AnalyticBridge
Data Intelligence, Business Analytics
How can a large company detect the root cause for a sudden increase in churn, requested credits or charge backs?
Here are two ideas:
- Survey your customers, including those who left or who are causing problems.
- Automatically analyze / summarize what users post on Twitter and other social networks about your company. Look for shifts in what users are posting now, vs. 3 months ago.This can be performed with very inexpensive / open source tools. We do it with Perl scripts, web crawling and advanced data mining.
Views: 34
Replies to This Discussion
-
Permalink Reply by Adam Hood on -
We don't do much in the way of survey's, however we've leveraged SAS to pull in various rss data to detect differences. Additionally, we predefine segmentation and monitor changes in segments or migration between segments. Sudden changes in a given segment can tell you where to focus your search for answers.
-
Permalink Reply by Minethedata on -
AS suggested what difference do you foresee between the segments.Can you elaborate on this more?
-
Permalink Reply by Minethedata on
-
This solution is by doing some modeling on the internal data. Carry out logistic modeling on the data belonging to the period in which the increase in churn is seen. The logistic coeffcients will show which variable is the most contributing factor and can be indicative of the root cause of the churn.
-
Permalink Reply by Tom Wolfer on -
Once an analysis has identified a churn issue: whether this be through a time-series analysis or another method, you may not have to survey customers at all to learn the root causes. If you have a sales force, consult with them to learn what customers/prospects are saying in the field in order to learn what is happening in the marketplace. This may provide some guidance as to reasons behind the churn issue: upcoming competitors, pricing issues, product/service quality problems etc. This is an inexpensive way to validate your model's findings, and, it may justify the costs involved in surveying customers more formally. Make sense?
-
Permalink Reply by Ralph Winters on -
Agree with Tom. Once a problem has been identified, and you are looking for a root cause, you need to get out there and brainstorm within/ and external to company to ascertain what is going on. There comes a time when further data modeling will not help. You can certainly use things like fishbone diagrams, but the important factor is causation, and not necessarily association, so there needs to be a lot of why's . Of course you do need good stated facts as evidence.
-Ralph Winters
-
Permalink Reply by Tom Wolfer on
-
Ralph, what is a fish bone diagram? I am not familiar with this term. Maybe I have seen one before, however, can you send me an image? Thanks
-
Permalink Reply by Ralph Winters on
-
Also known as Ishikawa diagram. The first one answers the question of why it's called that
http://www.google.com/images?q=fishbone+diagram
-
Permalink Reply by Tomas Keller (formerly Ohlson) on -
How many tweets /month do you need to implement idea 2? Any rule of thumb?
/ Tomas
-
Permalink Reply by Vincent Granville on -
Not just tweets, but any RSS or blog data. How many... depends on how striking the difference between today tweets and tweets that are 2-month old. If people have shifted their mind 180 degrees, you should be able to draw conclusions from 20 tweets or less.
If the tweet volume is really high, you should break it down into segments, and check what people say within each segment.
-
Permalink Reply by BR Deshpande on
-
A related CRM question: what types of modeling techniques are commonly used to *predict* churn as opposed to explaining root causes post facto? Does anyone have benchmarks about such cases? Thanks.
-
Permalink Reply by Tom Wolfer on
-
Hi BR, if you define your target 'churn' variable correctly and are using historical data, then you can use logistic regression, an inductive decision tree, regression to name a few.
-
Permalink Reply by BR Deshpande on
-
Hi Tom,
Appreciate the response. Thanks.
Are you (or anyone) aware of the performance of such models? We have used several models/techniques, but unfortunately, the speed of business limits us from really quantifying the ROI and helping us identify the best fits or performers. I am interested in learning if there are documented ROI results.
Top Content
Follow us
© 2013 AnalyticBridge.com is a subsidiary and dedicated channel of Data Science Central LLC
