How to predict user behavior in the application

    At the Epic Growth Conference , CEO of App in the Air, Bayram Annakov, reviewed practices that help increase the user’s retention rate in the application.


    Read the transcript below.

    The user always wants to tell us something.


    Someone on the slide below sees poor planning. Here I see a message of pedestrians to architects. An analogy can be drawn with the product. How do users really go from point “A” to point “B”.



    We design the interface the way we think the user should use it. The task of the product manager is to understand what users want, based on their path along the application and screenshots.

    Why study user messages?


    1. The user is not a product manager


    There is a proverb: "Man proposes, but God disposes." You can apply it to the work of web services: "The product manager assumes, but the user disposes." We cannot always predict exactly how people will use the application. In order to analyze the situation in time and work on errors, it is important not to “hammer” on user messages, but to study them.

    2. Funnel is not everything


    Funnel is the most common way to study messages from users. However, the problem with the funnels is that they are not enough at a certain point. There are several reasons for this: The

    funnel greatly simplifies the complex, rich user behavior.
    On the left, the slide shows how users behave on the screens of your product. On the right is how you reduce this behavior in a set of stages, assuming that the stages of user movement are sequential.



    The user can go onboarding and then switch to another application, because an SMS with the message “go back” has arrived.

    The user does not go clearly to the goal that you set for him. He is guided by complex behavior, and the funnel greatly reduces his behavior. This does not allow you to see the complexity and richness of the user's message.

    User is wasting time


    The funnel does not take into account the aspect of time. There are accurate screens where users spend more time and where less. In our application, for example, we know that if a user reads privacy policy, then most likely he will leave the application and will not return.

    At some point, you feel that the funnel no longer answers the questions that arise. Then you have to study the user trajectories.

    3. The user graph is the key


    What is a trajectory? Imagine: you have standard events (Google Analytics, Firebase, Amplitude). Events have a time sequence. You represent user behavior as a sequence of actions with transitions from one event to another.

    Nodes are events (as a rule, these are screens). Transitions are jumps between screens. When we draw a screen layout, we use about the same tool.

    It would be cool to analyze all the trajectories of all users, find patterns in behavior and what they are trying to tell us. But when the number of users exceeds 100 million per month, there is not enough time for manual analysis. I have to use an automated tool.

    4. Frequency analysis = benefit


    We have developed a set of tools for tracking the trajectory of users who buy and do not buy our product. We use the product usage frequency matrix.



    Along the edges of the slide are different cohorts of users. Two charts show the percentage of users who buy our subscription from each cohort. On the X axis - we see an indicator of the frequency of use of features, on the Y axis - users.

    When you build a similar matrix, you begin to see the fundamental differences between one cohort and another. Knowing that as a result there are differences between what proportion of users subscribe and which fraction does not, you can understand which screens, events and actions lead to the understanding of the user.

    5. Through the group graph you can see insights


    We are interested in looking at the sequence in which users use features, and constructing what we call a “group graph” - a graph that characterizes a particular group. For example, the key features they use.

    Further, depending on your application or tasks, you make people move along the path that gives you the maximum result.

    If you clearly understand that your product is suitable for different categories of users, then build onboarding. You can also sharpen the entire portion of the product for this use case.

    6. Cycles lead to an outflow of users


    When you get a tool that automatically analyzes the graphs, and built a transition graph on one of the cohorts, you begin to see losses in this graph.

    For example, we lost about 5% of users after one of the onboarding screens on which the user could connect a calendar.

    This happens due to the loop: the user walks around a set of screens, repeats the same actions, and then closes the application. Cycles are very easy to find if you build a mathematical graph - because the more cycles a user makes, the lower his retention coefficient.

    7. Dynamic counting


    We found out which cycles of the sequence of user actions collected in the trajectory make the greatest contribution to the fact that a person leaves. We started flashing these cycles.



    Using trajectories, you define patterns of user behavior. To do this, you can impose ready-made mathematical tools, for example, the search for cycles - they will quickly show which cycles lead to the fact that people leave.

    You dive into these cycles, do a cross-check on a couple of users, view the entire cycle, understand what the problem is, and flash the cycle. This instantly gives a profit in the user retention rate.

    A good example: Imagine that your user arrives at a Dubai airport and gets lost. This is one of the most incomprehensible airports in terms of navigation. At some point, he is noticed by an airport employee and points in the direction of the exit. For your service, you can dynamically change the UI to maximize retention.

    We thought: “It's cool to do this inside the company. But it’s even more fun to compensate for all of these tools and enable product managers to use them. ”

    Work with Google Analytics or any analytics tool. A set of tools will help you automatically build graphs and make predictions of a person’s departure for the latest X-events.

    How is analytics evolving in many companies?


    Let's imagine that we have two axes. One end is “I know,” the second is “I don't know.” The second axis acts on the same principle. Observation of many companies and the evolution of analytics has shown that we are all moving within this quadrant.



    What is the position of retenshing and the described tools in the quadrant?


    1. “We only know what we know”

    This is usually the main dashboard of the analytics system. We know how many downloads we have, users, how much is our income. At this level, “factology” occurs. This can not be called "analytics", just statistical information. Many companies still remain at this level.

    2. “We know that we don’t understand something”

    They know, for example, what retention coefficient or which LTV. They begin to measure this in many ways in order to predict the future.

    Why measure retention? To predict the future number of active users. Why measure LTV? To understand how much we spend and how much in the end we earn from the user. How to relate this data to each other? When we are at the “we know what we don’t know” stage, we gradually consider them and try to look into the future.

    3. “We don’t know what we know.”

    This is a place of retensoring and many machine learning approaches. We already know how to measure user trajectories. We know that users are trying to tell us something. But we do not analyze this information. Tools help us pull messages from users and get insights to improve the product or, conversely, turn it off.

    4. “We do not know what we do not know”

    When you deal with retensoring, you need to move in this direction. This stage can be described as going astral in analytics. You are constantly looking for ideas, trying to apply them in your product, check and analyze the results.

    More reports on product marketing can be found on the @epicgrowth Telegram channel .

    Also popular now: