A two state Markov chain model is used to describe and forecast the over time behavior of the best retail customers at Merrill Lynch. This model has 4 behaviorally meaningful parameters which capture the effect of recently being a prime customer, the differing average commissions generated across customers and the exiting of some of these customers from the Merrill Lynch system. This model helps management to understand the dynamics of the prime customers' behavior. In particular, the forecasts generated by the model allow for better analyses of possible strategies for providing special services for these very good customers. The model which was developed with 1976–1979 data is validated against the actual 1980 behavior of Merrill Lynch customers.