This paper proposes an approach to assess the ability of an Enterprise to efficiently predict churn (develop a systematic churn prediction model). It attempts to cut through the hype that surrounds ‘churn prediction modeling’.
Churn in the highly competitive telecoms environment has been a major pain for Enterprises, steadily decreasing their profitability. Especially for telecom Enterprises with a high & inelastic installed capacity (in terms of network, customer service & sales infrastructures), the customer base reduction can have a significant profitability erosion effect.
Due to the high demand for an efficient churn management framework (tools, customer retention initiatives & relevant processes), many relevant services are offered by the market. Certain Consultants claim that they can offer an ‘out of the box’ churn prediction model which is immediately applicable to the specific Enterprise. This very simplistic approach could easily mislead an Enterprise with no prior churn prediction initiatives.
Certain cases believe that they fully understand the key churn drivers, therefore churn prediction is meaningless. Others believe that the success of a churn prediction modeling initiative, depends solely on the successful implementation of a data mining project. The Enterprise has to gather all available data and this will suffice.
Based on historical telecom market data (source: relevant TM Forum papers), customers leave their Service Providers because of the following main reasons (the sequence is indicative of the ‘weight’ of each factor towards this decision): 1. Price 2. Billing Dispute 3. Lack of Network geographic Coverage 4. Lack of Value Added Services/features 5. Lack of carrier responsiveness 6. Brand Image 7. Numerous tariff options available to customers offered by competitors. 8. End of Service Contract 9. New competitors & offerings in the market
In order to assess the readiness of an Enterprise to understand the satisfaction level of a specific customer (vis-à-vis), One needs to assess the degree to which customer satisfaction levels are reflected in the customer’s behavior which is systematically captured by the Enterprise. In other words, does the Enterprise capture information in which the customer satisfaction level vis-à-vis the above factors, can be traced.
Price satisfaction
This is the top churn factor, according to all relevant studies. The ‘weight’ (or percentage contribution) of this factor in the churn decision, is expected to be high. However the exact weight can only be derived from a customer survey (voice of the customer). A proxy for price satisfaction is the ‘distance from optimum price plan’ (internally compared to other Enterprise price plans or against the competition). The smaller the distance, the higher the expected customer satisfaction level or the ability of a competitor to convince on its cost leadership. However, it is not easy to estimate this ‘distance’. In order to do this, a telecom enterprise needs to gather usage data (CDR’s) of a specific customer and reprice them against all available & eligible (internal & competitor) price plans (a pricing analytics platform is needed). Often this capability is not available.
Billing Dispute
The CRM system ideally captures all customer interactions and billing disputes. The outcome of the dispute needs to be evaluated and captured as well. The perceived level of customer satisfaction or frustration from the settlement should be assessed & captured in the customer history.
Lack of network geographic Coverage - Generally such a shortcoming is known internally by the Enterprise. A factor not easily reflected in captured customer behavior (a mobile Customer would not file a complaint on that, since the coverage is not guaranteed). On the order handling process, the lack of geographic coverage leading to failed sales could be captured, though not easily. I.e. an anonymous search query on the product web site on ADSL availability would not be captured or if captured (as a request) it can hardly be associated with a specific Customer. (though this sales process failure is not directly associated with churn).
Lack of Value Added Services/features - Generally known by marketing & competitive intelligence. It could be derived from a VOC (voice of the customer) survey on churners & by database profiling.
Lack of carrier responsiveness
Responsiveness can be measured. The time lapse from a complaint submission to the reply can be measured. The time lapse from a fault submission to its resolution can be measured. Overall carrier responsiveness can be evaluated per customer based on that history.
The following additional factors: • Brand Image • Numerous tariff options available to customers offered by competitors. • New competitors & offerings in the market, are identified as potential churn factors. However these factors are common to the Customer base of an Operator, thus they do not differentiate per Customer and cannot be used for individual Customer churn prediction. On the other hand, if a competitor’s offering relates to a specific product/service, then Customers subscribed to this service may be more affected.
End of Service Contract
End of Service Contract date should be monitored, since the time period near this date is a period in which the customer may be evaluating churn options. Tenure band analysis can be used to analyze high risk periods.
Posted: February 2008 C: Business Intelligence Posted by: K. Panayotakis
[to be continued]