Website Data analysis: How to reasonably define user churn

Intermediary transaction http://www.aliyun.com/zixun/aggregation/6858.html ">seo diagnose Taobao guest stationmaster buy cloud host technology Hall

Haven't updated the blog for a long time, this article again write about "user churn" content. Before the release of active users of the site and the loss of users this article on the site's active users, loss of users and the loss of new users have done a definition, where the loss of the user's English name, the general loss of users commonly used in English as "churn user", before the wastage, Moz, Lost are not too standard. Later, the friends who do correlation analysis ask the loss of the user's length of time to choose how long is reasonable, in particular, "site analysis of actual combat" after the publication of this book, I have mentioned in it how to more accurately define the length of the loss of time, may explain the relatively simple, or have a friend message feedback this aspect of the problem, So here's another article to explain.

Lost Users and return users

The definition of the lost user please refer to "Active users and lost users of the website" This article, to explain how to reasonably define the length of the user lost time period, we need to introduce a new concept of indicators: return visit users. A return visit here does not refer to Google Analytics above returning Visitor (relative to the new user, referring to the site before the user visited the site again), where the return visit users refer to the loss of users once again visit the site, that is, the user has been lost, Meet the lost time limit completely without access/login site conditions, but then re-access/login site. Then, according to the number of users return visit can be calculated by the user return rate, namely:

User return rate = Return to users number of users x100%

The value of a return visit to the user rate can indirectly verify the rationality of the definition of user churn. Under normal circumstances, the user's return rate should be relatively low, from the business point of view, if the definition of the loss is reasonable, so it is difficult for those who have lost interest in your site to visit your site again. Under normal circumstances, the site's user return rate should be below 10%, in the value of about 5% is more reasonable, for the mature site users return rate will be slightly higher, and the new site users return rate is usually lower, especially like mobile phone app such users easily lost products.

Loss period and user return rate

The length of the loss period is inversely proportional to the user's return visit rate. We define the user lost in the continuous use of not access/login site for the longer term, the loss of this batch of users return to the site after the probability will be lower, and with the definition of the loss of the duration of the increase, the user return rate must be decreasing, and gradually approaching 0. So if you choose the appropriate length of the loss period? We can set different length of the loss period, further statistics the user return rate of each loss term, and observe the user return rate with the definition of the loss period of the convergence rate increases. If you set the duration of the loss in the unit "Week":

We can use the inflection point theory (Cay method) to select the most appropriate loss cycle according to the change curve of the user return rate of the different loss cycles set.

Inflection point theory: An increase in the number of values on the x-axis will result in a large gain in the y-axis value (minus the benefit), until after a certain point, when x increases the data gain (decrease) of y is greatly reduced, that is, the marginal income in economics decreases sharply, that is the point of the chart. For example, when the loss cycle increases to 5 weeks, the rate of reduction of user return is significantly reduced, so the 5 weeks here is the inflection point, we can use 5 weeks as a definition of the user's loss of time, that is, a previously accessed/logged in user, if after 5 consecutive weeks without access/login, then define the user drain.

So, after this approach, can be more reasonable definition of the loss of the user's statistical logic, and before doing is to select different loss of time to calculate the user's return rate, and then use the statistics to generate such as a smooth line with a scatter plot, the problem will be solved.