>
I am currently working as a Business Analyst at a major telecom company (let’s call it MTC). This “Business Analyst” title can be expanded as Statistician/SAS Programmer/Predictive Modeler and can be expressed as “Data Miner” in general. There are about 15 members (including one director and 4 managers) in our group named Customer Analytics. At MTC, we are known as "Modelers", and belong to the Marketing branch of the organization hierarchy.
In a nutshell, we do data mining, statistical modeling and clustering & segmentation which enables us to make business predictions that can help the marketing group to take business decisions.
Let me briefly talk about the need for Business Analytics in a telecom company.When a telecom company acquires a new customer, they have to spend money. This spending includes everything they have to do for getting a new customer: advertising, giving subsidies on handsets, additional network capacity to accommodate the new customer calls, more care center representatives needed to handle calls to the company etc. It takes months before MTC can recover from the acquisition spending for each new customer and start making money from that account. If a customer leaves the company after, let’s say two months, the company actually loses money. (Even the early contract termination fees can not balance the loss in revenue.)
Moreover, it's cheaper and easier to retain an old customer than to acquire a new customer.
Thus, while trying to attract new customers, telecom companies are always more concerned with retaining the existing customer base.
Here’s where the role of our group comes into the picture. MTC needs to know which customers are 'more likely to leave', and when they are likely to do so. If we knew that, then the Marketing group can take necessary actions to retain those customers. To find out whom to contact, and when to contact them they depend on us. (On our predictions based on statistical models, to be more specific.)
Our job is to study the customer history (cell usage, call details, demographics, late payments, purchase patterns etc.) and look for behavioral patterns that can explain what triggered previous MTC customers to leave. Based on the historical data, we build Statistical (predictive) models using SAS. These models can be applied to the current customer base (whose behavior is known to us) to decide which customers are likely to leave (in, let’s say, next two or three months).
After the very interesting phase of data discovery, modeling and prediction, comes more challenging task. We have to present our model to the Marketing managers (decision makers) and explain them what’s going on with our customers. Obviously, we have to present our findings in such a way that they are easier to digest for a non-technical marketing folks. The marketers will take this result and run various campaigns (offers, incentives) to retain those customers. In doing so, they will be spending millions of dollars, so our results must be: (a) as accurate as possible, (b) meaningful, and (c) actionable.
This Churn/Risk Modeling is one among the many other things we do. Here are some scenarios in which the seek help from our group:
− The company launches new products and they need to know which customers they should target. So we will build a model that predicts which customers are more likely to respond to that specific product related offer. (MTC has millions of customers, so they can’t go and target all customers. We need to find out which customers will be most interested and target them to increase our ROI.)
− The company wants to identify which customers are more valuable (profitable). To get the holistic view of a customer; in addition to look at “what’s the customer worth as of today” it’s necessary to look into the future and see “what’s the potential value of the customer” too. Using various statistical modeling techniques, we can predict (a) future tenure, and (b) future revenue stream of a customer. (Interestingly, when you call the customer service, your “value” decides which call center will handle your call. A call from a low value customer might get transferred to international call centers, while a high value customer will get to talk with a US based call center. Also, if you decide to terminate your contract, your “value” can heavily restrict how much you can negotiate on contract termination fees.)
The use of data mining and statistical modeling is, of course, not limited to telecom industry. Think about an insurance company that wants to identify which customers are likely to be frauds, a bank or credit company that needs to know who is likely to close the account or which transactions are likely to be fraudulent, a pharmaceutical company that wants to identify which drug component is likely to affect as a remedy on a decease while developing a new drug, a giant retailer that needs recommendations about which products they should keep together in the store (which products do customer buy together?), or an online retailer (like Amazon) that gets help from data mining to come up with recommendations for next purchase.
