city is often needed.

Specifically, usage information should be extracted from the operational systems into a datawarehouse, which shall store the historical information on usage and periodically produce F and M scores for the Customer base and sort the Customers in F and M quintiles. Moreover, it can perform time-series analysis on FM scores. Finally, in order to produce an F or M score for any Banking or Telecom Customer, the Business has to aggregate Customer behaviour from all accounts this Customer holds. Prerequisite for this, is to have a Customer-centric data structure. If the Business operates on unlinked accounts of the same real Cus

Frequency (F) and Monetary (M) analysis, form together with Recency (R) the framework of RFM analysis. Though recency is the strongest predictor of future behavior, frequency and monetary analysis act in a complementary mode (to recency), to create a complete picture of the Customer behavior. There are many cases in which Recency analysis not coupled by Frequency & Monetary analysis, can give a misleading picture. For example, a new subscriber is very recent, but appears to have a low monetary value because she started using the service recently. The Business cannot tell, whether this new Customer will be profitable (will use the service a lot). On the other hand, an old Customer can be ranked low in R but high in F and/or M. This is probably a valuable Customer who is late in interacting with the Business. This is probably one of the candidate cases for a retention plan. In general F is a stronger predictor to M. If the service usage is producing a relatively stable monetary value, then M does not yield substantial additional insight to F. However there are exceptions. Should a Business try to offer an expensive service, the M predictor has an increased power. Those that spend a lot on a service are more likely to spend on an expensive additional service, than those who spend less. Subscription based services involve the continuous usage by a Customer, often based on a contract.

Common examples are: bank accounts, credit cards, fixed & mobile telecommunication services, internet access service. In the case of subscription based services, the frequency of use can be easily derived, since each service usage is recorded. In the case of bank accounts or credit cards, each account or card transaction is recorded. These transactions are handled by banking systems in order to produce monthly balances or Customer invoices. In the case of fixed & mobile telecommunication services, each phone call is captured in a CDR record (call detail record) along with details about the call (calling & called number, time of day, duration, cost band). These usage records are processed by the billing systems, in order to produce Customer invoices. Therefore the banking and telecoms industries, capture rich information on service usage in order to bill Customers. Thus the information for the execution of an F and/or M analysis, is there. These service usage transactions are executed by millions of Customers very frequently and may be producing a huge number of service usage records on a monthly basis. Therefore, in order to execute F and M in these industries, substantial computational & storage capacity is often needed.

Specifically, usage information should be extracted from the operational systems into a datawarehouse, which shall store the historical information on usage and periodically produce F and M scores for the Customer base and sort the Customers in F and M quintiles. Moreover, it can perform time-series analysis on FM scores. Finally, in order to produce an F or M score for any Banking or Telecom Customer, the Business has to aggregate Customer behaviour from all accounts this Customer holds. Prerequisite for this, is to have a Customer-centric data structure. If the Business operates on unlinked accounts of the same real Cus

the other hand, an old Customer can be ranked low in R but high in F and/or M. This is probably a valuable Customer who is late in interacting with the Business. This is probably one of the candidate cases for a retention plan. In general F is a stronger predictor to M. If the service usage is producing a relatively stable monetary value, then M does not yield substantial additional insight to F. However there are exceptions. Should a Business try to offer an expensive service, the M predictor has an increased power. Those that spend a lot on a service are more likely to spend on an expensive additional service, than those who spend less. Subscription based services involve the continuous usage by a Customer, often based on a contract.

Common examples are: bank accounts, credit cards, fixed & mobile telecommunication services, internet access service. In the case of subscription based services, the frequency of use can be easily derived, since each service usage is recorded. In the case of bank accounts or credit cards, each account or card transaction is recorded. These transactions are handled by banking systems in order to produce monthly balances or Customer invoices. In the case of fixed & mobile telecommunication services, each phone call is captured in a CDR record (call detail record) along with details about the call (calling & called number, time of day, duration, cost band). These usage records are processed by the billing systems, in order to produce Customer invoices. Therefore the banking and telecoms industries, capture rich information on service usage in order to bill Customers. Thus the information for the execution of an F and/or M analysis, is there. These service usage transactions are executed by millions of Customers very frequently and may be producing a huge number of service usage records on a monthly basis. Therefore, in order to execute F and M in these industries, substantial computational & storage capacity is often needed.

Specifically, usage information should be extracted from the operational systems into a datawarehouse, which shall store the historical information on usage and periodically produce F and M scores for the Customer base and sort the Customers in F and M quintiles. Moreover, it can perform time-series analysis on FM scores. Finally, in order to produce an F or M score for any Banking or Telecom Customer, the Business has to aggregate Customer behaviour from all accounts this Customer holds. Prerequisite for this, is to have a Customer-centric data structure. If the Business operates on unlinked accounts of the same real Cus

ion based services involve the continuous usage by a Customer, often based on a contract.

Common examples are: bank accounts, credit cards, fixed & mobile telecommunication services, internet access service. In the case of subscription based services, the frequency of use can be easily derived, since each service usage is recorded. In the case of bank accounts or credit cards, each account or card transaction is recorded. These transactions are handled by banking systems in order to produce monthly balances or Customer invoices. In the case of fixed & mobile telecommunication services, each phone call is captured in a CDR record (call detail record) along with details about the call (calling & called number, time of day, duration, cost band). These usage records are processed by the billing systems, in order to produce Customer invoices. Therefore the banking and telecoms industries, capture rich information on service usage in order to bill Customers. Thus the information for the execution of an F and/or M analysis, is there. These service usage transactions are executed by millions of Customers very frequently and may be producing a huge number of service usage records on a monthly basis. Therefore, in order to execute F and M in these industries, substantial computational & storage capacity is often needed.

Specifically, usage information should be extracted from the operational systems into a datawarehouse, which shall store the historical information on usage and periodically produce F and M scores for the Customer base and sort the Customers in F and M quintiles. Moreover, it can perform time-series analysis on FM scores. Finally, in order to produce an F or M score for any Banking or Telecom Customer, the Business has to aggregate Customer behaviour from all accounts this Customer holds. Prerequisite for this, is to have a Customer-centric data structure. If the Business operates on unlinked accounts of the same real Cus

cord) along with details about the call (calling & called number, time of day, duration, cost band). These usage records are processed by the billing systems, in order to produce Customer invoices. Therefore the banking and telecoms industries, capture rich information on service usage in order to bill Customers. Thus the information for the execution of an F and/or M analysis, is there. These service usage transactions are executed by millions of Customers very frequently and may be producing a huge number of service usage records on a monthly basis. Therefore, in order to execute F and M in these industries, substantial computational & storage capacity is often needed.

Specifically, usage information should be extracted from the operational systems into a datawarehouse, which shall store the historical information on usage and periodically produce F and M scores for the Customer base and sort the Customers in F and M quintiles. Moreover, it can perform time-series analysis on FM scores. Finally, in order to produce an F or M score for any Banking or Telecom Customer, the Business has to aggregate Customer behaviour from all accounts this Customer holds. Prerequisite for this, is to have a Customer-centric data structure. If the Business operates on unlinked accounts of the same real Cus

city is often needed.

Specifically, usage information should be extracted from the operational systems into a datawarehouse, which shall store the historical information on usage and periodically produce F and M scores for the Customer base and sort the Customers in F and M quintiles. Moreover, it can perform time-series analysis on FM scores. Finally, in order to produce an F or M score for any Banking or Telecom Customer, the Business has to aggregate Customer behaviour from all accounts this Customer holds. Prerequisite for this, is to have a Customer-centric data structure. If the Business operates on unlinked accounts of the same real Customer, then it may be unable to carry out an accurate F or M analysis. Copyright 2006 – Kostis Panayotakis