QNT 5485
Assignment “Lost Sales” (15 points)
(based on the case “Lost Sales” by Marlene Smith, JMP Statistical Discovery, 2017)
In many industries throughout the world, suppliers compete for business by submitting quotes for work, services or products. A key criterion used to determine the winning quote is the dollar amount of the quote, but other factors include expected quality, estimated delivery time of the product, or quoted completion time of the work.
The focus of this case is a supplier of equipment to the automotive industry. The products of interest in this case are various precision metal components used in a range of automotive applications, such as braking systems, drive trains, and engines. Some of the products will be used in the manufacture or assembly of new automobiles (i.e. original equipment), while others will be used as replacement parts in automobiles already on the road (i.e. aftermarket).
Task: The supplier wants to increase sales and expand its market position. Many of the quotes provided to prospective customers in the past haven’t resulted in orders. Do the data provide any indication why? Are there certain situations that make it more or less likely that a customer will place an order?
The data set contains 550 records for quotes provided over a six month period. The variables in the dataset are:
- Quote = the quoted price, in dollars, for the order
- Time to Delivery = the quoted number of calendar days within which the order is to be delivered
- Part Type = OE = original equipment; AM = aftermarket
- Status = whether the quote resulted in a subsequent order within 30 days of receiving the quote: Lost = the order was not placed; Won = the order was placed.
Analyze the current stats of the company’s ability to win orders and determine the drivers of the company sales by estimating classification models, including Classification (Decision) Tree, Neural Network, K Nearest Neighbors, Naïve Bayes, Random (Bootstrap) Forest and Boosted Tree.
- What are the predictors of the order status? Summarize the impact of each predictor variable.
- Are there certain situations that make it more or less likely that a customer will place an order? Each model has section Profiler which allows you to obtain probabilities of losing/ winning order and the predicted classification by moving the thresholds (red dotted lines) shown on Profiler’s graphs. What probability of losing an order is predicted by the classification tree, neural network, and random forest for two orders:
(1) the order of a part for original equipment to be delivered in about 10 days (i.e. 2 business weeks or week and a half of calendar week);
(2) the order of a part for the aftermarket part to be delivered in about 45 days (months and a half). You might use the Profiler option to assess the probabilities.
- Analyze the misclassification rates for these models higher, lower, or about the same? Given the significance of the terms in the model, is this what you’d expect?
- Based on misclassification rates, which model does the best job of predicting Status?
- How could these models be improved? For example, are there any variables missing from this data set that might improve the model’s predictive ability (i.e., lower the misclassification rate)?
- Prepare a business report summarizing your findings in previous questions and recommending the supplier on how to increase sales and expand its market position.
