List Price Optimization Using Customized Decision Trees
There are many data mining solutions in the market which cater to solving pricing problems to various sectors in the business industry. The goal of such solutions is not only to give an optimum pricing but also maximize earnings of the customer. This pape
- PDF / 104,468 Bytes
- 7 Pages / 439.37 x 666.142 pts Page_size
- 87 Downloads / 229 Views
Abstract. There are many data mining solutions in the market which cater to solving pricing problems to various sectors in the business industry. The goal of such solutions is not only to give an optimum pricing but also maximize earnings of the customer. This paper illustrates the application of custom data mining algorithms to the problem of list price optimization in B2B. Decision trees used are mostly binary and pick the right order based on impurity measures like Gini/entropy and mean squared error (for CART). In our study we take a novel approach of non-binary decision trees with order of splits being the choice of business and stopping criteria being the impurity measures. We exploit proxies for list price changes as discount %age and SPF discounting. We calculate transaction thresholds, anchor discounts and elasticity determinants for each SKU segment to arrive at recommended list price which gets used by pricing unit. Keywords: Non-binary decision tree List price SPF discount Classification Regression Entropy Log-loss
B2B
Anchor
1 Introduction VMware (VMW) is a virtualization, end user computing and cloud company with annual revenues of USD 6 BB (as of 2014) and a market cap of USD 25 BB [1]. VMware sells products in the Software Defined Data Center (vSphere, VSAN, NSX for computing, storage & network virtualization respectively), end user computing (Airwatch, Horizon, and Fusion/Workstation) and cloud. These are all sold to B2B customers. The prices of products at VMW have been rarely changed over time. However Sales reps could offer discounts via SPF flag which a special discretionary discount was typically given to large orders. The Pricing Business Unit was keen to figure out a way to get to optimal list prices at VMW.
1.1
Objectives
The Advanced Analytics & Data Sciences team came up with the following objectives with the Pricing Business Unit. © Springer International Publishing Switzerland 2016 P. Perner (Ed.): ICDM 2016, LNAI 9728, pp. 438–444, 2016. DOI: 10.1007/978-3-319-41561-1_33
List Price Optimization Using Customized Decision Trees
439
– Analyze historical prices and related volume movements and understand levers of discount % and SPF (sales person specific discount) Flag – To come up with recommended list price along with level of confidence for all SKUs
2 Solution Framework 2.1
Strategy
A traditional list price optimization would use price changes versus quantity changes, but we never change prices. Discounting practices are an alternative inference method but requires additional steps. Discount% and SPF requests are useful indicators to infer the customer’s assignment of value to a product. Segments will be arrived at for discount % and SPF flag. SPF flag is a discount that can be given by a sales representative on request – mostly given on large order sizes. List price is used to measure VMware’s assignment of value to a product. Imbalance between assignments of value then indicate tension in List Price. Following detailed steps were planned – Understand segment
Data Loading...
