How to use data to mine potential users?

This article mainly focuses on the ways to use data to mine potential users. The author provides five practices for data people by giving specific examples, which effectively solves the problem of mining potential users. It is recommended for data people who want to mine users.

"Which customers have potential?" is a difficult problem for many students in their daily work. What is considered potential? How to use data mining? It is not clear. Often, students who work on data work hard, and then the business complains: "I knew it a long time ago, what basis do you have for what you said!"

1. Common Practices

Let’s look at a specific problem scenario:

An Internet advertising platform, whose main customers are Class B enterprises, now needs to tap into potential customers. How should it conduct the analysis?

Note: The title is about toB business, and the biggest difference from toC business is that B-side customers have fixed industry affiliations (automobiles, beauty, fast-moving consumer goods, games, etc.), and B-side customers also have differences in business scale (which can be obtained from the number of employees in the company, publicly available financial reports, third-party industry reports, etc.). Therefore, if you do your homework in advance, you can label the company and classify the industry and scale.

So many students start to do cross analysis as soon as they get the question. They use the industry and enterprise size labels as classification dimensions and cross analyze the amount of user advertising. They get a table like the one below and then start to say: Because enterprises in industry A place more advertisements, enterprises in industry A are all potential customers... (as shown in the following figure)

Of course, this result will be criticized by the business:

"I knew it!"

“Since it has potential, why can’t we do it?”

"Other companies are not investing much at the moment. Can you guarantee that they won't invest much in the future?"

So, what's the problem?

2. Difficulty of the Problem

Essentially, the difficulty of the problem lies in how to define the word "potential" in business terms.

The data can only tell you:

• The industry and size of this user
• How much money has this user invested in the past
• What type of advertising has this user run in the past?

However, these are just descriptions of the current situation and cannot be used as a standard for judging "potential". If we do not talk about the standard and simply describe the current situation, problems and doubts will arise:

• You said that what was invested more in the past was potential, I knew that a long time ago...
• You said that what was less invested in the past was potential, why didn’t I achieve it?

There is no solution to the similar chicken and egg problem.

Therefore, the solution to this problem is not to crisscross the data here and there, but to get back to the business and see what the so-called "potential" of the business means in different scenarios.

3. Differentiate between scenarios

As an advertising business, there are three key customer attributes that determine whether customers will continue to place ads:

• Industry attributes: Some industries have a long-term and rigid need for advertising (games, beauty, clothing, etc.), while others have seasonal advertising peaks (real estate, automobiles, durable goods, etc.).
• Customer familiarity with the platform: Different advertising platforms have different ways of working, and customers need to familiarize themselves with the process.
• Customers’ earnings on the platform: If the advertising is effective, they will naturally invest more; if it is not effective, they will invest less.

Therefore, when distinguishing scenarios, customers must first be distinguished according to the following dimensions to facilitate subsequent identification work:

• Customer industry classification (industry name)
• Customer size (number of employees/industry ranking/turnover)
• Customer lifecycle (onboarding, trial, peak, decline, churn)
• The customer's past advertising results (based on advertising ROI: good, medium, poor)
• Customer value stratification (by sales contribution: high, medium, low)

Secondly, as a toB business, customer churn may not be entirely a problem of our platform, but may very likely be a problem of the customer’s own industry/enterprise.

for example:

• The client industry is in decline
• The client industry is fine, but the business is declining
• The client company has no problem, but they don’t know how to play

If it is case 3, it can still be regarded as potential, after all, there is still an opportunity for educational incubation, but if it is case 1 or 2, it is really hopeless. Therefore, it needs to be treated differently. Note: Data collection here will be very difficult. Because it may involve dozens or hundreds of industries and tens of thousands or even hundreds of thousands of companies, it is too difficult to collect them one by one. Therefore, it is necessary to do a good job of basic classification first, and then focus on them one by one.

Thirdly, as a toB business, the business side has more one-on-one customer communication opportunities and the possibility of guiding customers. Therefore, it is necessary to do a good job of benchmark analysis to find out the successful cases/successful benchmarks in the key customer groups. This can help the business side better follow up on potential customers and also answer a very fatal question that the business side often asks: "Why do you say it is potential, but it is not manifested?" The answer is: "Dear, that's because you didn't use the right method..."

After completing the above preparations, you will have the basic materials for analyzing potential customers and can officially start work.

4. Defining Potential

Considering that the amount of data that can be collected is different in different life cycles, it is recommended to use life cycle as the main division dimension and make judgments based on other information.

1. Initial entry period

The overall idea of ​​the initial analysis is to make one order. Try to activate the newly acquired users to recharge/first launch. Because there is no user launch data in the initial stage, it can only be classified according to basic information such as industry and scale. Here, you can conduct a portrait analysis of existing high-value users, find out the characteristics of industry, scale, etc., and focus on customers with similar characteristics in the initial stage (as shown in the figure below).

2. Trial period

During the trial period, some user data is available, and users have begun to differentiate: some users have indeed made money after the campaign, while some users have had a terrible effect and their activity is already very low. At this time, the idea of ​​tapping potential is to keep the seeds. Because it is too difficult to convince users with poor results to continue the campaign, you can focus on keeping those customers who have the potential to become high-value users.

3. Peak period

During the peak period, except for users who lost during the trial period, those who can enter the peak period either have sufficient funds for the campaign or have made great gains in previous campaigns, so they all have the opportunity to continue to tap potential. The ideas for tapping potential are also different. If you have made great gains before, you can promote cross-selling and new campaign products. If the effect is average, but there is still a lot of investment, you can promote competitive product cases in the same industry in the form of services, and improve the quality while increasing payment. In short, the two directions of tapping potential are: to increase the average order value.

4. Decline

During the decline phase, prioritize identifying the customer’s decline reasons.

• Cyclical fluctuations: There is no point in saying more, people will come back anyway
• Industry-specific issues: Talking is useless, and no one will listen.
• Corporate issues: There is no point in talking too much, people have to save themselves
• Efficiency issue: If you have a chance, you can try to redeem yourself with the case.

Because the above problems all involve data collection, we can stratify the user value and focus on high-value users to solve the problem of large customer decline first. At this time, the goal of tapping potential is to keep the existing investment and save as much as possible.

5. Churn period

The churn period is similar to the decline idea. First, we need to distinguish the reasons for churn. Eliminate normal deaths and focus on high-value users. Then, when the user industry recovers and new success stories emerge in the user industry, we can explore potential, with the goal of recalling users.

5. The Big Picture

After the above classification, we have a preliminary definition of "potential users" in each life cycle. However, we do not have to push all "potential users" to the business at once. Because the business focus is different at different stages, so so much information may not be needed.

For example, if the current business focus is to attract new users, then focus on solving the potential problems in the initial entry/trial period. If the current focus is to improve productivity, then focus on tapping the peak/decline period users.

There is another situation, that is, the business has already had an action plan, and hopes that the data can find users who meet this action plan. At this time, the "potential" is actually very clear: customers who have the potential to participate in XX activities and have the potential to purchase XX products.

The approach at this point is completely different from the above, because the meaning of "potential" is already very clear, it's just a matter of matching. I will share this type of approach later.

Some students may ask: Can we build an algorithm model here without using business rules to deduce? Answer: Yes. But please note: what we need to build is not an algorithm model, but segmentation scenarios and modeling based on data characteristics. Otherwise, if we do not do everything at once, the model we build will have problems.

Author: Down-to-earth Teacher Chen

Source: WeChat public account "Down-to-earth Teacher Chen (ID: gh_abf29df6ada8)"