HPDirect case study - Grade: A PDF

Preview

Summary

test...

Description

Hewlett Packard uses Operations Research for profitable growth of HPDirect.com Background: Hewlett Packard (HP) is one of the largest information technologies, software services, Infrastructure and solutions provider. HP was facing stiff competition from Dell during the starting days of online consumer sales business with its launch of HPDirect.com. HP focused different areas to improve the processes with different kinds of analytical solutions to sustain in online consumer business. Below are some of the key focus areas company choose to develop data driven solutions.  Increase traffic (i.e., attract a greater number of visitors to the online portal) to HPDirect.com product pages.  Convert visitors to customers with a suitable targeting strategy.  Identify potential customers who likely to buy the products  Estimate the demand of the products based on customers purchasing history and identify potential customers who likely to buy products.  Develop a strategy for third-party warehouses to generated demand and deliver the products on time.  Determine best marketing channel to reach the customers  Allocation of budgets marketing channels and product categories by using optimization

Solutions: The data scientists in HP Global Analytics (GA) addressed this need with three different solutions such as Solution A, Solution B and Solution C. Solution A helps marketing teams to allocate the budget to various marketing channels and enables HPdirect.com team to identify the key drivers of online traffic and quantifies their relative impact Solution B helps marketing teams to identify which customers are most likely to purchase and when, technology needs of customers and enable the teams to target specific customers with messaging channels (ex: email, printed mail). Solution C helps to increase the efficiency of warehouse operations by improving the customer order forecasts.

Solution-A Implementation: Solution A helps the marketing team to replace legacy approach of based on simple moving average forecast and it consists of demand generation models for traffic of product subcategory web pages and optimizing return on investment. SAS Enterprise Guide, Oracle’s Crystal Ball, and JMP are used to develop time-series and forecast

models. The diagram depicts the steps performed as part of solution A implementation. Visualization and Data codification: Visualization helped the HP team to identify seasonality and trends based on historical data and they used data codification to relate trends and seasonality to specific marketing activities and sales seasons over the period of two years. Time series Modeling uses ARIMA (autoregressive integrated moving average) models to identify the main components of web traffic such as seasonality and trend based on historical values of web traffic. MLR model: Multiple Linear Regression(MLR) models uses data from data codification as independent variables and weekly traffic as dependent variables. This step helped the team to find below significant variables which are influencing the web traffic.  Marketing spend on search and display advertisements  Use of coupons from affiliate sites  Holiday events Model Integration: Time series model forecasts and MLR models weekly forecast are integrated to provide the recalibrated forecast values based on upcoming marketing activity. Model validation: Historical data is divided into test and training tests to validate the forecast values. User Interface: A spread sheet tool is developed with baseline forecasts from time series models and coefficients from regression models to help planners in developing alternative budget allocations Allocation of Market budget: Marketing and budget teams used optimization techniques to allocate the budget efficiently for online marketing activities and promoting various product categories. They used linear programming to perform this activity. HPDirect.com team used historical ROI values to forecasts revenues and expenditures. Time series models used to identify the expected ROI. This expected ROI used for allocation across the products. Again, they used previous ROI’s to revise future budget allocations and it is a continuous improvement process So, Spending budget for different marketing channels and product combinations are the decision variables and marketing managers are responsible for managing the traffic to the website from specific source(e.g., search marketing, affiliate marketing, email marketing, display advertising) so they prescribe the minimum and maximum spend constraints in consultation with DPdirect.com director.

Below is the mathematical formula of linear programming. Objective function Maximize total sales =ROI1 * SPEND1 +ROI2 * SPEND2+_ _ _+ROIn * SPENDn Constraints SPEND (1,2,3,4,..,n) ≤ MAX SPEND(1,2,3,4,..,n) SPEND (1,2,3,4,..,n) ≥ MIN SPEND(1,2,3,4,..,n),

where 1,2,3,....,n represent all marketing channel and product category combinations; SPENDn is the suggested budget allocation for combination n (i.e., the decision variable); ROIn is the ROI for combination n; ROI is defined as the ratio of revenue from a marketing vehicle and dollars spent on that vehicle; MAX SPENDn and MIN SPENDn represent maximum and minimum boundaries on SPENDn.

Solution B Implementation: Solution-B is a customer targeting solution and they call it as intelligent cube. It helps to improve the effectiveness of the direct marketing campaigns. HP used SAS software to develop OR solutions and other techniques which are part of intelligent cube. Intelligent cube consists of four different dimensions such as customer segmentation, product needs, purchasing time, Marketing channel preference. Customer segmentation: This dimension helps marketing team to understand to whom it should sell. It creates customer segments based on the attitudes towards technology and survey results. So, HPDirect.com developed customer specific profiles after segmenting them to six various groups by using linear discriminative analysis (LDA). To measure the impact, they have followed a test-versus-control approach. Results were encouraging to them, so they used this technique in different marketing campaigns. Product Needs:

This is a product recommendation model which helps the team to predict the most likely product that a given customer will purchase next from HPDirect.com. They have developed product matrix. It contains the probability of buying particular product X, given that the customer has purchased Y and then estimate conditional probability PXY as the proportion of in which X followed Y in all transactions that followed Y and is assumed to be stationary for the period of interest. They have used Markov chain model to do this activity. Again, they have followed test-cs-control approach to see the impact of the model and results were amazing. Purchase Timing: This model addresses two fundamental questions such as likelihood of customers stop buying products and rate at which customers buy HP products. So, they developed Bayesian hierarchical model to estimate the values of churn probability and rate of transaction. A Markov chain Monte Carlo algorithm is employed to sample from the posterior distribution of churn and transaction rate parameters. The prior distribution is chosen from the following parameters.  The number of transactions made by jth customer follows a passion process with rate parameter  Probability of dropping out after each purchase is binary with probability They computed score for customers after obtaining the parameters from regression model, which will help them to understand possibility of customers who can make a purchase from HP after obtaining the parameters from regression model. Marketing channel Preference: They have used logistic regression to build individual response model for each customer and they have used campaign response data, which consists of information such as channel used to make the purchase and other data sources (e.g demographics).Later, they assign each customer a propensity score, which denotes the likelihood of buying via the respective channel. For example, if the customer has high score for email channel than that obtained for the printed mail channel, then they infer that the customer is more likely to respond to an email campaign. Below are some of the significant variables they found in their validations.  Customers length of relationship with HP  History of buying electronics items  Credit card usage

Solution C: This solution helped HP to improve their downstream warehouse operations by increasing demand predictability and it helps them to ensure reaching of products on

time to the customers from their third-party warehouses. So, they developed model to forecast orders based on below parameters  Seasonality, including day of week, month of year and back-to-school season  Special events, including black Friday, cyber Monday and tax holidays  Marketing events, including holiday sales, promotional emails, coupons. Initially they developed simplistic multiple logistic regression, including 90 independent variables that primarily comprised of warehouse data. In phase II, HP developed hybrid forecast model to incorporate effects of marketing activities on warehouse orders and to the improve accuracy. This hybrid model includes both time series modeling and MLR modeling and HP team could able to see 15% improvement in the resultant model

Impact and Benefits: Operational Research solutions helped HP’s decision-making process across ecommerce value chain. Below are some of revenue improvements HP could able to see with the help OR solutions.  2.6% increase in their web traffic which resulted $44M improvement in sales.  15 % increase in order and 60% of conversion rate resulted $63 million additional revenue.  Better inventory management helped them to save $2 million and this equates 10 million in incremental sales. Since 2009, these solutions have driven an award-winning customer experience and incremental revenue of $117 million.

Conclusion: The OR solutions brought cultural shift in how the fast-faced and dynamic business can be managed and helped HP to solve business issues in its e-commerce business are novel and scalable to improve key business drivers and ensure profitable growth. As a result of these solutions in HP business, now the company is planning to develop them in 23 countries as part of its integrated e-commerce platform.

References Rohit Tandon, Arnab Chakraborty, Girish Srinivasan, Manav Shroff, Ahmar Abdullah, Bharathan Shamasundar, Ritwik Sinha,Suresh Subramanian, Dave Hill, Prasanna Dhore, (2013) Hewlett Packard: Delivering Profitable Growth for HPDirect.com UsingOperations Research. Interfaces 43(1):48-61. http://dx.doi.org/10.1287/inte.1120.0661...