How developers use data analysis to increase revenue

Programmers who do not understand data analysis are not good product managers.

I have previously written an article on traffic acquisition, which mentions that data analysis plays a crucial role in improving revenue, the first job I had the honor to do after graduation was related to data mining and analysis. I did website statistics and analysis. The company covered hundreds of thousands of small and medium-sized webmasters on a statistical platform, every day, various indicators and data are floating in the mind-PV, UV, IP, new users, active users, user attributes, page Jump rate, page arrival rate, access depth, and access duration ...... then I learned to analyze how to increase the user stickiness of a page on the SEO factor level. While maintaining the independence of variables, I analyzed where the advertisement should bring the best click rate; how to reasonably typeset and reduce user loss without changing the existing content. In subsequent work, we gradually discovered the role of data analysis in product operation and decision-making. This is not just a task that enterprise BI personnel need to do, in the early stages of product design, promotion, O & M are everywhere, so there is a saying at the beginning.

 

At the mobile Internet level, it seems that data mining is not so easy. udid replaces pc-side cookies and becomes the only way to identify users. With the increasingly improved anti-virus and protection systems, the behavior and attributes of mobile device users seem to be more difficult to understand. It is difficult to analyze the inaccurate data sources. As a developer who relies on traffic monetization, I have accumulated some ways to use data analysis to increase revenue from my previous subtle work experience. I hope to discuss it with colleagues.

To do their best, you must first sharpen your tools.

Without proper tools, we can't analyze it with our eyes alone. Fortunately, we are in a free and open-source Internet environment, and naturally there will be many excellent auxiliary tools, on the web side, we have Baidu statistics, cnzz, and on the Mobile End, We have Flurry, umeng, and TalkingData. If you are interested in shoes, you can also study Google Analytics. According to a simple investigation, most developers in China use umeng statistics, so here I will focus on how I use it.

The features and interfaces after the revision of umeng are similar to those of Flurry. Although I prefer Flurry's Dashboard (which may be a favorite of Ios systems ), however, umeng is quite satisfied with its interface style and functions. After all, it is a tool and practical!

 

Learn multi-dimensional analysis

I chatted with many colleagues about the usage of umeng and concluded that many people look at the metrics such as new users, active users, and reserved users on a daily basis, if you are a bit eager to compare the data during the same period, you can see whether the data is increased or decreased. Then, you can compare the data and income on the ad platform. It is considered normal if the data is not big enough, if there is a large discrepancy, you will be asked for questions. In fact, during my use, umeng's many functional modules are still very practical, for example, in the [application trend], you may find that your startup status fluctuates by hour, for example, this application related to the yellow calendar operation clearly shows that the maximum number of startup times is reached at every day, because it provides the function of pushing every day, in this case, you can add two ads appropriately to get twice the result with half the effort.

 

 

In addition, the "funnel model" in statistical analysis is often used out, so that you can easily see the depth of the user's screening at each step. After the data is visualized, you can analyze the impact of a specific step and how to increase the conversion rate of a specific step. For example, I added a point wall to the game's main interface, for the first time, a user needs to obtain certain points before entering the level. This application has a large number of new users, but the unit user income is low. In order to study the effect of the model set by the point wall on the game itself, the page access path in [function usage] shows that when a user opens an application, 5.5% opens the integral wall, 30.6% leaves the application, and 63.9% tries to select a game level, when we repeat steps 2 and 3, the percentage of users who actually access the points wall to obtain points is only 1.8%.

 

I found the reason from Figure 2: The first time a user enters the credit wall, the traffic loss rate is as high as 44%! The traffic loss rate during step 2 and Step 3 is also 17%, which means that it is not a wise choice to limit the number of users when they enter the game, it is not feasible for me to increase the revenue of a unit user through the attractiveness of the game itself, so I am willing to give the user the first level for free, the following levels require certain credits to activate, which not only increases the retention rate of users, but also increases the income of some unit users.

 

 

The number of wall opening users has increased to a certain extent:

 

At the same time, umeng also provides a more practical tool-[event conversion rate], which aims to analyze the path through user-defined events. In fact, such features are already popular in web statistics, the only drawback is that for webmasters, there are too many links, buttons, and content on a page, so when the website traffic is not that large, the funnel analysis that defines an event will produce a certain amount of data distortion, leading to analysis deviations. The advantage of mobile end is that it just avoids this point-generally, a single page of mobile applications, it does not give users multiple choices, so that the analysis of event conversion rate is more effective when traffic is concentrated.

 

For example, the above three-step event model is ad pop-up window, AD (Application) details, and application download. In fact, we are most concerned with the last step-ad application download, you can analyze several problems by combining the two pictures above and below:

1. Whether the conversion rate of step 2 and Step 3 has reached the expectation

2. How to quickly and effectively increase the conversion rate of step 3

3. Is the process design reasonable?

First, there is no major problem with the ad download process from the pop-up window-display-user click-download. Obviously, we have a high expectation for the ad download rate, because it determines the final revenue, and the pop-up window of 1.6%-download is not very high, it can be learned that there is still much room for improvement from the pop-up window of advertisement to the display of advertisement details, if we try to increase the retention rate of the user's displayed ads, will it help the download rate?

Here I made another version: When an advertisement pops up, users are not allowed to delete it, but must click to enter the details page to delete it. The following results can be seen:

 

In this way, you will understand why many domestic advertising platforms intentionally design the buttons for eliminating advertisements in a relatively small way, or intentionally design the advertisements with an extremely attractive design, which may lead to user errors, this is the simplest and most crude way to increase the conversion rate. Although it forces and induces users to some extent, it is also favored by advertising platforms and developers.

View data rationally

Nowadays, big data and cloud are everywhere. I have the honor to attend several big data forums and summits. I feel that there are many advertisements and many servers to sell, there are few real analyses. As a developer, the key data on advertising revenue is unclear. How can we measure the merits of an advertisement and the quality of an advertisement platform, it is easy to be fooled by the so-called "big companies", "big platforms", and "Big Data. In fact, you don't have to stare at the data every day to study the number of new users and the number of active users. You only need to take the "conversion rate" and "unit user income" as the core for overall consideration, then, you can evaluate them separately in different advertising forms.

Conversion Rate: average AD downloads/clicks for each new user

User income per unit: average revenue of each user

Then, based on the Banner's fill rate, click rate, opening volume of the integral wall, promotion volume, display volume, click rate, promotion volume, and other indicators, as long as the data fluctuation is within a reasonable range, you don't have to worry too much about it. For example, if you see that you are counting more active users today, but the ad revenue is less, you need to look at the number of new users and the number of ads displayed, are most active users not opening the ad list?

In addition, umeng's new statistics and active users are relatively loose, and their latest sdks provide developers with three data transmission policies.

1. Send at startup

2. Send by Time Interval

3. Send when exiting

As we all know, mobile terminal user information is located through Udid. I have studied this. The statistics of old users of umeng have a certain life cycle. That is, after a period of time, this user will access it again, it becomes a new user, and the old users of the advertising platform are basically permanent. I believe this does not have to be explained too much. Therefore, the statistical data of umeng is basically greater than that of most of the current advertising platforms, if you do not understand it, you will think that the amount is deducted.

In fact, it is very easy to verify whether an advertising platform deducts traffic. You just need to check the data:

1. unit user income

2. Real-time Data-that is, how long it takes to update

3. Data stability of time-based statistics

 

View more analysis cases

Although mobile ad revenue analysis on the market is relatively small, you can learn a lot from other types of analysis reports. My personal experience is, iResearch, DCCI, and yiguan can read a macro data, and they will not give you very detailed data. They will visit 36 kill and 199IT more, some analysis methods can be learned from some reports translated from abroad, such as RFM model analysis and social network analysis. You can go to some well-known data mining blogs, recommended blog:

Http://idmer.blog.sohu.com/

Http://shenhaolaoshi.blog.sohu.com/

My blog:

Http://www.cnblogs.com/LilJim/