The data come from a firm that sells DSL services, to both residential and business customers. The firm operates services in the United States and globally. Some initial subscription required a one year contract, but customers can terminate the service at any time, with a penalty if the contract is terminated prematurely.
The technical question include, software or hardware related issues, question regarding installation, dial up, user identifications, or passward, and downed services or network outages. Transactional questions include inquiries about billing, emailing accounts, product news, product services, and registration.
According to the company measures the service duration as the total time of service encounter from the time the agent picks up the phone to the time the problem is solved. This measure includes time to speak to the customer, as time speaking to the customer is on hold and the agent is processing the customer request.
Research methods used
The calibration sample used contains the service history of 9643 calls. the holdout sample contains 1053 customers, who made atonal of 4661calls.they have access to detailed information about each call, such the callers location ,time stamps, call reason, service allocation, call center agent, call center manager, and total service duration.
Data analysis methods
They assume that each time a customer x of type q decides whether to stay with the firm, they follow extant marketing literature and customer satisfactory.
Intuitively, service duration could be determined by the trait of service center, question type, and customers.
Heterogeneity and estimation
They use maximum likelihood to estimate alpha and beta which parameterize the effect of service allocation decisions on duration and customer retention. To account for customer heterogeneity they adopted a latent class approach, such that segment membership depends with computers and type of customer.
To improve customer experience and use offshore service center more effectively ,a firm might match each service call with the right center according to individual customer preference .therefore they formulated the service allocation decision of the firm as a stochastic dynamic programming problem.
To solve the dynamic program problem, they undertook two iterative steps; first, the firm continuously learns each individual customer by analyzing customer information according to the customer's revealed reactions to the firm's most recent interaction. the second ,it adapts its decisions according to its most recent knowledge about each customer, during these integrated and iterative processes, updated knowledge continuously adjusts the firms decisions, and the resulting customer reactions again inform the learning process. Thus learning and decision making are integrated, which they refer to as adaptive learning. Adaptive learning enables the firm to follow customers and improve the accuracy of its knowledge about each individual customer. Because the state variables are continuous, they face the problem of large state space and therefore adopt the interpolation method that Kean and Wolpin (1994).developed to calculate the value functions.
Adaptive Learning of Customer Heterogeneous Preference
When customer i calls at time t, his or her preference remains unknown or uncertain to the firm. To acknowledge that companies usually conduct segmentation analysis using demographic variables and know the average probabilities of a customer belonging to a segment, we define the prior belief of customer type Pri0(m) as the probability of segment membership, resulting from the latent class approach in the estimation. In addition to static demographic variables, accrued information obtained by observing customer feedback about the firm's most recent interventions might reveal customer information.
There are at least two such information sources: observed prior service duration and observed customer retention. The same customer usually shows a consistent pattern over in terms of the length of the service duration. For example, retired customers have more time to talk on the telephone and likely incur longer service durations. Customer retention observations reveal the customers' reactions to service allocations. For example, if a customer leaves because he or she is serviced by an offshore center, this implies that the customer is sensitive to offshore centers.
The role of call centers has shifted from a cost to be saved to proffered and prevalent channel to handle integrated marketing fuctions, which makes it an increasingly important corporate strategic asset.reserch could develop more sophisticated learning routines to allow for dynamic changes in customer preference. Using panel data pertaining to service allocation, they provide empirical evidence about how the firms service allocation decisions affects service duration and customer retention. The offshore studies have comparative advantages over onshore centers when it comes to technical questions.