User operation examples, creating a lost user recall system

The author will share with you his exploration and experience in building a "Lost User Recall System", and strive to help readers have a complete review and understanding of the work of "Lost User Recall".

Before writing the article, let me do some calculations for everyone.

In the app market or other distribution channels, according to conservative estimates:

• The average download and activation cost of a tool product is 1 to 5 yuan per person;
• The average download and activation cost of an online education product is between 10 and 50 yuan per user.
• The average download and activation cost of an e-commerce product is 50 to 200 yuan per user;
• The average cost of downloading and activating a financial product is between 100 and 500 yuan per person.

If we acquire 1,000 customers through channels every day and have a 40% retention rate on the next day, 60% of users will never open our app the next day and eventually become lost users, with an average daily loss ranging from several hundred to tens of thousands of yuan.

Although the loss seems small compared to the financing of millions to tens of millions, it is regrettable that the next-day retention rate of most products is far less than 40%. This means more users will be lost! And for many transactional products, guiding users to download, activate and register is just the beginning. They also need to be guided to convert into paying users.

Since the paid conversion rate of most products is generally below 10%, the actual daily losses will be more than you see.

If any company only spends money on promotion to increase users but is indifferent to lost users, it is not only a huge waste of resources but also a form of chronic suicide. Therefore, it is very important to establish a "complete lost user recall system".

Today, I will share with you my exploration and experience in building a "Lost User Recall System". I would like to point out that before writing this article, I noticed that there are numerous articles on the market about recalling lost users, but after reading them all, I think these articles are too fragmented and lack completeness.

So my article today is slightly different: it not only builds a complete system, but also teaches the detailed execution, in an effort to help readers have a complete review and understanding of the work of "recalling lost users".

So let’s get started.

1. Key points of the lost user recall system

I personally think that there are several key points that are very important in building a "lost user recall system".

The first key point: focus on text messages

I hope everyone must understand that even though there are a lot of spam messages bombarding users every day, it is still our most important means to recall lost users.

Because it is low-cost (3 cents per message on average), there is no need to worry about users uninstalling the APP or blocking product pushes, and messages can be sent as long as you have the user's mobile phone number. The displayed text can go directly to the main screen of the mobile phone. It is clear and concise and can jump to the link. After connecting to the SMS platform, large-scale group messaging can be carried out, and the reach rate and display effect have incomparable advantages over PUSH and email.

So our content today is all about how to send group text messages better.

The second key point: high frequency recall

When I communicate with other operations teams, I often find that most operations teams are very casual and have no plans when it comes to the work of "recalling lost users." For example: if you suddenly think of recalling users, you can arrange for operations to send a wave of text messages; or when a certain festival or event comes, you can send a wave of text messages.

This kind of casual, long-interval, and unsystematic recall not only lacks complete coverage of the user churn cycle, but also lacks polishing of details such as user classification, copywriting, and acceptance, so the recall effect can be imagined.

If you want to achieve a more ideal recall effect, the operation team must have the courage to recall users frequently, improve the system, polish the copy, monitor data, and optimize costs during the continuous high-frequency recall . Only in this way can the recall effect get better and better.

The third key point: from the simple to the complex

This is the most important point I think.

“From the simple to the complex” is not only our approach to recalling lost users, but also the method to carry out this task.

Since recall work involves a series of trivial tasks such as periodic screening, user classification, AB testing, copywriting polishing, process acceptance, data monitoring, etc., how to maintain steady progress among these trivial tasks and adopt a strategy from shallow to deep becomes very important.

In this article, I will continue to remind everyone to pay attention to "starting from the shallower to the deeper."

Okay, let’s move on to the practical stage.

2. Practical stage

After you apply with the company, find a reliable SMS supplier, and finally connect the SMS system to the product.

The first task of “recalling lost users” is to determine the user’s churn cycle. That is, if a user has not opened our product for a long time, he will be considered lost during this period.

There are many general churn cycles, such as:

• Second-day churn: If a user downloads and activates our product and opens it on the first day but does not open it on the second day, it is counted as second-day churn.
• 7- day churn: When a user downloads and activates our product, opens it on the first day, and then does not open it for 7 consecutive days, it is recorded as 7-day churn.

This is determined based on the user's registration date, and can also be determined by natural month or week.

for example:

• Natural weekly churn: If a user registered for our product last week, but did not open our product this week, it would be recorded as natural weekly churn.
• Natural month churn: If a new user registered for our product last month but did not open our product throughout this month, it will be recorded as natural month churn.

Whether it is the next-day loss, 7-day loss, 14-day loss, or natural monthly loss, determining these is just to facilitate our construction of the "user loss recall system".

I think a recall system does not necessarily have to fully cover all user churn cycles, but it should at least cover the short, medium and long stages after new users come to the product.

For example: My churn recall system includes: natural month churn recall (long), natural week churn recall (medium), and next day churn recall (short).

I will recall users who have lost in the natural month at the beginning of each month. For example, in April, I will recall users who lost in the natural month of February.

I pulled a total of XXXXXX users who "downloaded, activated and opened the product for the first time in February, but did not open the product for the entire month of March." These users are the natural month churn users because they churned for a whole natural month. I pulled all their phone numbers and did a text message recall.

Every Monday, I will recall users who have churned during the natural week. For example: On Monday of this week, I will pull all the users who downloaded and activated the app last week, but did not open the app in the whole week. They are the users who lost users in the natural week because they lost users for a whole natural week. I pulled all their phone numbers and did a text message recall.

I will also recall users who churned the next day every day. For example: Today, I pulled all the "users who downloaded, activated and opened the APP the day before yesterday, but did not open the product yesterday", a total of XXXX. They are the next-day churn users because they churn the day after downloading and activating the product. I pulled all their phone numbers and did a text message recall.

This kind of work is repeated every day, every Monday, and at the beginning of every month. You may ask me, is it necessary to harass and recall users so frequently?

It is very necessary. Because user loss is often silent, if you cannot quickly recall lost users in the first place, it will be very difficult to recall them later.

You may ask: If that’s the case, why not increase the frequency? For example: recall again after three days of churn, recall again after seven days of churn, and recall again after two weeks of churn.

No, because this will lead to diminishing marginal benefits. If the time is too long, the recall effect will not be good. If the time is too short, the effect will also be poor, wasting text messages and money. Only through continuous small tests can you develop the most cost-effective recall system. But in any case, a high-quality lost user recall system must at least cover the short, medium and long stages.

Of course, the frequency of the recall system also needs to take into account the characteristics of its own products and users. For example, for information products, users use them multiple times a week at a high frequency. Correspondingly, the user recall should be compact and intensive. You can set up next-day churn recall, 7-day churn recall, natural two-week churn recall, natural monthly churn recall, and so on.

But for some products, such as e-commerce, investment and financial management, recruitment, etc., there is no need to keep calling users at an ultra-high frequency. However, weekly recalls, biweekly recalls, natural monthly churn recalls, and quarterly recalls are still necessary.

In short, your lapsed user recall should be systematic rather than random.

Classify lost users

After building the initial recall system, the next task is equally important: classifying lost users.

We all know that the core of user operation is to carry out targeted operations on different types of users. So of course, we also need to classify lost users.

So how do we classify it? There are a few simple ways:

1. Classify by core operating behavior

For example: E-commerce website A discovered in March 2019 that among the 500,000 new users in January, 300,000 users did not log in once in February and were in a state of loss, so it decided to recall them.

The next step is to classify lost users.

E-commerce website A decided to divide users into two categories: one is new users who have made at least one purchase after registration; the other is new users who have registered but have not made any purchases.

For these two categories, e-commerce company A has formulated different plans. For the first category of users, it gives coupons of lower amounts to promote repeat purchases; for the second category of users, it offers large discounts to promote first-time purchases.

Different copywriting was written for the two types of users for recall.

2. Divide by user operation attributes

For example: Product B is an online novel APP. They found that a total of 80,000 users registered in the first week of February, of which 60,000 users did not open the app in the second week of February. Considering that novel apps are high-frequency products, if a user does not open an app for a week, it is defined as loss.

Product B classifies lost users according to the attributes of the content that each user browses the most. For example, some lost users read a series of martial arts novels before they lost, so they are classified into the martial arts category; other lost users read a series of white-collar novels before they lost, so they are classified into the city category...

1. Specific copywriting and content recommendations for each type of user

The above are two small examples I made up: First, I hope to help everyone understand how to classify lost users; second, I hope everyone must understand that effective classification means more targeted actions in the future. Only when you perform targeted operations on users can you achieve the best recall effect.

However, I would like to emphasize one point here - if your product data warehouse is not perfect and the developers cannot pull out the behavioral data and attributes of lost users for you, please do not be discouraged. I suggest that when you are initially recalling lost users, you can classify them in a rough way, or even without any classification. Just run it first, and then gradually refine it after you get stable data.

Please remember the most important core point of our recall system: "from shallow to deep" . Regardless of whether we can achieve a more detailed classification, we can still continue to push this matter forward. Don't get too hung up on it.

SMS copywriting

Okay, now that we have talked about user classification, let’s move on to the next step of the lost user recall system: polishing the SMS copy.

Since the final execution details of the recall are sent via text messages, it is very important to polish a high-quality text message copy.

My phone receives many recall text messages every day. Let me read them to you:

• 【XX Game】Happy holidays to our respected users! This is a European and American Q-version strategy game. Let’s jointly defend the glory of the kingdom!
• 【XX Singing】Come and create a friend room! Two people can start the game! Sing whenever you want with your friends! Compare hand speed!
• 【X Insurance APP】Medical insurance users have priority experience, critical illness insurance, starting from 35 yuan per year, with a maximum compensation of 500,000!

To be honest, these marketing text messages cannot arouse any interest among users because they are too awkward, too self-conversational, and too hard to resonate with users.

After a series of copywriting tests and explorations, I think the following type of SMS copywriting is more effective and can be used in a variety of products.

This type of SMS copy is: scene resonance + profit seeking.

An excellent SMS copy should meet the above two points.

Profit-seeking is easy to understand, which means that your SMS copy should clearly state the tangible benefits provided to users. So what does “scene resonance” mean?

That is, your copy should be able to evoke resonance in his specific life scenes or usage scenarios.

For example, I think this text is very good: [XX Financial Management] Just move your fingers and you’ll have money for breakfast every day. 50,000 yuan of trial money has been deposited into your account. Go and claim your earnings!

"You have money for breakfast" is very relevant to the user's life scenario, and "50,000 yuan of trial funds have been deposited into the account" is a profit-seeking point. This is a text message copy that resonates with the scene and is profit-oriented, and the logic is very smooth.

For example, I also like this text message copy: [XX e-commerce] Lisa next door has a new bag, why are you still wearing old clothes? The 100 yuan clothing exchange coupon has been deposited into your account!

It also conforms to the scene resonance + profit-seeking. Therefore, I believe that if operators can refer to this type of copywriting and polish it when doing SMS recall, I believe better results will be achieved.

Process and business succession

It is not enough to just have copywriting, it is also very important whether the implementation is smooth.

This undertaking consists of two parts:

1. One is the process of taking over
2. One is business acceptance

The process includes the user receiving a text message - clicking a link - opening the landing page - waking up (or downloading) the APP - entering the target page - subsequent experience.

It is very important whether the entire link is smooth enough, and there are many optimization points to be explored. For example: when a user clicks on a text message link, if you add the word "poke" or "point" or "click me" before the link.

Even a small change can help increase your link click-through rate by at least 1 percentage point.

For example: If your SMS link can be shortened to the shortest possible length, it will help with the neatness and click-through rate of the entire SMS.

Since most SMS texts cannot directly wake up the APP through a link, but will instead enter an intermediate page (landing page), optimization in this regard must also be done well. For example: After opening the link, the landing page you enter should be opened as quickly as possible and should not take too long to load.

For example: The landing page should have a button that clearly guides users to open the APP. At the same time, the button can be used for judgment: if the lost user has not deleted your APP, the landing page button will directly wake up the APP to open; if the lost user has deleted your APP, the APP will directly call up the browser to download.

Another type of undertaking is business undertaking, which is also very important. For example, for e-commerce products, the recall of lost users can be greatly enhanced if coupons are issued or specific promotional venues are prepared. Once users are recalled, they will quickly convert into paid purchases.

For example, for news and information products, the order of content recommendations will be adjusted when recalling different types of users. Let lost users see targeted content when they come back and open the app, thereby better improving retention rate.

When taking over these businesses, operators can think about them based on their own product forms.

It is important to emphasize here that not all lost users will return by clicking on the link in the text message. When recalling lost users through SMS, an important point is that a large number of users will directly open your product after seeing your SMS.

So the path becomes as follows: user receives SMS - opens SMS - closes SMS - opens the product. Therefore, we must also be able to take over the users along this path in terms of business.

So how do you monitor the number of this group of users? We will talk about it later.

Continuous A/B testing

Because the operation of "recalling lost users" involves thousands or even hundreds of thousands of users each time. Multiple links, such as user classification, sending cycle, sending time, copywriting, jump, etc. will affect the recall effect, so it is very important to do a large number of AB tests.

For example: At the beginning of December 2018, according to the plan, I will recall all new downloads and activations in October, but users who did not log in to the product in November. Since the number of this group of users is very large, if we rashly conduct SMS recall, the recall effect will be highly uncertain, so we first conducted multiple rounds of testing.

We randomly selected 5,000 users from the entire lost user base, divided them into five groups of 1,000 people each, and conducted five rounds of testing.

In the first round of testing, we divided the first group of 1,000 people into five teams, with 200 people in each team. We arranged 5 teams at different times of the day and sent text messages with the same content. Finally, we found that the link opening rate and wake-up rate during the 9 pm time period were much higher than those of other teams. Finally, we decided that 9 pm was the best time to send recall messages.

In the second round of testing, we divided the 1,000 people into four teams, with 250 people in each team. We sent text messages with different styles and focuses to each team. As I said before, the awakening rate and click rate of the "scenario + profit-seeking" copywriting far exceed those of other teams.

After a few rounds of testing, we finalized the subdivision dimensions of the categories.

For example, we classified users who performed different operations and sent different text messages to groups, and found that the wake-up rates were not much different. Therefore, it can be imagined that the difference in operating behavior is not an important point of recall (this conclusion only applies to our products). So we simply divided user behavior into two categories without doing too much segmentation.

When all the tests were completed, we developed a targeted recall plan based on the test results, and then sent large-scale text message batches to lost users. Therefore, a large number of small-scale tests are very important for the work of recalling lost users. The cost of testing is actually very low. Testing thousands of people will cost less than 200 yuan at most, which is still very worthwhile.

Testing not only improves effectiveness, but also saves the cost of recalls.

For example: Among the above content, some friends will definitely ask us, since recall is so important, why not do more frequent recalls?

Of course we have thought about this question. After the frequency of recalling our lost users covered the next day, natural week, natural month, natural bimonthly, and natural season, one day we suddenly had an idea: should we do a 3-day recall?

We quickly divided the 1,000 users who churned within three days into two groups, one group that received a recall SMS and the other group that did not. The test results showed that the return rate of the [text messaging group] was only 2 percentage points higher than that of the [non-text messaging group]. This means spending hundreds of yuan to bring back dozens of users, which is too costly.

We conducted several more tests and finally concluded that the 3-day recall cycle was a thankless task.

Having said this, I just want to tell you: since user churn recovery is a long process, the corresponding various tests cannot be interrupted.

• Because you always need to change the text of SMS messages, how can you ensure that the replaced text is effective? Let’s test it first!
• You always need to explore some new user dimensions for recall, right? Let’s test it first!
• You always need to continuously optimize the costs and benefits of issuing coupons to find a better price-performance ratio. Let’s test it first!

Therefore, continuously conducting AB testing on a small scale is also a very important task in the systematic recall of lost users.

Data monitoring and cost monitoring

After the user churn recall text messages are sent, the issue of data monitoring and cost monitoring arises. I suggest that if you are just starting to do this kind of work, based on the principle of "starting from the shallow to the deep", the initial assessment of user return can be more extensive.

For example: When we are trying to recall lost users, as long as we can determine that after sending the text message, the user can reopen the APP within 24 hours, then we will consider the recall successful.

Is this result good when opening the APP? Of course it is calculated. Think about it, if the user has not opened the product for 7 days or 30 days, and you send a text message and the user comes back to open the product, then it definitely means that your SMS recall is successful!

Let me remind you again here: please always remember that not all users will open your product by clicking on the SMS link. Many users will close the SMS and open your product directly. Therefore, it is important to monitor all users after sending to avoid missing data.

Well, in the early stage, we successfully recalled lost users and opened the APP through text messages. Next, we need to have higher requirements for ourselves.

We can set some more advanced evaluation indicators for recall. For example, we hope that the recalled users will not only open the APP, but also complete some core operations after returning.

• For food delivery apps, if the recalled users not only open the app but also order food delivery, this is a more successful recall;
• For the accounting app, the recalled users not only opened the app but also kept accounts, which is a more successful recall;
• For financial management apps, recalled users not only open the app, but also invest and manage their finances, which is a more successful recall.

If we can achieve this level of recall of lost users, the value will undoubtedly be extremely high. But please remember to start from the basics and gradually move on to the difficult. Don't be impatient at the beginning and take your time. First, get returning users to open the app, then get them to perform more core actions.

No matter how many text messages you send, it costs a lot. As long as the contribution value of your recalled users is greater than your cost, you should continue to send recall text messages.

• For example: Although I spent 10,000 yuan to send text messages, I recalled 50,000 lost users and asked them to open the APP. The cost of each opening was 20 cents. Is it worth it? It’s totally worth it. If these 50,000 users didn’t open it, they might have been lost forever and my loss would have been even greater.
• For example: A financial product spent 10,000 yuan and recalled 3,000 users, but in the end only 100 users invested. The average cost of recalling one investing user was 100 yuan per user. Is it worth it? It’s totally worth it. An investment user in the market needs to spend at least 1,000 yuan.

Therefore, as long as the cost of SMS recall is lower than the cost of attracting new users of the same size, the recall of lost users should be carried out resolutely. Otherwise, the cost of acquiring new users in the market will be higher.

Heading towards real user operations

In addition to recalling real users, I believe that the work of recalling lost users will also drive the product and operation teams to move towards real user operations. It is like an alarm, a wound that is constantly being bandaged, constantly pushing each module to optimize its work.

• For example: Intuitively speaking, the operation team can recall this work through text messages, drive the product to make more optimizations in user registration, and increase the user's mobile phone registration rate. Because the more mobile phone numbers registered, the larger the sample size recalled.
• For example: the operation team can drive data analysts to record more user operation behavior data and build a complete data warehouse, so as to divide it into more precise dimensions during recall.
• For example: the operations team can drive technology to establish an "automatic sending mechanism" to recall lost users. For example, a certain Internet company has automated the entire lost user recall process. If a user does not log in for a certain number of days, he or she will automatically enter the "lost user database", which will then send text messages to the user within a specified period of time. If the recall is successful, the user will be removed from the warehouse. If the recall is unsuccessful, the system will continue to send recall content to the lost users according to the rules. For such an automatic recall system, all the operations staff need to do is to regularly change the copy and monitor the data, without having to do any more manual operations.

Therefore, do not simply regard the recall of lost users as an independent task. When the entire system is gradually built up, you should use this system to drive more business departments to optimize products.

In addition, I personally believe that building a user recall system is very helpful for primary operations to establish basic user operation thinking.

First of all, this job is very easy to get started. You just need to learn how to operate the SMS platform and be able to write copy. You don’t even need to learn skills such as event planning and PPT writing. As long as you are careful enough, you can complete the work independently.

At the beginning, you only need to distinguish simple user dimensions, and you can see the most intuitive effect even without distinguishing them.

In the process of building the system, this work will constantly strengthen your concepts of user testing, cost, and data, and help you develop very good work habits. At the same time, in the process of doing this job, you will deal with many business departments. For example: pulling data, running phone numbers, communicating with the SMS platform, communicating with the marketing department, etc., will also exercise your communication skills.

The entire work of recalling lost users is a long-term one. During the continuous recall process, you can optimize your operational ideas and continue to grow.

Summarize

Well, having said this, we have come to the end of this article. Let me give you a complete summary.

First of all, any company must consider the recall of lost users as an important matter and implement it firmly in the long term. The cost of a new user is very high, and bringing back lost users will save the company a lot of costs.

Secondly, for the user churn recall system, we must grasp the following core points:

• Emphasize SMS because it is low-cost, highly visible, and can be reached even when uninstalled;
• For high-frequency recalls, don’t do it piecemeal, recalling only when you think of it, but establish a complete recall system that covers the short, medium and long term;
• Start from the basics and work your way up to the complex. Don’t be too greedy in the early stages. Test step by step, achieve the basic goals first, and then continue to expand.

During the implementation process, the recall cycle must be established first. The recall cycle should refer to your own products and user patterns. The recall period for high-frequency products should be frequent, while the recall period for low-frequency products should be loose.

After determining the recall cycle, the next step is to classify lost users and write corresponding copy and recall plans for different categories. The initial classification can be simple, but as the work progresses, the classification can become more complex and detailed. It can be classified based on user operation behavior or operation attributes, etc., but it must be based on its own product, form and user form.

After user classification, the copywriting needs to be polished. The effect will be more gratifying if the copy is more "scene resonance + profit-seeking". At the same time, after the copy is successfully polished, the corresponding succession should be done well. The succession is divided into process succession and business succession. The process should be continuously optimized, and attention should be paid to details at every step to create high conversion; business succession should pay attention to cooperation with other departments to give returning users a better experience and conversion.

After all these are prepared, a large number of combined tests of various factors are needed, and finally the most effective solution is selected through testing. At the same time, testing should become a long-term auxiliary work accompanied by recall, constantly testing the best copy, the most accurate classification, and the most matching solution.

After the test, pay attention to monitoring the data results. Generally, it is based on the user's return status within 24 hours after sending the text message. In the initial stage, you can only require the user to return and open the APP to be successful, and then gradually improve the standards and monitor more operations of the returning users.

Finally, whether it is ordinary operations personnel or operations teams, do not treat the creation of a lost user recall system as an independent job, but make it a part of driving products and individuals.

For companies and products, long-term and stable recall of lost users can drive products, technologies, and markets to continuously optimize themselves, reduce user churn rates, and improve user recall effects.

For individuals, especially new operators, although the task of recalling lost users is simple, it can help you establish a complete user operation mindset, cultivate the habit of testing and data monitoring, exercise your communication skills, and prepare for more difficult work in the future. So no matter what, this work should not be underestimated or ignored.

Well, I have finished talking about building a system to recall lost users. I hope it can give you more thoughts and inspiration. I will also write more systematic articles on user operations for you in the future.

postscript

Around mid-2018, I was invited by Phoenix News to give a lecture on user growth to their operations team. After the lecture, a friend present asked me about the issue of recalling lost users.

I think my answer at that time was not very good. After returning to the company, I did a lot of testing, practice and research in this regard. Together with the team, I helped the company build a simple but effective system for recalling lost users. This article was finally completed, which can be regarded as the final answer to this question.

Source: Liu Weidong's Operation Notes