It’s been a little over a week since Pokémon GO was released, and as a result of an increase in the number of users seeking more “real-time information,” I predict that the number of Twitter users is on the rise. What about at
The theme of this article is to reconsider the merits of cohort analysis, regardless of Pokemon GO . If you are reading this blog, I think that there are many people who think that “cohort analysis is that function of Google Analytics”, but this time I would like to proceed with the discussion on Twitter instead of Google Analytics. increase.
About cohort analysis
What is cohort analysis?
Cohort analysis is an analysis method for visualizing “changes in user behavior over time”, and it is common to use a chart like the capture below.
How many days (weeks) does it take from first contact to conversion (purchase) on EC sites that are difficult to convert on first contact? can be displayed in an easy-to-understand manner. Also, by using cohort analysis on SNS sites, it is possible to understand the retention rate of users. You will be able to gain insights such as “ There is.”
Furthermore, if the cohort segments (vertical axis) are time-series items such as daily, weekly, and monthly, “From a certain timing (= timing when some site renovation was performed), the user retention rate improved greatly.” You can also get insights like In addition, by using this segment (vertical axis) as the inflow route information of the first contact instead of the time axis, it is possible to say, “This campaign had a high CPA, but the user retention rate was high, and overall performance was high. You can also get the insight that
How to make a cohort
Google Analytics also provides a report for cohort analysis in the form of a beta version, and if the cohort segment (vertical axis) is only the time axis, it can be used from the Google Analytics management screen. Also, using the Google Analytics Core Reporting API v4 released in April 2016, it is possible to create not only time-series cohorts, but also cohorts by first contact referrer, medium, and campaign.
Also, even if it is not Google Analytics, it is possible to create a cohort chart from the original data such as Excel using a pivot table etc., although it will take some work. The above capture is actually a cohort chart created using Google Sheets. For example, if you want to create a cohort chart for “mobile phone contract cancellation status”, prepare an Excel table for the new contract date and cancellation date for each subscriber, create a pivot table from them, and add some formulas. cohorts can be created by
Real-world cohort example
overview
This time, I created a program that uses the Twitter API to obtain follower data for a specific Twitter account on a daily basis. Since this program has been running since the end of March, I have about 4 months worth of follower data for the relevant Twitter accounts (7 accounts). You can generate follow and unfollow dates (or are you currently following) for your account. In other words, based on Twitter follower data, it is possible to express the follower retention rate as a cohort diagram .
Now, let’s take a look at the cohort chart for 3 of the 7 accounts below. Also, think about the insights you might gain from the generated cohorts.
Example 1
This is a cohort chart of accounts with around 12,000 followers and media related to web marketing.
What we learned from the cohort
First, looking at the weekly cohort vertical axis “number of acquisitions”, it has increased significantly since week 20, and over 700 new followers have been acquired in week 22. Even if the media article goes viral, it is unlikely that the number of followers will continue to increase for 5 consecutive weeks, so I think that at this point in time, “I used Twitter ads to increase my followers.”
Next, let’s look at the horizontal axis. From the 13th to the 15th week, it seems that the continuation rate drops significantly from the 2nd to the 3rd week. In addition, it decreases smoothly after the 4th week, and there seems to be no significant attenuation point. Then, how to keep followers for 3 weeks becomes important. Now, looking at the decline rate from the 2nd week to the 3rd week in the 20th to 24th weeks when Twitter ads were supposed to be implemented, all of them maintained a continuation rate of 80% or more. No weeks below 80%. In addition, the 20th week, which is the first week, has a very high figure of 88%.
Example 2
It is a cohort chart of accounts that have around 22,000 followers and are doing a bit of a maniac consumer service.
What we learned from the cohort
First of all, looking at the vertical axis of the daily cohort, “Number of Acquisitions”, we continuously acquired a large number of followers from April to early May, but after May 10th, the number of followers increased significantly. depressed. After May 10th, there are many days when the number of followers is in the single digits, and there are four days when no followers are obtained. That said, there are some days like May 31st and June 1st, when there is a big increase, and these are considered to be due to external media (or regular tweets) other than Twitter Ads.
Next, let’s look at the horizontal axis of the weekly cohort. In the first example, the follower retention rate for 3 weeks was around 80%, whereas for this account, the follower retention rate for 3 weeks was mostly over 90%, maintaining a high retention rate. I can say Furthermore, this maintenance rate continues for about 5 weeks, so I think it can be said that it is an amazing continuation rate. Also, comparing the retention rate from weeks 13 to 18, which is thought to have increased followers through Twitter ads, to the retention rate from week 19 onwards, the follower retention rate during the Twitter ads period was high, making it a very good target. It can be said that we are able to reach the demographics.
In normal Twitter ads for gaining followers, “whether or not the follow button on the ad was pressed” is considered as a performance indicator, but in some cases, “users who followed once but immediately unfollowed” can also be considered as “acquisition”. It will be evaluated. As a result, depending on the targeting settings, there is a good chance that “I was able to keep the acquisition cost low, but it disappeared in a few weeks.” In order to avoid such a situation, it is recommended to build a state where you can see an indicator such as “Acquisition unit price for maintaining follow-up for XX weeks” for each campaign.
Example 3
A third example would be a cohort of well-known blogger accounts with around 68,000 followers. This is data from a larger account than in Examples 1 and 2.
Cohort chart (by week)
What we learned from the cohort
First, looking at the vertical axis “number of acquisitions” of the daily cohort, it seems that the average number of followers is over 100 per day. The number of followers acquired on May 23rd and the number of followers acquired during the week are abnormally high, but when I checked the data, it seems that the data for May 22nd was not acquired properly, and as a result, 5 It seems that there was an abnormality in the data on the 23rd of the month. If you are going to investigate an account of this scale, it seems necessary to carefully consider in advance whether there will be any problems with data acquisition processing. From now on, I will ignore the figures for this day and this week.
Next, let’s look at the horizontal axis of the weekly cohort. Again, the follower retention rate for 3 weeks is around 80%. Furthermore, in the first and second examples, the follower retention rate remained almost constant after the third week, and there was no significant decline, but in the third example, the follower retention rate was even higher at the eighth week. The follower retention rate has fallen to around 75%, and after 10 weeks, the retention rate seems to have fallen to about 70%.
summary
This time, I created a cohort using information other than the website, and tried to mention what I learned from the obtained cohort. Cohort analysis is an analysis method that can be used in areas other than the web. Also, in the (time series) cohort chart,
- How to increase the number of new acquisitions by looking at the actual value on the vertical axis
- By looking at each percentage on the vertical axis, the change in the retention rate
- By looking at each percentage on the horizontal axis, the timing of the high withdrawal rate
It leads to knowing Furthermore, the second and third of these are difficult indicators to notice without looking at the cohort chart.
Why don’t you start by looking beyond the website and looking for areas where cohort chart analysis can be used?
