E-commerce case: Detailed explanation of data-driven strategies for retaining old customers!

background

E-commerce platform A has been established for nearly 10 years. It is not the top in the industry, but it is also a leader in its niche field.

This year's economic downturn has seen the daily active users of the APP drop from 500,000 at the beginning of the year to now, with the operation still struggling to keep the bottom line of 400,000 , and the boss racking his brains to coax investors .

Meeting.

"We can't go on like this. We have to think of a solution." The boss put out the cigarette that was not even half burned.

"Let's talk about a few cooperative platforms and exchange resources to get some traffic ?"

“Invest in a few content KOLs , or some advertisers?”

"Let's do a few more promotions, and push them together to increase the daily activity."

BD, marketing, and operations offered suggestions, but the boss remained silent.

The mobile sector was in a downturn in 2018, and the industry was trying to preserve its volume, so no one dared to invest heavily in it. Promotions may be able to stop the losses, but to completely solve the problem, new traffic must flow in.

After a long silence, the boss asked, "How many old customers have left the platform over the years?"

"About eight or nine million."

"Think of a way to win back some of these users."

method

The boss is right. From experience, recalling lost customers is definitely cheaper, more accurate, and has better traffic quality than attracting new customers.

What to use to reach?

Eight or nine million old customers have left their mobile phone numbers, names, addresses and other information during the registration and ordering process. It is not difficult to reach them. The simplest, most efficient and low-cost way should be text messages .

Tell them what?

What kind of text messages should I send to attract them back?

As far as e-commerce is concerned, there are generally three ways to play:

1. Promotional offers, such as: 70% off on all popular summer cosmetics, SK-2 Facial Treatment Essence for only ¥698, hurry up and grab them! ;

2. Give away coupons, such as: The weekend is coming, give you a 88 yuan food coupon, click the link to receive it!

3. Product collection, for example: You haven't come for too long, the boss said I was derelict in my duty, so I'll give you a ray mask worth 59 yuan, come and collect it!

The first method is less expensive (only the SMS fee is required), but the response rate is poor. It is generally used in conjunction with promotional activities and is not suitable for recalling old customers. The second and third methods generally respond well in SMS or Push. However, as a non-self-operated platform, coupons mean a high 1:1 subsidy cost . At the same time, for lost old customers whose purchasing needs are unclear, coupons are not a necessity and have limited appeal. Relatively speaking, “get hot-selling products for free” is not only highly versatile, but also has lower costs for large-scale procurement.

Therefore, we adopted the third method: sending SMS to notify these lost old customers and invite them to return to receive free gifts.

Where to attract them?

Where to get it? APP?

Lost old customers are those who have not visited the app for 180 days , and most of them have uninstalled the app. The cost of reactivating the app is too high and impractical.

If not the APP, where to go?

In the e-commerce industry, “free giveaways” are generally accompanied by social activities such as group buying, sharing, and bargaining, which further buffer costs by attracting new customers through fission , such as Pinduoduo’s bargaining assistance and MissFresh ’s 0-yuan fruit eating. The best breeding ground for social fission is undoubtedly mini programs . As an extension of the WeChat ecosystem , the A platform mini program is lightweight and fully functional, making it the perfect base for returning old customers.

In summary, we will notify lost old customers via SMS and offer them gifts when they return, introduce them into the mini program, and generate a round of fission through social gameplay.

The activity flow is as follows:

In the above process, whether it is the host or guest state, starting a group or joining a group, after the group is formed or fails, it does not mean the end of the operation work. When the user participates in this stage, we have obtained their WeChat OpenID and the permission to push template messages. At this time, based on the user's status, we can push promotional products, coupons or other activities to encourage and undertake further conversion of users .

Now that the event background and planning are clear, let’s talk about how to use data to track and evaluate the effectiveness of the event.

data

We evaluate the effectiveness of the above activities from four dimensions:

1. Response Data
2. Fission data
3. Value Tracking
4. Cost Assessment

These four dimensions are basically suitable for evaluating all micro-operation activities .

Below we will expand on each one and give examples based on the previous activities. Based on business rules, the data will be trimmed within a reasonable range.

1. Response data

The response data is what we often call the "funnel" .

Suppose we put an advertisement on Toutiao . Over a period of time, 1,000 people clicked on the advertisement, 200 people registered after clicking, 50 people downloaded the APP according to the prompts after registering, and finally only 10 people completed the consumption conversion on the APP.

Here, from 1000 to 10, it's a simple funnel. Any activity, from the first specific action of the user after being reached to the final conversion, is a funnel . The funnel shows the direct results of an activity and breaks them down into individual steps, providing attribution for the results at each level .

The following is the conversion funnel of this old customer recall activity:

The “password input ratio” is low, which may be because the text message copy is not sharp enough and the product selection is not attractive enough ;

The ratios of "number of people sharing the link" and "number of people starting a group" are both good , indicating that after users enter the password, the guidance and process of forwarding and starting a group are clear and concise, and the user operation threshold is low ;

The overall “group formation rate” is 7.79%, which is lower than the industry level . It is speculated that it may be because three people are too many to form a group and it is too difficult, or it may be because these users find it troublesome or are unwilling to abuse their social value .

Through the funnel, we can not only see the overall conversion level of the activity , but also examine the data performance of each link in detail, so we can see at a glance where the problem lies .

“Where is the problem” is only the first problem that the funnel solves; the more important one is “how to solve it” . The above is the "trunk" of the funnel. Here are the "branches" of the funnel:

Regarding the issue of "low input password", we guess it may be related to copywriting, product selection, delivery time, etc. Taking product selection as an example, for comparative testing, we set up a control group with a small number of people (5,000 people) for each group before the event, and sent text messages to verify what kind of products are more attractive to this group of people .

Obviously, users under 24 years old are more willing to get eye drops for free, facial masks are more attractive to users aged 25-29 and 30-39; users over 40 prefer toothpaste .

We found that the above table includes not only the variable "product selection", but also the intersection of the two variables "age stratification" and "product selection". In an activity, the variables that affect the results of a certain link are often numerous and intertwined. Taking this activity as an example, the factors that affect the "ratio of password input" are not so much a single variable as a combination of variables such as "age and product selection", "age and copywriting", or even the optimal combination of all relevant variables such as "age", "product selection", "copywriting", and "sending time" .

Therefore, the "branches" of the funnel are to seek the optimal combination of various variables . In practice, the combination of variables will produce exponential growth. Three or five variables may combine into hundreds of situations, and it is impractical to test them one by one. Therefore, variable optimization must be based on business judgment. For example, in this example, based on business experience, users of different ages should have different product demands, so we conducted variable tests on "age" and "product selection"; similarly, we believe that there is no clear correlation between "product selection" and "SMS sending time", so we use "SMS sending time" as a separate test variable. Whether it is verifying a hypothesis or mining insights, data is definitely just a tool for the business, not the business itself .

2. Fission data

In the planning of an operational activity , if you do not take fission as the starting point of the entire activity, you should at least regard it as a key value link.

Whether it's the red envelope sharing of Ele.me , the bargaining assistance of Pinduoduo, the fruit eating for 0 yuan of MissFresh, the ticket grabbing assistance of Ctrip , or the recent circle of friends phenomenon of NetEase Jungian psychological test and "My Coffee Shop" of Lian Coffee, these textbook-level activities or gameplays, whether conversion-oriented or communication-oriented, all revolve around the value point of "socialization". Even for activities outside the WeChat ecosystem, such as Alipay 's Spring Festival Five Blessings Collection, the highly social "Giving Blessing Characters" is also the core source of participation and conversion in its activities.

The reason why activities have the potential to achieve great results with little effort is that they rely on social communication behaviors such as forwarding, sharing, and liking.

Having said so much, I simply want to prove that as operational activities become more and more social, tracking and analyzing "fission data" is by no means dispensable, and is gradually becoming an important data indicator for evaluating the effectiveness of activities.

Let’s return to our case.

As can be seen from the activity SOP diagram, there are two main ways of fission in this activity:

• Main state fission: In order to get the product for free, users shared the activity to 3 different groups. When group members see the link, they can also share it 3 times to get the opportunity to start a group with 0 yuan and become the new leader.
• Guest fission: After the main user opens a group, he invites new users to join. If the invitation is successful, a guest fission is completed.

The first round of fission demonstrated the first round of fission effect brought by SMS push users. The subsequent fission refers to the uncontrollable fission caused by the activity link being shared by SMS push users to 39,859 groups. Due to the embedding of activity pages, subsequent fission cannot continue to be tracked in layers, although this is necessary (the groups forwarded to by users of different ages may be of completely different quality).

The number of shares and the number of groups opened in subsequent fission represent the main fission ; the number of participants represents the guest fission .

• From the table, the first round + subsequent fission were shared to 39859+13477=53336 groups in total. However, only 11211 people clicked on the link, and the average click rate for each group was only 0.21 . This is a very low number, indicating that there are probably big problems with the presentation and interest points of our links .
• Compared with the first round of fission, the subsequent fission has a significant decrease in the number of shares per capita, the proportion of group participants, and the group formation rate. On the one hand, it is because the two groups of users are exposed to different information . On the other hand, it is because the people who received the text messages are old customers of platform A and are more accurate, while the users generated by fission are uneven and cannot be targeted, and the accuracy is poor .
• We have also analyzed the problems of low group formation rate and small number of group participants in the previous article.

In summary, the fission of this event is not ideal. The display, product selection, copywriting and other aspects of the sharing session need to be optimized. If it is still not ideal after optimization, the logic of the entire event may need to be reconsidered.

3. Value Tracking

Holding the response and fission data, the operation team walked into the boss's office with confidence.

He reported the key points of the conversion and fission data to his boss in a concise manner, made business-level attributions, and put forward some suggestions and arrangements for optimizing the next execution plan.

The boss looked at the data and nodded slightly, thinking about something at the same time.

At this time, the young operations manager said a summary, which angered the boss.

“I think the effect of this activity is pretty good. If the logic of fission is optimized, it can be done in large quantities.”

A scolding.

In fact, no conclusions should be drawn before completing the value tracking of converted users .

The conversion of each user means real money costs and subsidies , which must be offset by the profits brought by subsequent purchases . If the user repurchase and activity brought about by the activity are extremely poor, then no matter how high the conversion rate is, the activity is a failure . Because your event attracted a group of people that the operations industry fears - the freeloaders .

It should be emphasized that the so-called "conversion" does not only count when a group is established . Users who jump in any link such as sharing, registration, and joining a group are considered conversions as long as they have a relationship with us , but the degree of conversion is different.

In order to improve the subsequent performance of users after conversion, operators usually use push or text messages to provide further incentives. As in this example, for users who successfully start a group, fail to start a group, successfully join a group, fail to join a group, and share the link less than 3 times, we will push personalized coupons, discounted products, event links, etc. to them within 24 hours according to their behavioral tendencies based on the degree of conversion . This is to promote subsequent conversions and do our best to retain these users.

The value tracking data is relatively complex. Based on user stratification, time stratification, behavior stratification, etc., a huge value tracking system can be formed . Considering the length of the article, the author has simplified the value tracking for this example and divided it into the following two parts:

(1) Key value tracking

There are three key points in this table:

1) Who to track

Of course, it is the converted new customers. However, it is necessary to track in layers . In addition to the age layering of SMS push, the conversion level layering should also be included. This way, the value differences between users with different conversion levels can be tracked.

As shown in the table above, we can see that "users who started a group and formed a group" have a higher repurchase frequency than "users who started a group but did not form a group"; the value of fission users is inferior to that of the push group in all aspects .

2) How long to chase

Determined based on product usage frequency and user life cycle . Generally, e-commerce companies track data for 90 or 180 days. If data support is urgently needed for activity optimization, short-term tracking data can be provided.

3) What to chase

It must be the most critical value indicators that track users for the product . For self-media , it may be clicks, reposts, and rewards; for games , it may be activity and payment; and for e-commerce platforms, it is naturally transaction data such as GMV, number of orders, and average order value.

At this point, we have figured out how much GMV each tiered user contributed in different time periods, and then multiplied it by the platform transaction fee, it is easy to calculate the profit earned by the platform . Through these profits, on the one hand, we can intuitively judge the value of users in different tiers ; on the other hand , we can judge the effectiveness of the activities by comparing with the costs and decide whether the activities can be productized or replicated on a large scale .

When the product is relatively simple and the activity is small, the above value tracking can basically prove the value of the activity. However, if the product ecosystem is complex and user consumption behaviors are diverse, it is necessary to track user behaviors in layers .

(2) Behavior hierarchical tracking

In the same product, different users contribute different values ​​to the platform based on different behaviors, and jointly build a complete ecosystem.

Taking Zhihu, the most representative UGC content platform, as an example, its user behaviors and corresponding values ​​are roughly divided as follows:

Compared with Zhihu, the ecosystem of e-commerce platforms may be more complex :

In the e-commerce ecosystem, it is not as simple as just “buying” and “selling” . Search-and-buy users contribute most of the GMV and are the basis for the survival of the platform; users who like to "browse" are keen on discovery-based shopping. In addition to contributing considerable GMV, they also provide feedback on the popularity of different theme banners, modules and products through clicks, stays and other behaviors, thereby pointing out the direction for operations; those users who are keen on group buying and helping to bargain gather on WeChat to exchange their labor for platform discounts, constantly introducing new traffic to the platform; even most of the current e-commerce platforms have introduced content sub-ecosystems such as guides and reviews, and users who like to write reviews are the core support of this sub-ecosystem.

Therefore, we cannot judge user value based solely on purchasing power. In other words, we cannot conclude that an activity has failed simply because the users attracted to the activity do not buy enough . Because these users are likely to contribute other important ecological values.

Judging from the data, the users brought by this activity are more prominent in starting groups, helping to bargain, and following service accounts . This shows that users are mostly gathered in WeChat mini-programs and have good communication value . It is speculated that this may be because the activity is played in a group, so the traffic brought in is more or less a bit of wool attribute, which also reduces our expectations for its repurchase value. These users are more suitable for peripheral dissemination of the platform in mini programs .

4. Cost data

“Cost data” and “value tracking” are the yin and yang of the activity. The difference between value and cost basically represents the initial benefit of the entire activity.

For this activity, the costs are mainly incurred in three places:

• The cost of SMS is about 0.03 yuan per SMS for lost old customers;
• Free goods subsidy costs , all group orders need to pay about 20 yuan of goods costs;
• Subsequent incentive subsidies ; after the user conversion, we may push some coupons to promote repeat purchases. Coupons are the main cost in this process.

In the third step, we track the total GMV of repeat purchases, multiply it by the platform fee to get the profit, and then subtract the sum of the above three costs to get the profit of the activity.

The logic of cost data is simple, the only thing worth noting is whether to stratify. Different layers have different user values ​​and consume different costs . If you use the same layer to track value and cost, you can calculate the ROI of different layers, find the break-even point, and selectively push it to increase the revenue of the entire activity . But most of the time, due to limitations such as limited user lists and data embedding, it is difficult to achieve layered ROI optimization. Therefore, whether to divide into layers or not depends on the business situation.

Finally, here is a mind map: