Hewlett-Packard: Delivering Profitable Growth for HPDirect.com Using Operations Research and Analytics

The Problem

With Internet penetration increasing and customers taking to the Web over the past decade, building a strong online sales channel became an important strategic need for Hewlett-Packard. HPDirect.com was established in 2005 to fulfill the need for an online sales channel and to help realize HP’s vision of enabling a more holistic buying experience for the consumer across multiple touch points—retail stores, the Internet, and by telephone. However, for a company where the systems, processes, and people were all wired to sell to large organizations and retailers, building the capability to sell directly to consumers was challenging. In addition, HPDirect.com, like other e-commerce companies, is constantly faced with the need to improve the volume and quality of online traffic, increase conversion of visits to transactions, and increase order sizes, all factors directly impacting revenue and profits. The ability to accurately predict online demand also impacts fulfillment costs and lead times, and impacts profits and customer satisfaction.

The Analytics Solution

HPDirect.com and data scientists at HP Global Analytics developed a set of solutions based on mathematical programming, Bayesian modeling, regression modeling, and time-series forecasting to address these key business issues. Demand generation models help quantify the impact that online marketing activities have on customers visiting the online store. These models, based on a hybrid approach of time-series forecasting and multiple linear regression, have been helping HPDirect.com management prioritize marketing dollars across online marketing activities to maximize traffic. Because they provide predictions of weekly Web traffic that are accurate within 15% of the forecast, these models have become integral to the planning of demand generation activities at HPDirect.com. Next, a customer targeting framework, based on a combination of regression models, Bayesian hierarchical models, and discriminant techniques, predicts a customer’s product choice, timing of purchase, and marketing channel preference.

The Value

This solution helped ensure that the customer received the right offer at the right time, leading to increased conversion rates and order sizes. Enhanced customer targeting using this framework has yielded an average incremental conversion rate of 1.5% on an average base conversion rate of 2.5% across a series of marketing campaigns. These up-sell/cross-sell models have helped drive a 20% improvement in average order size. Further, these analytics-based solutions in the period 2009-2011 have increased online customer traffic, improved conversion and up-selling/cross-selling, and reduced warehouse inventory costs to generate $117 million in incremental business impact. These capabilities and new analytics-based solutions developed to grow HP’s online sales channel in the United States are now being leveraged across similar HP businesses in 23 countries globally. By creating desktop-based solutions where the technical sophistication of the models and algorithms developed is embedded in simple user interfaces, the Global Analytics team has ensured high levels of business-user adoption.