Channel promotion: How to bring better quality volume with less money?

The concept of the "second half of the Internet" was proposed by Meituan CEO Wang Xing in 2016. The general meaning is that China's Internet demographic dividend has been used up after more than 20 years, and the development mode of Internet companies has changed from "speed and traffic" in the first half to "depth and innovation."

If 2016 is taken as the dividing line between the first and second half of the Internet, then the second half has already entered its third year. Many people still haven’t figured out the rules of the first half, but we are forced into the second half.

The first half can be traced back to 1995, and it has been 21 years by 2016. If it is based on a football match, the first and second halves should be of equal duration, but is this really the case?

The only thing I can be sure of is that we don’t have much time left, and the difficulty is comparable to the appearance of the Chinese national football team.

We have all seen that ofo completed the E2-1 round of financing of 866 million yuan, and we have also seen that the number of refunds exceeded 10 million;

I have seen Pinduoduo become popular overnight, and I have also seen tens of millions of dollars stolen overnight;

People all believe that the Internet is a money-burning industry. Burning money does not necessarily guarantee success, but not burning money will definitely not lead to success.

This is how we got through the first half. Regardless of who is right or wrong, what we care about next is what we should do and what we can do to survive and not be eliminated by the times.

As a result, more people suddenly realized the importance of user growth. Yes, our topic in the second half is user growth.

In my personal opinion, anything that can be done with money is not called a thing, nor is it called growth, and anyone can do it. But how to use less money (or no money) to bring better quality volume, that is called user growth, that is knowledge, and it is also the core gameplay in the second half.

Today we will share some tips on user growth around channels. The details are as follows:

1.Channel Status

2. Model Overview

3. Model Algorithm

4. Application of the model

5. Promotion of the model

1. Current status of channels

Three major channel challenges in the “second half”:

1. Difficulty in acquiring user traffic

2. Less investment

3. Difficulty in user retention

The current channel traffic problem can be described in one sentence: small budget, seeking users, and poor quality.

What’s worse is that channel traffic cheating is rampant, with virtual machines, repeated device activation, abnormal devices and other methods of brushing traffic being common.

How can we ensure refined channel operations and achieve user growth in this difficult environment? Let’s talk about the channel evaluation model today. By evaluating the quality of channel traffic from multiple dimensions, we can spend our money wisely.

Before talking about the model, let’s briefly introduce several common settlement methods for channel delivery:

1. CPA: cost per action, paid by user activation
2. CPC: cost per click
3. CPD: (1) cost per download (2) cost per day
4. CPS: cost per sale is charged based on sales
5. CPT: cost per time
6. CPM: cost per mille is charged according to the thousand impressions

Usually, the method commonly used in the Internet finance industry is CPS, while consulting platforms such as Autohome and Bitauto mostly use CPC and CPD methods for settlement.

The Internet finance industry has relatively little water in attracting new customers (only relatively). The Internet finance business model plus the CPS settlement method requires user information such as mobile phone numbers and ID cards, as well as actual investment, which makes the cost of new cheating in the channel relatively high.

The subsequent scoring of the channel is also relatively simple, and is mostly based on indicators such as conversion, first investment amount (GMV), product purchase period, product cross-purchase, reinvestment rate, and capital reinvestment rate to comprehensively judge the quality of the channel's new user acquisition.

Most consulting platforms settle accounts based on CPC or CPD, which is relatively watery because users are not forced to register and log in. They can log in as guests, which results in the key information of many users not being captured or being captured incompletely.

Therefore, the phenomenon of machine or human brushing is beyond your imagination. Coupled with issues such as later retention, GMV, UGC, effective behavior, etc., it is necessary to evaluate channel quality with multi-dimensional factors.

For example, for a consulting app, the settlement method of channel A is CPD, which means charging by download. The following situations may occur:

(1) Users spend a lot of money but only download the app but never open it

(2) You spend a lot of money, and users download and open the app, but retention is extremely poor

(3) A lot of money is spent, users download and open the app, and retention is OK, but there is no other effective behavior

(4) After spending a lot of money, users download and open the app, and retain valid behaviors, but it is found that the virtual machine, device abnormality, or repeated activation (uninstalling the app and reinstalling it)

(5) Users download and open the app, retain effective behaviors, and the machine is also good, but it costs a lot of money

The above is just a simple list of a few situations. In fact, there are more problems in practice.

What is a good channel and what is a bad channel? It is not as simple as high retention, good user behavior, low cost or high GMV to say that the quality of a channel is good. How to comprehensively consider the quality of the channel? Today I will introduce the channel evaluation model to you.

The model is explained using the example of attracting new customers through consulting channels in difficult mode . I believe that if you understand this type of channel evaluation model, you should be able to use other types of platforms with ease.

2. Model Overview

I will show you the finished product first so that you can have a preliminary impression of the model, and then I will explain the details in detail.

Model home page display picture

Algorithm Page Diagram

Enlarged image of the algorithm page

Simply put, the model uses different variables (indicators) on the algorithm page to standardize the scoring of channels, and then summarizes a total score based on the different weights of each indicator. Each indicator can be broken down into first-level, second-level, or more-level indicators.

In the sample model:

Total channel score =

Quantity score weight 1 + behavior score weight 2 + business score weight 3 + cost score weight 4 + quality score * weight 5

Now that the general model has been introduced, let’s take a closer look at the model algorithm.

(If you are not interested in algorithms, you can skip Chapter 3 and go directly to Part 4 on model application and model promotion)

3. Model algorithm:

Criteria for inclusion in the rating database:

(1) All new channels

(2) Channels with abnormal data that cannot obtain correct information will not be included in the scoring

Scoring method:

(1) Each primary and secondary indicator is scored within a range of 0 to 100 points, with no passing score.

(2) For quantitative indicators with available data, score them in a standardized manner; if the magnitude of the data varies greatly, you can first take the logarithm and then standardize it (to reduce the impact of the magnitude difference).

(3) From the company's perspective, set positive and negative indicators. For example, the number of new employees is a positive indicator, and the cost is a negative indicator. That is, the higher the cost, the lower the score.

Take the newly added score as an example:

Among them, Qi is the number of new people in a certain channel, Q is the maximum value of the number of new people in all channels in the database, or Q can also be a preset upper limit.

Take cost as an example:

Among them, Gi is the cost of a certain channel, MAX(G) is the maximum value of the per capita cost of all channels in the library, or MAX(G) can also be a preset upper limit value.

For example, if G is set to 10, then users whose scores are higher than 10 yuan will be rated as 0.

MIN(G) is the minimum per capita cost of all channels in the database, or MIN(G) can also be a preset lower limit.

Weight establishment method Analytical hierarchy process (AHP)

Let me introduce the analytic hierarchy process (AHP). AHP can hierarchize the decision-making thinking of complex systems and combine qualitative and quantitative factors in the decision-making process.

The final result obtained through the establishment of the judgment matrix, sorting calculation and consistency test is convincing, avoiding the phenomenon that the weight prediction is inconsistent with the actual situation due to human subjectivity, and has wide application value in the field of multi-objective planning.

First, we established five primary indicators: quantity score, behavior score, business score, cost score, and quality score . The main calculation steps are as follows:

Constructing a judgment matrix

(PS: The AHP software can also be downloaded for free from the Internet)

So there are the following first-level indicator weights

Weight distribution diagram

Second & Third Level Index Screening (Clustering)

1. Blind selection: All existing candidate clustering variables obtained based on experience are included in the model, without considering whether some variables are suitable for the time being.

2. Contribution analysis: Through variance analysis, observe whether there are significant differences in classification and kick out variables that do not contribute to model clustering.

3. Similarity matrix analysis: Output the similarity matrix and analyze the correlation coefficient. If the correlation coefficient between two variables is close to 1, it means that the two variables can replace each other, and one variable is kicked out to achieve the purpose of dimensionality reduction.

1. By screening the variables through the above three steps, we can not only kick out the variables with lower contribution, but also integrate the variables with strong correlation, and finally output the variables with independent attributes.

The weights of the second- and third-level indicators are the same as those of the first-level indicators. The weights are allocated using the analytic hierarchy process, and the final indicator skeleton diagram is obtained, as shown in the figure below.

Model indicator skeleton

Model indicator skeleton diagram

Skeleton filling and BI display

Fill the data into the skeleton, and then make the corresponding BI display according to the dimensions you want to see. This article monitors channels by week and month, and can also monitor on a daily basis as the finest granularity. The following Chapter 4 explains some BI displays and applications.

IV. Application of the Model

After using the model, we can comprehensively evaluate the advantages and disadvantages of the channels. Just like a dish needs to have color, fragrance and taste, the channels also need a comprehensive evaluation of quantity, behavior, business, cost and quality.

After the model is built, the next step is application. As the saying goes, any data analysis that cannot implement the strategy is a hooligan.

The following is just an introduction to some basic BI displays and applications. In fact, there is a lot more to explore in depth, and readers are welcome to continue exploring.

The most granular channel total score display chart (I)

The trend chart of each channel in recent weeks (months), this week's score, last week's score and month-on-month value, as well as the switch button between month and week at the top, can observe the channel in different time dimensions.

Simply and clearly display the channels at the finest granularity, close or reduce the volume of inferior channels, and transfer the budget to good channels to maximize the advantages.

Paid channel overview (Part 2)

The overall trend chart of paid channels monitors the situation of all paid channels on a weekly and monthly basis to see the situation of the market.

For example, it is very difficult to acquire customers on Double Eleven, the cost is high, the user stay time is shortened, the GMV is reduced, etc. We can draw conclusions from the summary and make strategic investments for festivals such as 618, Double Eleven, and Double Twelve.

Channel total score breakdown chart (Part 3)

In addition to the total channel score, you can also break it down to look at the first-, second-, and third-level indicators to better understand the attributes and characteristics of the channel and perform refined operations based on the characteristics of different channels.

When you need to increase the volume, which channels should you invest in? Because although some channels are excellent, they will encounter bottlenecks in the amount of new volume. My colleagues in the channels often tell me that they can’t spend the money. This is the reason.

When it comes to commercial conversion, it doesn’t matter which channel you invest in. It’s good if there is enough retention volume, but the performance in commercial conversion is average.

These are all the attributes of the channels. If you understand the channels well, you will get twice the result with half the effort and can also complete your KPI at the same time. It is more than just "stable".

Comparison chart of paid and organic users (IV)

Comparing the portraits of paid and natural users, we can see that natural users’ behavior A, behavior B, retention and propagation coefficient K factor are higher than those of paid users, while paid users’ behavior C, behavior D and commercial conversion are higher than those of natural users.

By using user growth thinking, we can continue to explore whether behaviors A and B will affect user retention and propagation coefficients; and whether behaviors C and D will affect user commercial conversions.

Take the successful case of Twitter as an example. Twitter found that if new users follow 30 friends within 30 days, the retention rate of these users will be very high. The additional 30 days and following 30 friends become the magic numbers, which are your behaviors ABCD or behaviors EFG that have not yet been discovered. Finding it will open the door to user growth.

5. Promotion of the Model

This model is not limited to channel evaluation, but can also be used for activity evaluation, user quality evaluation, user points rating system, etc. This model can be applied by replacing the corresponding indicators and assigning corresponding weights.

User rating example

So far, the entire model process has been introduced. There are many more analysis methods and strategies to follow. I will not go into details here.

Finally, let's sort out the process of today's analysis:

Process fishbone diagram

(1) Model establishment

(2) Establishment of database inclusion standards

(3) Establishment of scoring criteria

(4) Weight calculation

(5) Variable index screening and debugging

(6) Results display

(7) Analyze and formulate strategies

This is the end of the standardized rating model I will introduce to you today. Commonly used rating models include the standardized model introduced in this article. In addition, regression models also have a wide range of applications. In the future, I will also share with you in detail how to use another type of model in actual business.

Postscript: I hope this article can help the majority of operators and allow users to understand how the platform operates. At the same time, I welcome colleagues and enthusiasts to communicate and learn together and put forward your valuable suggestions.

Source: (niaoge8)