Product operation: 5 steps to build a complete user recall system!

This article mainly introduces how to build a recall system for information products from 0 to 1.

Peter Drucker, the father of modern management, once said: The purpose of business is to create and retain customers.

"Creation", as the name suggests, is to bring users the "aha moment" and allow customers to access core functions more quickly. Then you can start to build a high-speed growth train on this basis to achieve user growth.

However, some people are not on the bus at all, or get off halfway. At this time, another way is needed, that is, user growth means to retain users, which is the lost user recall system we will introduce today. Today I will introduce how to build a recall system for information products from 0 to 1.

The recall system is divided into 5 parts, the specific contents are as follows:

5-step diagram of user growth system:

1. Find the problem
2. Propose an idea
3. Expected Results
4. test
5. Replay Analysis

These five steps form a complete user growth system. Steps 1-5 should be repeated continuously for a period of one to two weeks. We call this the user growth cycle.

The growth team needs to hold a meeting every cycle to review and analyze the tests in the previous cycle, identify problems and propose new ideas, and finally decide on the tests that should be conducted in the next cycle.

1. Discover the problem

When discovering a problem, you should think carefully about a few questions:

1. What is the current loss situation?
2. Is it necessary to recall users?
3. What is the cost of acquiring new users?
4. How much should it cost to recall users?

First of all, you need to understand the current churn situation . Before launching a recall plan, you need to know the total number of churned users, the number of new churned users per month, the time distribution of churned users, etc. The following figure shows the overall situation of lost users.

The blue bar graph represents the absolute value of new churn, and the red curve represents the new churn ratio.

From this ratio we can see that the rate of loss exceeds the rate of new additions, and the absolute value of loss increases, which is mostly in the channel expansion period. Through this chart we can understand the basic situation of the current channel.

Next, let’s discuss the next question: whether it is necessary to recall lost users.

Some companies do not recall their users. For example, some Internet finance companies have already subsidized new users a lot when they are attracting new users. Since the marketing expenses in a user's life cycle are limited, they will not spend more money on lost users. This is due to the excessively high customer acquisition costs of financial products.

Generally speaking, the average customer acquisition cost of financial products is between 100-500 yuan per person. Taking all factors into consideration, Internet finance companies will focus their operations on the pre-churn stage rather than churn.

Having said that, the customer acquisition cost of information platforms is approximately 5-30 yuan, which is relatively low, and there is more abundant operating expenses to support the recall of lost users.

Frederick Reichheld of Bain & Company once did a study that found that for every 5 percentage point increase in user retention rate, profits will increase by 25-95 percentage points.

To sum up, information platforms must have plans to recall lost users. Not only must they do it, but they must also do it periodically.

Finally, let’s analyze how much it should cost to recall users.

When introducing the Internet finance article before, I mentioned a method, which is to allocate the cost of recalling users according to the user's life cycle contribution value (recall cost = user life cycle contribution cost - market cost - operating cost).

This method is applicable to industries such as e-commerce and Internet finance, but for other industries such as information, the life cycle contribution value is not as clear as the above two industries, so it cannot be directly calculated using this formula.

For information platforms, magic numbers are used to set recall costs in the early stages of a recall.

According to industry statistics, the cost of acquiring a new user is five times the cost of maintaining an old user. In other words, the cost of maintaining an old user is 1/5 of that of a new user, which is the lower limit of the cost. According to the length of time that lost users have been lost, some users who have been lost for a long time can actually be considered as new users if you change your way of thinking, so the upper limit cost is equal to the cost of acquiring new users.

Here, the cost of acquiring new users is taken as the industry average of 15 yuan (including information flow, app store, etc.), so our recall cost range is as follows:

The cost range before the growth experiment of the recall system begins has been preliminarily determined. As for how much money is reasonable within the range and how to balance the recall rate and cost, we will continue to improve and perfect it based on later experiments and re-analysis to find the optimal recall cost.

2. Propose an idea

The idea was proposed to prepare for the fourth phase of testing, and to conduct splicing experiments based on modules such as user quality, reach tools, SMS copywriting, reward gifts, landing pages, etc. As shown below:

According to our ideas, users are classified into A1, A2, A3...AN; means of reaching are B1, B2, B3...BN; SMS copy is C1, C2, C3...CN.

In this way, a total of N x N solutions enter the reserve pool, but the company's resources are limited, and it is impossible to put all ideas into practice. How to prioritize the many ideas? This is where the "ICE scoring system" specified by Sean is used to organize the ideas of the second stage.

ICE stands for impact, confidence, and ease.

For example, as shown below:

There are currently N solutions, but our resources only allow us to conduct 3 experiments at the same time. So we use ICE scoring and finally select 3 solutions for experimental testing based on the scores from high to low.

3. Expected Results

I have seen many user growth experiments that ignore the step of expected effects. Some people think that user growth experiments are about rapid iterative experiments and reviews, but they forget to estimate the effects before the experiment. This is a wrong approach.

If the direction is wrong, efficient execution will be a disaster.

Therefore, we need to split and analyze the current status indicators before we do it. Taking the recalled users as an example, we need to know the various indicators of naturally returning users (users who return naturally after loss, not through testing methods) before the experiment, as well as split the natural return indicators of each type of users and set expected goals. As shown below:

The above figure shows the data of natural return of lost users. How to interpret this data?

That is to say, if a user has not logged in for 30 consecutive days, we will consider him/her as a lost user. When these users are not exposed to any recall experiment strategies such as SMS and PUSH, the natural return rate is as shown in the green column in the figure above.

These users are then classified according to certain behaviors they had before churn, because different behaviors before churn also determine the difficulty of natural return and experimental recall . For example, we see that the natural return of former member users is higher than other behaviors.

With the natural return data, it is natural to set goals based on this. In this way, there will be a specific target direction before the experiment.

4. Testing

The testing phase is to put the ideas from step 2 into practice. The strategies have been formulated. As in step 2, we will divide the experiments into 3 groups:

• Test group 1: A1+B1+C1+D1+E1,
• Test group 2: A1+B3+C1+D1+E1,
• Test group three: A1+B3+C1+D1+E3.

The lost users are sampled and grouped to ensure the uniformity and consistency of the three groups of samples. Next comes the specific implementation phase:

• A: Screen high-quality users (based on RFM or behavioral models, etc.);
• B: Prepare the reach tools. For example, SMS requires a mobile phone number, PUSH requires IMEI and IDFA, and activities require a user ID.
• C: Write the text of the SMS and get it confirmed by the legal department;
• D: The form and time of gift distribution, for example, whether the gift is sent when the returning user clicks the SMS to enter the APP, or is it distributed on a T+1 basis, etc.
• E: The landing page entered after clicking the SMS link or PUSH.

Everything is ready, JUST do IT!

5. Review and Analysis

Review is the last step in the user growth system and also the most critical step. It determines the direction of the next growth cycle. Therefore, it is necessary to analyze the data in a multi-dimensional and three-dimensional manner. A good review system will achieve twice the result with half the effort.

Let's talk about the review analysis. The review analysis mind map is as follows:

Next, we will follow the mind map step by step, using retrospective analysis to evaluate the pros and cons of the activities, identify problems, and develop strategies for the next growth cycle.

1. How much did it cost?

In the first stage, that is, the stage of discovering the problem, we set a recall fee of 3 yuan. This price was determined based on experience. As for whether it is really reasonable, it needs to be verified in the step of review and analysis.

The actual recall cost = reach tools (SMS, PUSH) + reward cost + subsequent operating costs. For example: the SMS fee is 0.3 yuan, the reward points cost 3 yuan, and after the user is recalled, he participates in the rebate activity on the platform, then the recall cost = 0.3 yuan + 3 yuan + 1.7 yuan = 5 yuan.

As for whether this price is reasonable, let's do some calculations: if we only consider the login cost, it costs 15 yuan for a new user to enter the platform, which will bring 12 logins within a year, which means that the average login fee is 1.25 yuan.

Recalled users will log in 3.7 times after the recall, which means it costs 1.35 yuan per login.

By this calculation, the login cost of recalling old users is greater than the new cost. In this case, the recall cost must be appropriately reduced, and the reward cost should be reduced to below 2.625 yuan, that is, the total recall cost of less than 4.625 yuan is a reasonable price.

2. How effective is the experiment?

The Gantt chart above lists the recall rate, return quality, recall cost, strategy, and implementation time of each experiment. The bubble chart below can more intuitively show the quality of the activity.

The horizontal axis represents the recall rate, the vertical axis represents the quality after recall, and the diameter of the circle represents the recall cost.

It can be seen that D, E, and F have better usage rates. By analyzing the activity matrix, we can see that the main factor affecting the recall rate is column A of the test matrix, that is, user classification. A1 represents high-quality users, so the recall rate will be better.

This also verifies our hypothesis that users with high quality before churn are easier to recall, and their quality after recall will not be too bad. On the contrary, it is observed that A, C, and G belong to the A3 category of low-quality users, and their recall rate and quality after recall are also relatively poor.

Of course, if the situation is completely opposite to the actual result, we will adjust the strategy and redefine the test matrix.

For example, the actual recall rate and quality after recall of high-quality users that we calculated are very poor, while the average-quality users perform very well in both aspects. Then we need to dig deeper to see which variables are highly correlated with the recall rate and quality after recall. In the next test cycle, we will reclassify the users according to the new model method.

Subsequent output situation (recall quality): The subsequent output situation (recall quality) has been shown in the bubble chart. It is the result of observing the subsequent performance of recalled users and quantifying indicators such as effective behavior, retention, and core behavior. Here, it is scored on a scale of 1-5. The higher the score, the better the subsequent performance.

RFM Analysis:

Under normal circumstances, the RFM model is used to classify users. Users are classified into high-, medium-, and low-quality users based on the length of time before churn (Recency), login frequency (Frequency), and effective behavior before churn (Monetary) .

The M in RFM originally stands for amount, but here I changed it to effective behavior because the financial attributes of information products are relatively low.

Through the above bubble chart, recall quality analysis can also verify the rationality of RFM model classification and continuously correct the user classification model.

3. How to improve later?

From the above-mentioned Gantt chart, bubble chart, recall quality and RFM analysis, we can dig out a lot of valuable information and problems, such as how to make the good parts better and how to improve the bad parts.

• Analyze from an operational perspective whether user classification, recall costs, choice of contact methods, choice of copywriting, choice of gifts, etc. are reasonable.
• From a product perspective , is there room for optimization in the landing page selection when users first open the APP, the waiting time from clicking the short link to opening the APP, etc. From a technical perspective, can the user growth system be automated?

So far, all the steps of a user growth cycle have been completed. Let's sort out the process of the entire user growth system today:

User growth system:

1. Find the problem
2. Propose an idea
3. Expected Results
4. Test Matrix
5. Replay Analysis
6. 1,2,3,4,5,1,2,3,4,5,1,2,3,4,5,1,2,3,4,5……

This is the end of the introduction to building a user growth system from 0 to 1. Here I want to tell you that growth is accumulated bit by bit from small successes. Every test is an opportunity to learn and improve. Continuously learn and improve, gain more success, and eventually form an overwhelming advantage.

Author: Liu Changlin

Source: Mars Operations (ID: hxyy233)