Controlling ISP Churn: Borrowing from the Telco Model - Industry Trend or Event

Telecommunications, Nov, 2000 by Dana Kreitter

Customer retention is one of the largest concerns among service providers. Creating a network metering infrastructure can help.

Some of the factors behind ISP customer churn include undifferentiated service offerings that make switching services easy. For example, in a flat-rate, one-size-fits-all service environment, some customers would likely be much better served by a specifically tailored service bundle. But ISPs typically do not know how their customers are using their services or how they might be segmented. This is, in part, a consequence of the distributed nature of the IP network itself.

In the telephony world, service providers are able to extract CDR (call detail records) from switches to learn about network usage. This data was initially used for time-and-distance billing, but later for defining new services and pricing plans, such as calling circles. With the advent of Internet services, however, the task of data collection has become much more challenging, and it is only over the past year or two that major ISPs have started tackling the problem. The new IP mediation infrastructure enables the same business functions previously found in the public network: usage-based billing, usage analysis and improved operations management. Churn prediction is one application of usage analysis.

Service providers have been performing churn analysis for years. Specialized churn prediction application providers have built up data mining and modeling techniques of varying sophistication. ISPs can borrow from these models, however, a number of significant differences exist:

* Vast diversity of services. Internet services are numerous and it takes much more than sophisticated metering capabilities to derive meaningful usage information. A single session might entail a host of communication options, e.g., e-mail, basic Web surfing, VPNs and video streaming.

* Distributed nature of user data. In contrast to telephony CDR collection, usage data in an IP network must be extracted from a number of metering points. Data sets from various sources must be correlated before meaningful usage records result.

* Scale. The volume of data to be processed, particularly at the network level, is far greater than for CDRs. IP networks require high-performance collection systems, with highly flexible aggregation and correlation capabilities to make raw data more manageable.

* Real-time data processing. Traditional CDRs can be processed relatively infrequently--often hourly or a few times per day. However, due to the high volume and the dynamic nature of IP data, ISPs need to operate much more quickly. Real-time or near-real-time data collection is often necessary for many applications.

* Churn prediction models. Even with the data collection infrastructure in place, the challenge of defining effective churn prediction models for IP services remains. Internet connection time does not necessarily imply active service usage. Some leverage from existing wireless service churn models exists, but service providers and business intelligence application vendors need to take a fresh look at the range of relevant prediction variables.

While ISPs can borrow some techniques from the telco model, these significant differentiators require new usage collection solutions and potentially more sophisticated behavior models to evaluate customer traffic and turnover.

Churn Management

Churn management is a hot topic for ISPs and traditional telcos that are adding IP services. However, IP usage metering has just begun to pick up steam over the past 18 months. Until this point, IP service providers have typically been able to track only aggregate traffic without the ability to segment customers and individual services. IP network metering is the key enabler to getting the fundamental information for churn prediction.

The available network information includes upstream and downstream data volumes per subscriber, services accessed, connection time, disk space consumed and more. These metrics can be combined with other user data to develop prediction models (see Figure 1). Rated usage records must also be considered when determining which potential churners are of greatest value.

Once a network metering infrastructure is in place, service providers can analyze data records and segment users. They can create tailored service bundles and institute loyalty programs. The more services they can sell to a particular subscriber, the "stickier" those services are and the less likely it becomes that the subscriber will stray.

Network monitoring data also creates opportunities to cross-sell or up-sell specific goods and services. ISPs can offer discounts on future offerings. The airline industry, with its frequent flier programs, epitomizes the value of service usage data. Given the immense volume of Internet usage and the transaction value of Web commerce and advertising, it seems preposterous that even rudimentary subscriber usage tiers are unknown. As the Internet wars shift from customer acquisition to customer retention and brand loyalty, evaluation of usage data will become routine.