APP user growth data logic!

APP growth is inseparable from the application of data, and the depth of application determines the effectiveness of growth.

The basic application is to use data as an acceptance tool to measure and evaluate the effectiveness of each link of APP growth; the advanced application is to use data as important information, conduct a comprehensive insight into feedback data, and find the real needs of users; the more in-depth application is to use data as an important resource for APP growth, dynamically manage and mine data, establish data models to develop the value behind the data, and provide data basis and verification for the optimization of each link of APP growth.

This series of articles will disassemble a "growth" case completed in cooperation with a shared bicycle APP, interpret the role of data in the entire APP growth process and how to play its role.

Crowd Insights Scientific Verification Experience Prediction

Strategy is the cornerstone of APP growth, and crowd insight is the "digital navigator" that helps APP verify and clarify its target audience.

Before conducting crowd insights, the APP cannot act aimlessly, but must first have a preset target population. APP operators can combine industry understanding, experience summary and APP’s own data to conduct comprehensive analysis and formulate target groups. The shared bicycle APP in this case identified college students as the key target group for growth before conducting data insights.

After the APP has formulated the target population, the first thing it needs to do is to verify the development potential of the target population and whether it has "growth" value. Data insights can be verified through industry comparison from dimensions such as APP user portrait, user composition, and offline scenarios.

The case is compared and analyzed from two dimensions: the overall population portrait of the shared bicycle industry and the population portrait of the bicycle APP users.

By comparing the two groups of user portraits, the following conclusions are drawn:

• The 25-44-year-old group is the mainstream group of bike-sharing apps; the 18-24-year-old group accounts for 6%, which is a very competitive group outside the mainstream group;
• In the age structure of the bicycle APP population, people aged 18-24 account for 12.6%, which is slightly lower than the proportion of people aged 12-24 in the industry's age structure. This shows that there is still room for growth in the expansion of the bicycle APP to the 18-24 year old population.

In the user population structure analysis, the shared bicycle population was analyzed based on population tags, and the following conclusions were drawn:

• College students are the second largest user group in the bicycle industry;
• Among the users of this bicycle APP, college students account for only 7.3%, which is far lower than the industry share. The potential market space for college students is relatively large.

Offline scene insights break the spatial dimension wall

The dimension of APP’s user insights cannot be limited to online behavior, but also offline scenario factors must be considered. Because offline life scenarios will also have an impact on users' APP usage behavior.

By adding offline scenarios to insights, the APP can break the spatial dimension wall, achieve organic linkage between online and offline, and help user growth. Just as e-commerce companies are pursuing the concept of "new retail", its core lies in connecting online and offline scenarios and promoting growth by solving offline needs through online services. This growth strategy requires strong offline scenario insights.

In this case, the "growth" strategy of the shared bicycle APP also requires a close integration of online and offline. Therefore, in addition to the comparison of online populations, offline scene insights are equally important.

The case conducts a population heat map analysis of offline scenarios of shared bicycle users (sample locations: Shanghai and Chengdu), with a particular focus on areas with dense student populations such as university towns and entrepreneurial parks. The heat map shows the density of the crowd through blocks of different shades of color. The redder the color, the denser the crowd.

As shown in the above figure, whether in Shanghai or Chengdu, the locations of university towns are areas with highly concentrated regional populations (i.e. red blocks) and have expansion value.

The above data insights show that college students are indeed a valuable target population. It is feasible for bike-sharing apps to target college students as their growth target users.

However, the size of the college student population is limited, and they are not the mainstream user group in the shared bicycle industry. The APP's target growth group is only college students, which obviously does not cover enough. The APP still needs to find new growth target users to supplement.

According to previous crowd insights, office workers are the big users of bike-sharing apps, and bike-sharing solves the last-mile problem for commuters. Based on the understanding of the industry, the case shifts the direction of insight to commuters, and combines offline scenarios of transportation connections to conduct crowd insights at stations along the Shanghai and Chengdu subway lines.

This insight combines users’ online behavior preferences with offline scenarios for verification, and the results show:

• During the morning rush hour, there are more cyclists around subway stations;
• During the evening rush hour, the number of cyclists around subway stations is more active;
• Potential bicycle users are concentrated within a 1km radius of subway stations.

Insights show that the 1km radius area around the location station is a new growth scenario for shared bicycle apps, which can cover more active people.

The case ends here, and the data insights for the growth strategy formulation period come to a temporary end.

Through the above data insights, the shared bicycle APP not only verified that college students are a valuable target population, but also found that the target commuting population is relatively concentrated in the areas around subway stations and the APP activity is relatively high, which is an effective "growth" offline scenario.

Next, the APP can optimize growth strategies based on the results of data insights, such as strengthening promotion in scenarios with dense crowds of college students, such as university towns and entrepreneurial parks; and conducting bicycle dispatching operations and event operations along the subway lines during peak hours in the morning and evening.

Build a target population model to dynamically improve promotion accuracy

As the growth strategy is gradually implemented, the process of data insight will continue, and the stage of truly reflecting the compound value of data has just begun.

We believe that good data insights should run through the entire cycle of APP growth and promotion, and be able to generate data models, accumulate and process data in real time during the growth process, continuously iterate and optimize, and guide APP growth strategies to execute in the optimal direction.

In the subsequent execution and promotion of the case, how to locate these target groups and how to reach them more widely requires the APP to use the target group model in its delivery. If an APP wants to have a more accurate target population model, it needs to have more dimensional data features as a basis.

In this case, the target population model of the shared bicycle APP involves five dimensions:

1. Bicycle industry users: Bicycle APP installation and usage characteristics can be used to distinguish active users from silent users, and focus promotion on new APP users and silent users with awakening value;
2. College students: Based on the college student population label, add features that can improve differentiation, such as age group and permanent residence, to further target the student population;
3. White-collar workers: Based on the labels of white-collar workers, we add features that can improve the differentiation, such as working time period and professional scene, to further screen the white-collar workers;
4. Subway crowd: Use LBS geo-fencing technology to select subway station scenes to more accurately cover commuters;
5. High-density population: Find areas and populations with strong demand for bicycles through the combination of APP penetration rate and offline scenarios.

Among the above five dimensions, bicycle industry users and high-density populations cover, from a macro perspective, users who have not yet been reached by shared bicycle apps but have needs; college students, white-collar workers, and subway people are relatively micro, playing the role of locking in the target population. The target population model formed in this way is applied to APP growth promotion, which can not only accurately lock in the main target population, but also take into account the breadth of dissemination and influence more people.

The target population model is not static. It needs to be continuously iterated and optimized based on the data return from each delivery link so that its accuracy will continue to improve.

An effective promotion requires attention to every link in the entire conversion funnel. Each link requires collecting data, analyzing data, and carefully observing the problems reflected behind the data.

The APP needs to conduct in-depth research on the characteristics of the converted population based on the actual delivery effect, better deepen the understanding of the target population, further optimize the population targeting, and prepare for the next delivery.

Set the optimization effect attribution model for the natural traffic control group

In the process of growth promotion, effect attribution helps APP evaluate the effectiveness of the promotion combination on the one hand, and helps APP solve the "real problem" of how to spend money on the other hand.

The most important thing about performance attribution is to understand the various cross-channel interactions that led to conversions and the relative weight applied to each interaction. The more objective the data of effect attribution is, the greater its impact on the final delivery results.

However, the current mainstream attribution logic still has the drawback of being not objective enough. For example, the attribution models provided by Facebook and Applovin cannot eliminate the interference of natural growth traffic on promotion effects.

In this case, the target population is divided into a promotion sample set and a control sample set in a 9:1 ratio.

Promote to the population in the sample set and conduct statistics according to the mainstream attribution logic. The population in the control sample set will not be promoted, and their natural growth will be counted. In the final data attribution stage, the interference of natural traffic in the mainstream attribution logic is reduced by removing the natural traffic growth rate collected from the control sample set, so as to explore better promotion channels.

Of course, the natural installation rate of the control sample set is calculated through sampling, and there is an error compared to the actual natural installation rate, which requires the APP to make some optimization adjustments based on the actual situation.

Optimize APP promotion execution details through multi-data insights

During the growth of an app, there are still many details that can be solved through data insights. For example, identifying and protecting against black traffic, finding the optimal bidding range, the correlation between exposure times and promotion effects, and optimizing offline promotion scenarios.

In this case, data insights into exposure times and promotion effects helped the shared bicycle APP achieve the best promotion effect at the optimal cost.

Insights show that when the exposure times are less than 5 times, the number of exposures is positively correlated with the promotion effect, and the CPA cost can be controlled at around 6.24 yuan; when the exposure times are 5 times or more, the number of exposures has no direct relationship with the promotion effect, and the CPA price increases exponentially. On the other hand, the marginal CPA analysis shows that the promotion effect brought about by each additional exposure has not changed much.

From this we can conclude that high costs do not necessarily translate into good results.

The APP does not need too much unnecessary exposure. The solution of controlling the number of exposures to less than 5 times is the most cost-effective and can achieve the best results at the lowest cost.

In general, with the gradual implementation of data intelligence in the Internet industry, big data will become the new generation of growth "black technology".

Large companies have made arrangements in this area very early and have accumulated a certain amount of models and data. Other APP developers can also quickly implement data applications through tools and services provided by third-party data service providers.

With both hardware and software fully prepared, APP still needs a little more patience with data intelligence. After all, the iteration and optimization of data models requires slow and careful work.

Author: Operations

Source: Operations