For every marketer of mobile application, acquiring new customers certainly requires more effort in terms of time and money. On the other hand, firm can always focus on maintaining existing customer base and gain maximum out of them. If this is the case, then predictive analysis will be the correct approach for this situation.
The primary goal of this webinar is to predict segment of Mobile application users,
* Who will uninstall the app
* Remain inactive (which will be also termed as a churner) for quite long time and are expected to churn.
Churn analysis is the approach by which we will predict the likelihood of this event to occur.
Our webinar covers:
* How to extract data from Google Analytics using R
* How to build churn model in R
* Identifying the customer/subscriber segment that are classified based on past data pattern, who are likely to churn (Study customer behavior Patterns)
Watch Full Webinar - http://www.tatvic.com/webinar/churn-analysis-for-mobile-application/
➥🔝 7737669865 🔝▻ Bangalore Call-girls in Women Seeking Men 🔝Bangalore🔝 Esc...
How to Perform Churn Analysis for your Mobile Application?
1. #tatvicwebinar
A GACP and GTMCP company
How to perform Churn Analysis for your
Mobile Apps
March 19th, 2014 Free Webinar by
2. #tatvicwebinar
A GACP and GTMCP company
Your Speakers
• @parikh_shachi
• Technical Analyst @Tatvic
• Interested in jQuery, JavaScript and Data
Analysis
• @Mrugesh_soni
• Web Analyst @Tatvic
• Interested in Statistical Modelling of
data
5. #tatvicwebinar
A GACP and GTMCP company
Overview
What is Churn
Analysis?
Why we are carrying
out Churn Analysis?
6. #tatvicwebinar
A GACP and GTMCP company
Overview
What is Churn
Analysis?
Why we are carrying
out Churn Analysis?
What are the Benefits
Of Churn Analysis?
7. #tatvicwebinar
A GACP and GTMCP company
Overview
What is Churn
Analysis?
Why we are carrying
out Churn Analysis?
What are the Benefits
Of Churn Analysis?
What data will back-
up my churn analysis?
8. #tatvicwebinar
A GACP and GTMCP company
Overview
What is Churn
Analysis?
Why we are carrying
out Churn Analysis?
What are the Benefits
Of Churn Analysis?
What data will back-
up my churn analysis?
How to extract the
desired data?
9. #tatvicwebinar
A GACP and GTMCP company
Overview
What is Churn
Analysis?
Why we are carrying
out Churn Analysis?
What are the Benefits
Of Churn Analysis?
What data will back-
up my churn analysis?
How to extract the
desired data?
Why did we select this
data abstraction
method?
10. #tatvicwebinar
A GACP and GTMCP company
Overview
What is Churn
Analysis?
Why we are carrying
out Churn Analysis?
What are the Benefits
Of Churn Analysis?
What data will back-
up my churn analysis?
How to extract the
desired data?
Why did we select this
data abstraction
method?
Why Predictive
Model?
11. #tatvicwebinar
A GACP and GTMCP company
Overview
What is Churn
Analysis?
Why we are carrying
out Churn Analysis?
What are the Benefits
Of Churn Analysis?
What data will back-
up my churn analysis?
How to extract the
desired data?
Why did we select this
data abstraction
method?
Why Predictive
Model?
Which machine
learning Algorithm
will fuel our model?
12. #tatvicwebinar
A GACP and GTMCP company
Overview
What is Churn
Analysis?
Why we are carrying
out Churn Analysis?
What are the Benefits
Of Churn Analysis?
What data will back-
up my churn analysis?
How to extract the
desired data?
Why did we select this
data abstraction
method?
How to build
predictive model?
Why Predictive
Model?
Which machine
learning Algorithm
will fuel our model?
13. #tatvicwebinar
A GACP and GTMCP company
What is Churn Analysis?
• The term “Churn” refers to customer attrition
• The process of identifying those customers
who are most likely to discontinue the use of
your service or product is known as Churn Analysis
14. #tatvicwebinar
A GACP and GTMCP company
Why Churn Analysis?
• Customer Acquistion Cost > Customer Retention Cost
• Customer might Churn before you recover Acquisition Cost
• Signals that indicate whether a Customer is about to Churn
– Cancel Account/Service
– Uninstall Mobile App
• Difficult to Retain them with standard Retention Measures
• Calls for a Pro-active Approach
15. #tatvicwebinar
A GACP and GTMCP company
Business Objective for Churn Analysis
Reduce Customer Churn
• Predictwhether a customer will Churn and when it will
happen
• Understand why customers churn and act on these reasons
16. #tatvicwebinar
A GACP and GTMCP company
Revenue as a function of Churn Rate
If you cut your churn in half, then in few years your
revenue will be increase by 50 %
17. #tatvicwebinar
A GACP and GTMCP company
Impact of Churn on your business in terms of revenue
0
5,000
10,000
15,000
20,000
25,000
30,000
1 2 3 4 5
R
e
v
e
n
u
e
i
n
$
YearsTotal Revenue… Total Revenue…
2X $
X
$
After 5 years, a company
with 3% churn rate has
50% higher recurring
revenue than the
company with 6% churn
rate
18. #tatvicwebinar
A GACP and GTMCP company
Churn: Implications on Mobile Apps
Most of the Apps Lose Half of their Peak Users within 3 Months
1500
1,700
1,900
1,720
1,460
1,272
0
200
400
600
800
1000
1200
1400
1600
1800
2000
2200
1 2 3 4 5 6
#
o
f
u
s
e
r
s
Months
Total User
1500
1,700
1,900
2,020
2,116
2,193
0
200
400
600
800
1000
1200
1400
1600
1800
2000
2200
1 2 3 4 5 6
#
o
f
u
s
e
r
s
Months
Total…
Before Churn
Analysis
After Churn Analysis
19. #tatvicwebinar
A GACP and GTMCP company
1500
1,700
1,900
1,720
1,460
1,272
500 500 500 500 500 500
300 300
680
760
688 650
0
200
400
600
800
1000
1200
1400
1600
1800
2000
1 2 3 4 5 6
#
o
f
U
s
e
r
s
Months
Total
User
Before Churn
Analysis
Effect on Total Users : Before and After Churn Analysis
20. #tatvicwebinar
A GACP and GTMCP company
1500
1,700
1,900
1,720
1,460
1,272
500 500 500 500 500 500
300 300
680
760
688 650
0
200
400
600
800
1000
1200
1400
1600
1800
2000
1 2 3 4 5 6
#
o
f
U
s
e
r
s
Months
Total
User
Before Churn
Analysis
1500
1,700
1,900
2,020
2,116
2,193
500 500 500 500 500 500
300 300
380 404 423 439
0
200
400
600
800
1000
1200
1400
1600
1800
2000
2200
2400
1 2 3 4 5 6
#
o
f
U
s
e
r
s
Months
Total
User
After Churn Analysis
Effect on Total Users : Before and After Churn Analysis
21. #tatvicwebinar
A GACP and GTMCP company
How to begin with Churn Analysis?
• Identify the type of App (eg. Gaming/Social/Productivity/Travel)
• Identify the activities that users perform on your app
• Identify the frequency of usage of core features across your user-base
• Identify the average life-time for your user base
22. #tatvicwebinar
A GACP and GTMCP company
Churn Criteria w.r.t. App
• Cut-off date defines the date difference between Current Date and the Date when user last
engaged with your app
• E.g. User A last interacted with app on 9th Feb 2014 and has been inactive ever since
• Cut-off Date for User A is 40 days
23. #tatvicwebinar
A GACP and GTMCP company
Data Points for defining Churn
Other parameters can be:
• # of activities performed since the app was downloaded
• # of times the user visited your app
• Date of App install
• Whether the user interacted with Core Features on a regular basis
• And so on…depending upon the nature of your app.
24. #tatvicwebinar
A GACP and GTMCP company
Overview
What is Churn
Analysis?
Why we are carrying
out Churn Analysis?
What are the Benefits
Of Churn Analysis?
What data will back-
up my churn analysis?
How to extract the
desired data?
Why did we select this
data abstraction
method?
How to build
predictive model?
Why Predictive
Model?
Which machine
learning Algorithm
will fuel our model?
25. #tatvicwebinar
A GACP and GTMCP company
3 Steps prior to Model building
1
• Variable Selection
2
• Data Extraction
3
• Data Preprocessing
26. #tatvicwebinar
A GACP and GTMCP company
Variable(Feature) Selection
Category Variable
Location Country
Device Specific
Operating System
Device category
In App behavior
Count of sessions
Days since last visit
User action
User specific action1
User specific action2
User specific action3
User specific action4
27. #tatvicwebinar
A GACP and GTMCP company
Extracting Google Analytics Data into R
User performing
data extraction
Google OAuth2
Authorization
Server
Google Analytics
API
Access Token Response
Call API for
list of
profiles
Call API for
query
Access Token Request
Image adapted from: Google Analytics Core Reporting API Dev Guide
28. #tatvicwebinar
A GACP and GTMCP company
Overview
What is Churn
Analysis?
Why we are carrying
out Churn Analysis?
What are the Benefits
Of Churn Analysis?
What data will back-
up my churn analysis?
How to extract the
desired data?
Why did we select this
data abstraction
method?
How to build
predictive model?
Why Predictive
Model?
Which machine
learning Algorithm
will fuel our model?
29. #tatvicwebinar
A GACP and GTMCP company
Data Preparation Steps
• Exclude new user who have downloaded the app recently
30. #tatvicwebinar
A GACP and GTMCP company
Overview
What is Churn
Analysis?
Why we are carrying
out Churn Analysis?
What are the Benefits
Of Churn Analysis?
What data will back-
up my churn analysis?
How to extract the
desired data?
Why did we select this
data abstraction
method?
How to build
predictive model?
Why Predictive
Model?
Which machine
learning Algorithm
will fuel our model?
31. #tatvicwebinar
A GACP and GTMCP company
Overview
What is Churn
Analysis?
Why we are carrying
out Churn Analysis?
What are the Benefits
Of Churn Analysis?
What data will back-
up my churn analysis?
How to extract the
desired data?
Why did we select this
data abstraction
method?
How to build
predictive model?
Why Predictive
Model?
Which machine
learning Algorithm
will fuel our model?
32. #tatvicwebinar
A GACP and GTMCP company
Classification Problem
• In our case, we are trying to predict whether user will churn or not
• Based on data points, we are trying to classify between two outcomes
• This is a Classification Problem
Churn
“YES”
1
“NO”
0
33. #tatvicwebinar
A GACP and GTMCP company
Logistic Regression
• Logistic Regression is the technique when you are trying to predict the binary output.
• In our case Predictor(dependent) variable will be unique key(Visitor ID) for each visitors
• Predicted label would be
1 : Visitor will churn
0 : Visitor would not churn
• Using the Logistic Regression algorithm we predict the probability of a user getting Churn
or not
34. #tatvicwebinar
A GACP and GTMCP company
Model building Process
• Split Data Randomly into Train and Test Sets
• Build the Model on the Train Data-set
• Apply the Model on the Test Data-set (un-seen data)
• Calculate the Accuracy Metric for the model on the Test Data
70% 30%
Train Data Set Test Data Set
35. #tatvicwebinar
A GACP and GTMCP company
Model Accuracy
Accuracy = (Number of Correctly Predicted Labels) / Total Number of Labels
= (620 + 1024)/ (620 + 4 + 7 + 1024)
~ 99.34 %
Predicted
Labels
(Predicted by
running Model
on Test data set)
Actual labels (From Test data set)
Churn Not Churn
Churn 620 4
Not Churn 7 1024
Confusion Matrix
37. #tatvicwebinar
A GACP and GTMCP company
User types
Fencers
Tried App just
once in last x
days
Awareness,
Benefit, cost of
not using App
Activators
Tried App more
than once, used
more than one
feature
Increase
Knowledge ,
Motivation
Engaged
Tried App more
than once, more
than 2
features/levels of
app, more at
infrequent time
Learning,
practice,
improving skills
Loyalists
Regular usage of
app with most of
features, shares
app,
recommends,
provide feedback
Practice,
improve,
influence others
Characteristics
Messaging
0.87 0.56 0.45 0.35Churn chance
38. #tatvicwebinar
A GACP and GTMCP company
Overview
What is Churn
Analysis?
Why we are carrying
out Churn Analysis?
What are the Benefits
Of Churn Analysis?
What data will back-
up my churn analysis?
How to extract the
desired data?
Why did we select this
data abstraction
method?
How to build
predictive model?
Why Predictive
Model?
Which machine
learning fuel our
model?
39. #tatvicwebinar
A GACP and GTMCP company
Full Webinar Video
Watch full Webinar Video - http://bit.ly/1ChPrOe
Webinar Video
S: A small introduction about your speakers for the session. I am Shachi Parikh and I am a Technical Analyst at Tatvic. We also have with us Mrugesh Soni, our resident Web Analyst whose expertise lies in fetching actionable insights from data. Hi Mrugesh, shall we begin?
M : Sure, Thank you Shachi and thanks a lot everyone for taking out time for joining us today. Today I am going to spend some time to explain what is churn analysis and its application.
In this webinar we are focusing particularly on churn analysis for mobile apps. This will facilitate the mobile app analysts to target the right audience for retention and thus eventually it will help in boosting their revenues. Okay so lets get rolling.
So let me give you an overview first, which will provide the idea about what we are going to achieve today.
M : This is how the whole agenda has been divided from three different perspectives.
Which are 1) Understanding of churn Analysis 2) data extraction process and 3) Predictive model building
So without any further delay lets get started !
S : Mrugesh, I think it would be helpful for our audience to understand what is Churn?
M : Sure, Let us begin by defining Churn. Customer Churn is phenomena observed across telecom, finance and more recently Saas business models. The term Churn refers customer attrition. It is a possible indicator of customer dissatisfaction, and therefore it is very important for business owners to identify customers who are about to churn. And this is the goal of today’s webinar.
Now, why churn analysis? Then with reference to that - The discussion boils down to the fact that it is highly imperative for business owners to reduce customer churn.
This can be done in two ways.
First #1 Proactively identify customers with a maximum likelihood of churning and apply various Customer Retention Techniques.
This allows you to plug leakages in your customer growth.
But a long term strategy would be to identify the actionable factors leading to customer churn and act on those factors progressively.
Now that we have a handle over the objective and likely solutions, let us have a look at the economic value of the problem.
Though, before I show you actual numbers, here’s an interesting ‘tidbit’ <change word if req>
M : Churn acts as an inhibitor to Revenue Growth, hence if we were able to reduce Churn Rate by half, this has a noticeable impact on your Revenue. Let us map this down in more detail.
S : Interesting fact! Can you elaborate on this maybe with an example?
M : The graph depicted in this slide shows the impact of churn on overall revenue. Assume that there are two businesses, one having a 3 % churn rate while the other has a 6 % churn rate month over month. If we were to compare the revenue trajectories between both these businesses, it is very evident that the company with the lower churn rate would emerge as a winner. But the difference in magnitude of revenue is quite stark. The business with a 3% Churn Rate has a 50% higher recurring revenue than the company with a 6% churn rate. And as the company grows, the effect becomes even more significant. Now since we have a clear perspective of the impact of churn on revenue, let us think about a similar case for a mobile application.
M : It is widely known that since there are a multitude of mobile Apps on app stores, post App launch, if we observe the user growth as a function of time then the growth trajectory is linearly increasing. But after 3 months, the customers begin to churn and leave the app. In fact, a statistic by Mobile App Analytics provider Flurry states that Most of the apps lose half of their peak users by 6 months post launch. Conventionally, a way to counter this is to keep on acquiring new users. If we look at the wholistic picture, it is important not to just acquire new users but to focus on churned users as well.
M : The Green line in both the curves represents the Churned users month over month. In both the cases, the number of proportion of New Users remains constant. Carrying out Churn Analysis and focusing on the number of Churned users has a great impact on the overall customer growth.
M : If you focus on the graph to the left, then lets say that you have 1500 users at the beginning of the first month. You gained 500 new users and lost 300 users then you would be having 1700 users at the 2nd month. Lets say this trend continues then at the end of 6 months you would end up with fewer number of users than you started with. On the other hand, if you were to perform Churn Analysis and stop the user leakage then that would have a higher number of users at the end of 6 months.
S : Mrugesh, correct me if I am wrong, till now we have understood that decreasing the number of Churning users has a substantial impact on Revenue as well as Customer Growth.
M : so we understood that how crucial it is for any mobile app owners to carry out the churn analysis. Lets make an attempt to execute the churn analysis.
M : The Green line in both the curves represents the Churned users month over month. In both the cases, the number of proportion of New Users remains constant. Carrying out Churn Analysis and focusing on the number of Churned users has a great impact on the overall customer growth.
M : If you focus on the graph to the left, then lets say that you have 1500 users at the beginning of the first month. You gained 500 new users and lost 300 users then you would be having 1700 users at the 2nd month. Lets say this trend continues then at the end of 6 months you would end up with fewer number of users than you started with. On the other hand, if you were to perform Churn Analysis and stop the user leakage then that would have a higher number of users at the end of 6 months.
S : Mrugesh, correct me if I am wrong, till now we have understood that decreasing the number of Churning users has a substantial impact on Revenue as well as Customer Growth.
M : so we understood that how crucial it is for any mobile app owners to carry out the churn analysis. Lets make an attempt to execute the churn analysis.
< On hold >
What I was referring to, an app analyst should define the set of criteria, and if any users fall into that bucket he or she will be called as churner.
Now what are these criteria could be : you can first set the cut-off dates – which is difference between current date and last date when user performed last activity on your app
After this slide :
Let revise what we learn so far :
Now before we actually begin to churn analysis, there are 3 steps process, which will help us to build the model.
M : Variable selection is key process in analysis as it creates the base for your model. And here role of web analyst comes into the picture. While selecting variable analyst should keep in mind model accuracy along with the business perspective
There could be a lot a data points which help us in understanding our mobile app users for the purpose of Churn Analysis. On a higher level, these variables could be segmented into User Location, Device Specific features, In App behavior and User Actions
Corresponding to these four categories, we could have Google Analytics dimensions and metrics like Country, OS, Device Category. More
Now once the variable selection process is completed, lets take a brief detour and extract the data from Google analytics using R.
Here is the user who trying to extract data from GA using API and API are just gateways who protects GA’s data.
First of all we head towards authorization, we exchange authorization code with Google Analytics and Google Analytics in-turn gives us an access token. This access token enables us to extract Google Analytics’ data.
In next step we’ll retrieve the list of profiles associated with the account and once we have the list of profiles we’ll select the profile id for which
We wish to have data and post that we’ll select the dimension ad matrix and time frame for which we have the data requirement.
So this is how it works from backend.
Now lets perform this step by step which gives clear idea of data extraction process. For this I need to jump to R.
Now lets perform this step by step which gives you the clear idea of data extraction process. And for that I need to jump on R and lets extract the data.
This is how actually R look like…lets get familiar with it ! left window known as command window where you suppose to write the command and execute it and its result you can see on console window at bottom. On upper right hand side you can see your data file on which you are performing various actions along with list of variables and their data types. And at right corner bottom you can download various packages as per your need besides that you can also track record of at which location in your computer all this files are getting stored.
Now R has the concept of libraries and libraries are add on packages that are used to explore additional functionality. Which is not contain within the base version of R .
So today we are going to use one library which is known as RgoogleAnalytics , which will help us to extract the data from google analytics. And to do that we’ll load that library into R memory.
Now this command will help us to get the authentication, lets run this command and see what happens, as you can see it will redirect to browser and will ask me to accept the account. Post accepting the account we’ll reach to google oath server. Here we need to exchange the access token and for that I’ll click on it. API will exchange and will return the access token, which I’ll copy and paste it on the console window.
Next step is to get the list of the profiles and for that we need to execute this command. And it will give the list of profiles associated with the account.
You can check out the list of profiles from here, now identify the profile for which you want to extract the data and corresponding to each profile there is profile id associated with it.
We’ll use this id in next command which is query building command. (matrix and dimension along with start date and end date)
This is how we build the query. Now the query is ready and by selecting this specific command you can get the your data from you GA profile. You can check how many record you get from the console window…and to see the data click on data at right side…now let say you want to save this data and want to perform some action in terms of transformation or something than you can save this data in csv format by simply writing this command. And that’s it we have extracted the data from google analytics which we’ll be using for our model building.
Those who have any question can feel free to ask, we’ll keep this script on our website after this webinar from which you can download and can run on your system…
And we have completed the data extraction process.
M:
We need to exclude those users who have downloaded the app recently. This is because, those users would show a higher churn rate compared to old users.
Based on app charactetistics that we have take into consideration, 30 days of inactivity period implies user is potential churn.This is not the thumb rule for defining the churn. You can define as per the nature app morever you can also take other parmeters into consideration.
M : We saw until now how we could define the data points for the Predictive Model and how we get these data points into R from Google Analytics. With that in mind, we see
S : Thank you Mrugesh, that was really helpful. I hope the audience enjoyed it as well. We have a couple of questions.
#1 : Can I test the model on the same data-set as the one it was trained on?
#2 : Are there any other algorithms apart from logistic regression?
#3 :