Our old friends, the BehaviorBot and Jungermeister, are in two-period competition. In the first period all is as before: The BB has an accuracy of .95, and it costs BB $10 to process a case. The JM has an accuracy of .99, the buyer additionally incurs a labor cost of $10 to input the data, and it costs JM $85 to process a case. Mistaken diagnoses cost the buyer $3000 each. In this situation, JM has a competitive advantage of $35/case (as calculated in video 1-13). In the second period, each firm's accuracy improves based on the number of cases it processes (getting feedback on outcomes). Let x = the number of Period 1 cases a firm processed. Then suppose a firm i's accuracy in Period 2, pi2 is equal to its Period 1 accuracy, pi1, plus one-half of x/(1 x) times the gap between pi1 and perfect accuracy (perfection would be a probability of one): Pi2 = pi1 (1/2)*(x/(1-x))*(1- pi1). The 1/2 is there to limit the rate of learning in a market of reasonable sizethe fastest a firm could improve its accuracy is halfway to perfection (if it had an infinite number of cases). Now note that the