[Dry goods] Take you to get the ROI evaluation of new customers

about the author

@花花

Worked as a data analyst in Meituan, Tencent, and Toutiao.

After tens of billions of resource evaluations, we will grow and learn with everyone.

1 Introduction

In the previous issues, I shared the ROI evaluation methods of marketing activities such as AB experiment, PSM, DID, etc. These methods are mainly for the old customer group. New customers are a special group. When evaluating the ROI of new customers, a new one is often introduced. Concept LTV, use LTV/CAC as the ROI of new customers. Typical scenarios that require new customer ROI evaluation are as follows:

Off-site advertising channel selection: In the initial stage of new business user growth, it is often necessary to carry out advertising in different channels. At this time, how to invest limited resources into the channel with the highest ROI is an important decision facing the company.

Evaluation of the effect of new marketing tools on the site: There are many marketing methods for new marketing. Which marketing tool has the best effect?

New customer quality change monitoring: On the basis of the same cost input, how does the value of new customers change?

This article will involve some basic concepts and some English abbreviations. In order for everyone to better understand, these basic concepts are explained here.

ARPU: Average Revenue Per User, which is the average revenue per user, which refers to the total value of a single user.
LT: Life Time, the user life cycle, refers to the interval time from the first use of the product to the last loss.
LTV: Life Time Value, which refers to the value generated by users in LT. Different businesses have different definitions of value. Common methods of definition include revenue, net profit, transaction volume, etc.
CAC: Customer Acquisition Cost, or customer acquisition cost, refers to the average cost of acquiring a single user through various channels.
ROI: Return On Investment, the rate of return on investment, refers to the output brought by unit cost.

2 New customer ROI calculation

New customer ROI=Total LTV of the same batch of new customers in LT/Total CAC of this batch of new customers in LT. There are three issues involved: LT calculation, LTV calculation, CAC calculation.

01 LT calculation

2.1.1 Direct reference to LT=X

Defining LT directly as 1 year, 2 years or 3 years is one of the commonly used methods in the industry. The advantage is that it is simple and clear to determine a period of time and then observe the user's performance during this period. The disadvantage is also very direct, because this is not Represents the real life cycle of the user. There are also definitions of LTV as 18 months, mainly referring to the average period of users changing phones of 18 months. It is believed that after users change phones, many apps need to pay to acquire users again, but this method is only a rough estimate. It is not a reflection of the real user life cycle.

2.1.2 Complete period statistical method

Directly count the time from the first use of each user to the end of the first life cycle, and record it as the LT of a single user, and then add the LT of all new customers and divide by the total number of new customers to get the average LT. For example, if there are 300 new users in a batch, the LT for 100 users is 100 days, the LT for 50 users is 200 days, and the LT for 150 users is 300 days, then the average LT of the batch of users is as follows:

LT=(100 100+50 200+150*300)/300=216.7 days

When calculating the LT of each user, there is a very important point that needs to be confirmed under what circumstances the user counts the first life cycle to end. You can refer to the inflection point of retention rate stability and the inflection point of cumulative ARPU contribution to stabilize. But it is better to define it in conjunction with the business.

For example, in a relatively high-frequency business, it is often defined that the user has no visits/no clicks/no transactions for 1 month/2 months/3 months/6 months is regarded as the end of the life cycle. In most e-commerce industries , It is generally defined that a user has no transaction for 6 months is the end of a life cycle. As for no transaction for more than 6 months, but re-returning after a period of time, the re-returning can be recorded as a transfer of new users, and the second life cycle is calculated from the date of transfer to the end of the life cycle, so that new customers can be calculated LT for new users, LT for new users.

2.1.3 Retention rate estimation method: LT=1/(1-retention rate)

For example: For example, if the total number of new users is 10,000 and the retention rate for the next day is 10%, it is equivalent to: For this batch of new users, the number of users on the first day is 10,000, the retention of the second day is 1,000, and the third day The retention of 100 people, the retention of the fourth day is 1 person, and the fifth day is gone; then the total retention days of this group of people is 10000+1000+100+1=11111 days, then the average LT of this group = 11111/10000=1.1111 days, which is actually equivalent to 1+0.1+0.1²+0.1³... = 1/(1-0.1)= 1/0.9, which can be expressed by the following formula, where x represents the retention rate:

This method is suitable for businesses where the retention rate of the next day/week/month is relatively stable, and the retention rate can be selected from daily, weekly, and monthly dimensions. It’s worth noting that there are differences in the definition of retention here. Some think it is retention if you have visited, and some think retention is only after experiencing the core functions, such as playing short videos and placing orders. You must choose according to the actual situation. How the business appeal is defined.

02 LTV calculation

To calculate the value generated by the user in the LT, there are two types of calculations. The first is the calculation of historical data, and the second is that the LT that is not yet complete needs to be predicted. For the first type of historical data calculation is relatively simple, we can directly count, the point is that the definition of Value may be different. For the life cycle that is not yet complete, but need to calculate LTV, our common methods are:

Time percentage conversion estimate

Assuming that the value of the new customers in the current batch to be evaluated in X days is Y0, the value of the previous batch of users in the complete LT is Y, and the value in X days is Y1, then the value of the new customers in X days Approximately Y1/Y of the total life cycle. Then, according to this value, the value in the lifetime of the batch of users is calculated as Y0/(Y1/Y).

For a direct example, the user’s LT is 180 days, the value generated by the historical user in 180 days is 1,800 yuan, and the value generated within the first 30 is 600 yuan. So now the new batch of users has generated a value of 500 yuan in 30 days, but has not experienced the complete 180 days, then the value that may be generated in the entire life cycle can be estimated based on the value of this batch of new customers in the first 30 days: 500/(600/1800)=1500 yuan.

Algorithm estimation

It is transformed into an estimate of the value of this batch of users in the future. This generally requires a prediction model based on the relationship between LTV and LT. Commonly used are multiple linear regression, logistics regression, decision trees, random forests, etc., but for prediction The accuracy of the model has a greater test.

03 Total cost calculation

For non-transactional companies, the cost may only be the initial cost of guiding users to download and activate. However, for transactional companies, in order to stimulate users to continue to place orders, it often involves additional subsidies, Push SMS and other customers. To maintain the cost, it is generally recommended to include all the costs in LT to reflect the total input cost more truly.

CAC = channel new cost + subsidy cost + all other customer maintenance costs.

04 ROI calculation

After calculating LTV and total cost, ROI=LTV/CAC, the higher the ROI, the higher the output brought by the unit cost input. Analyzing ROI from different dimensions can have different business interpretations. For example, adding the channel dimension can evaluate the input and output of different channels of advertising investment, and adding the time dimension is the evaluation of the quality of new customers in different time periods.

In actual operation, if it is only related to the comparison of the ROI of different channels, sometimes it is not necessary to calculate the complete LTV, and sometimes choose 7-day user value/7-day user cost, 1 month user value/1 month User costs. Because if you compare the ROI of different channels alone, users' data within a week or a month can already explain some problems. You can also compare short-term calculations with long-term real values ​​to determine the error between using short-term data and long-term real data, and determine whether the responsible business can be estimated with short-term values.

3 Postscript

The ROI of new customers is only one dimension of channel quality evaluation. In channel extension, “quality” and “quantity” may be concerned at the same time. Different business development stages will give “quality” and “quantity” different weights. Especially when it is time to quickly seize the market, it is often necessary to preemptively seize users and form a scale advantage. At this time, "quantity" may be more important than "quality", and the problem of "quality" can be made up after stable development.

The private place of a data person is a big family that helps the data person grow, helping partners who are interested in data to clarify the learning direction and accurately improve their skills. Follow me and take you to explore the magical mysteries of data

1. Go back to "Data Products" and get <Interview Questions for Data Products from Big Factory>

2. Go back to "Data Center" and get <Dachang Data Center Information>

3. Go back to "Business Analysis" and get <Dachang Business Analysis Interview Questions>;

4. Go back to "make friends", join the exchange group, and get to know more data partners.