Know your customers by knowing who they know #paw

Copyright © 2009 James Taylor. Visit the original article at Know your customers by knowing who they know #paw.

Tim Manns presented on Optus SingTel’s work using social interactions in analytics. In this case the social network is the people you know based on who you call, and who you don’t call. Optus is part of SingTel and is a telco based in Australia that competes with the main telco and has about 35% of the market: cell phones, landlines, business, etc. Optus uses data mining in lots of ways: targeting communications, finding behavioral drivers, finding out what they want – understanding the customer.

100% of the data they use is, unusually, in a data warehouse from Teradata. This contains the Call Detail Records which amounts to 100 Million+ calls a day. The EDW contains information from usage, downloads, billing, point of sale information, contracts, customer information and more. This means all the data mining is done against the data warehouse rather than against raw data files extracted from transactional systems. An illustration of the power of keeping detailed records in the data warehouse rather than reporting-oriented summaries.

Churn and customer value are . …

Copyright © 2009 James Taylor. Visit the original article at Know your customers by knowing who they know #paw.

Tim Manns presented on Optus SingTel’s work using social interactions in analytics. In this case the social network is the people you know based on who you call, and who you don’t call. Optus is part of SingTel and is a telco based in Australia that competes with the main telco and has about 35% of the market: cell phones, landlines, business, etc. Optus uses data mining in lots of ways: targeting communications, finding behavioral drivers, finding out what they want – understanding the customer.

100% of the data they use is, unusually, in a data warehouse from Teradata. This contains the Call Detail Records which amounts to 100 Million+ calls a day. The EDW contains information from usage, downloads, billing, point of sale information, contracts, customer information and more. This means all the data mining is done against the data warehouse rather than against raw data files extracted from transactional systems. An illustration of the power of keeping detailed records in the data warehouse rather than reporting-oriented summaries.

Churn and customer value are critical to telcos. If a customer spend $50/month and you have 5M customers, then 0.5% churn is $1.25M/mo every month from then on. Similarly just $5 more to just 10% of customers is worth $2.5M/mo. Recent analysis has been focused on finding the social groups – identifying friends and family and measuring word of mouth influence for instance. Who you call, when you call them, how long you talk and more help identify this social group. These groups help Optus understand customers, see the impact of refer-a-friend promotions, and estimate the value of someone outside the network based on the people who call them. This kind of analysis really helps identify the impact of word of mouth – bad service stories, good service stories, etc. – because the referrals from friends can be much more influential than any corporate advertising.

Optus runs a data mining job inside their data warehouse that processes 600M records from the last four weeks. Numbers have to be cleaned to a common format and queries are run to summarize the customer calling relationship – outbound calls to other, inbound calls to other and then join them to identify the reciprocal relationships. Each relationship shows how many messages, picture messages and calls have gone back and forward between the two people.

It kept it simple to do things like identify someone’s spouse, work colleague, weekend buddy not on Optus, etc., seeing who has a relationship, for instance, to a non-Optus customer that can be a driver for a refer-a-friend program. Other insights included things like identifying family members based on calls when only they knew, previously, that they had the same address. It also found kids without separate accounts, because dozens of SMS from a phone all went to known teenagers, for instance. This last is a great example – the teenager is much more influential in terms of who they talk to.

Key takeways:

• Purchaser is not necessarily your customer
• Your cheerleaders/leaders are critical and can wreck expensive marketing campaigns
• Customer churn is worse than you think, as churning customers tell their friends and people can leave as a group, taking prospects with them also

Or, as he said, save the cheerleader, save the world. For instance, customers in social groups with recently churned customers are 4-5x more likely to churn. Similarly, prospects in a social group with someone who joins the network are 4-5x more likely to join the network! This kind of analysis allows you to be proactive with the friends of someone who leaves, for instance.

Their churn models predict churn in subsequent month using social groups, usage, billing, contract data and demographics. The model is excellent, predicting at 10x the random change where a perfect model would predict at 20x. They more or less doubled their accuracy in proactive churn prevention campaigns, for instance.

Link to original post