) At this time it is equal to 500. Assume that our product has just been put on the shelves at this time and is ranked on the last page among the 500 products. Then our product will be on page 11 (using pi(1≤i≤n)=pi(1≤i ≤n)/48 to calculate the number, while rounding up to an integer). Assuming that the traffic loss rate of each page is the same and 60%, then (0.4^11)×2000008≈4, which means that in this case our products can get up to about 8 traffic (actually estimated here The value is higher than the actual value. This is because the exposure click-through rate of the listing cannot be 100%, so it is difficult to convert 8 effective exposures or effective searches into 8 effective traffic. But when it is necessary to judge a certain category or some key points. When determining the traffic volume of word combinations, you need to use a new method to test the traffic of Amazon platform pages.
If you want to determine the single-day traffic of the keyword search page “dress”, you first need to search. Enter “dress” into the search engine.
We can see that there are many search results. As a major category of Amazon clothing, the search volume of “dress” is very large in a single day. The traffic of “dress” is Fi, then the total traffic of searching for “dress” within n days.
At the same time, because there is a drain effect on the traffic of each search page, the above mentioned factors also need to be used here. The value of traffic loss rate (L) means that the traffic will decrease due to the increase in the number of pages.
As mentioned above, when the traffic of page 1 is 10,000, the traffic of page 2 is. At 4000, the traffic loss rate is 0.6, and the traffic retention rate is 0.4. Then the total traffic of searching for “dress” and clicking into the listing within the first 2 pages is set again. The average conversion rate of these traffics is Pi, and the average conversion rate of order reviews is P. Then, the total number of reviews added to all listings on the first two pages of the “dress” search results within n days can be obtained.
This can be obtained. A conclusion: We can infer the daily search volume of “dress” based on the fluctuation in the number of reviews of all listings on the first two pages of the “dress” search results. However, there are still certain logical loopholes in this derivation logic, although we can count it. The number of reviews of all listings on the first two pages of the search results for “dress” has increased over a period of time. However, the growth of reviews does not entirely depend on the search traffic of “dress”, but is determined by all the respective keywords that can search for the products on these two pages. All search volumes owned by the combination.
Assume there are x keyword combinations, that is, keyword combinations. The traffic of each combination will bring products to the first two pages of the “dress” search results within n days. There are N reviews, and the number of reviews brought by the search for “dress” is N-d. Then the increase in reviews of all listings on the first two pages of the “dress” search results within n days.
From the above. It can be inferred that as long as we can obtain all the keyword combinations (keyword combinations) that can search for the first two pages of products that appear under the search results of the keyword “dress”, we can infer the total search volume of these combinations based on the increase in the number of reviews.
But if you really search for various keyword combinations of “dress” one by one, it is definitely impossible. Then is it impossible to judge the search volume of a certain keyword or keyword on the Amazon platform? ?Not (in Section 7.3, the ranking logic of the Amazon A9 algorithm for products will be elaborated in detail. On the Amazon platform, the search results for each keyword combination are different. The keywords with greater traffic will have different combinations. The greater the difference in product rankings. Therefore, for large traffic, keywords with slightly overlapping or non-overlapping attributes can directly determine the traffic size of different keyword combinations based on the difference in the number of reviews.
Hypothesis. If you need to compare the search volume of the two keywords “casual dress” and “dress” on the Amazon platform, you can search for these two keywords separately.
Because of the existence of traffic loss rate (L), and This value must be greater than 50%. According to the “28/20 rule” principle, only the reviews of products on the first 2 pages under “casualdress” and “dress” are counted. Whether it is through technical means@ or manual calculation means, we can know that in September 2018 On March 2nd, the review value of the product on the first two pages of “casualdress” was 33864 (the number of reviews on the home page excluding advertising products was 20126, and the number of reviews on the second page was 13738). The review value of the product on the first two pages of “dress” was 48613 (excluding advertising products) The number of reviews on the homepage is 28039, the number of reviews on the second page is 20574) and the review value of the first two pages of “casual dress” on September 3, 2018 is 33979 (excluding advertising products, the number of reviews on the homepage is 20217, and the number of reviews on the second page is 20217, and the number of reviews on the second page is 33979) The number of reviews on the first two pages of “dress” is 48813 (the number of reviews on the first page of “dress” is 28169, and the number of reviews on the second page is 20644).
From this, we can get, The reviews of the products on the first two pages under “dress” increased by 200 within 24 hours. The reviews of the products on the first two pages under “casualdress” increased by 115 within 24 hours. The ratio is 1.74:1 if some third-party products are used. Auxiliary software can obtain data: (data for a certain day in March 2018)
The number of searches using “dress” as the direct search object is 197878 times, and the number of searches using “casualdress” as the direct search object. The number of times is 111958, and its ratio of 1.76:1 is very close to the ratio of review growth.
However, it should be noted that when counting reviews, do not directly count the number of reviews on the first few pages, because Amazon will change it in real time. Permutation and combination, so when manually counting, you need to record the listing links on the first few pages after the first count, and click on the link to calculate the number of reviews during the second count. Otherwise, the calculated review value will have serious errors. When using this When predicting category traffic using this method, you also need to pay attention to the following two points:
(1) Test categories should not have too many overlapping attributes. If the two keyword combinations or keyword combinations tested have multiple attributes that overlap, for example, the words “casualsummer maxidress” and “casualparty maxidress” overlap with multiple attributes, then the ratio of the increase in the number of reviews at this time cannot reflect these combined searches. The size of the volume, because overlapping attributes cause overlap in product exposure, will cause large errors in traffic prediction.
(2) Categories with different root attributes cannot make traffic comparisons. The premise for predicting traffic after changes is that the review rate remains the same, but once the root attribute changes, this premise is no longer true. For example, the search results for shoe keyword combinations and skirt keyword combinations cannot be used as a reference for traffic comparison. Because there is a big difference between the review rate of the shoe category and the review rate of the skirt category: However, the T-shirt category and the top category can be used for traffic comparison, because the review rate of the two tends to be Consistent.
Therefore, by counting the growth ratio of the first few pages of search results for different keywords or keyword combinations, you can get the relative size of the search volume for different keywords on the Amazon platform. When there is a difference in traffic between a certain category and the current category, you can use this method to estimate the traffic of other categories to judge its market size.