Exploring the Cyber Frontier: Overstock.com

An ambitious data warehouse implementation—and great results—are all in a day's work for Overstock.com.

Over the past half-decade, Overstock.com has created a well-planned and thriving business. Initially focused on clearance merchandise and then expanding to books, videos and music, the company now has increased its product lines to include travel and auctions, with the goal of achieving at least 60% growth each year. To achieve and maintain that growth, the company has adopted technologies in its enterprise data warehouse (EDW) that transform supply chain and inventory management.

Online business tends to move quickly, but Overstock's goal, led by Jack Garzella who runs all of IT operations, data warehousing and reporting, was to build the EDW system and have it in production within 60 days. The goal was impressive, and any project that ambitious also has to be well-planned. Garzella led that planning effort as well as the project's implementation. Teradata Magazine talked with him about the company's experience.

Q: What were the primary challenges that led to a search for an analytical solution?
A: Bottom line? Our company had grown 100% a year for four years. Our business had changed and grown as well: We have added travel and auctions and greatly increased our books, video and music business. Throughout that period most of the company's (IT) focus had been on operational systems.

Quite simply, we had totally outgrown our decision support, reporting and analytic systems. Our current operational reporting systems could no longer function at the levels we required. There were scalability and performance issues as well as business rules that had to be modified. At the same time our competitors had become more intelligent so we took that into account as well. All of these were critical business issues that affected our revenue.

Q: Were you facing any challenges that were specific to Overstock's nature as an online retailer?
A: E-mail is critical to us, central to our marketing and our business. But our internally built e-mail system was struggling and was not easily modified for our future needs. It was a great send engine but it couldn't do any targeting. It was not good at processes such as hard and soft bounces, and there were increasing issues with our business partners including Yahoo and AOL.

Q: And how did these issues lead the company to commit to what was essentially a reinvention of its IT infrastructure?
A: The key word there is lead. Our CEO and CIO both wanted to leapfrog the analytic systems to where the business was and to be prepared for where we were headed while getting it done very quickly. Our initial timeline was 19 months, but we reduced that dramatically, with a five- to six-month timeline divided into three separate phases.

Q: So speed to deployment and implementation was your primary goal?
A: Not exactly. Putting a system in place that would let us manage our business better was the goal; doing it quickly was the need. We all had to be on our A game all the time. We knew what needed to be done. We knew we would be taking some shortcuts, but we didn't want to make any half-million dollar mistakes that had to be done over.

Q: To avoid those "half-million dollar" mistakes, were there any core principles you focused on to keep your operation on track?
A: We had a rule: The data must be right. We did not shortcut any of the data modeling and data access layer view definitions. If you write a report wrong, it costs you a few hours, if you model the data wrong, it can cost you weeks or months.

Q: Understanding those parameters, what led you to select Teradata?
A: Our decision was based on time to market and ability to execute. We were looking at 12 to 18 months with Oracle and no replacement of our e-mail system. Teradata was able to put a plan on the table that would get the job done in 12 months, and when we asked them to be more aggressive, they responded with a three-phase plan that had deployments staged at 60, 120 and 180 days. We used Teradata's Retail LDM and EDW Roadmap to help do this detail modeling very fast.

Q: Did the appeal of a brand-new architecture help guide the decision?
A: Our decision was based on getting the job done and done right. But there were great advantages to building from scratch. We had no legacy impediments to deal with, no old data structures. This proved important both for getting the job done right and getting it done on time.

Q: Your e-mail system was internally built. I assume that working in a 60-day timeframe you went with off-the-shelf e-mail and other applications?
A: This was another reason for going with Teradata. They offered an integrated stack of software that was good in all categories. Time to market was more important to us than best-of-breed applications, although every application we picked was number one, two or three in its category. We now use Postmaster's e-mail send engine, for example, and Teradata Warehouse Miner, Teradata Value Analyzer and Teradata CRM, among others. We use Teradata Logical Data Models based on their retail and travel models. There are 52 dual processor application servers feeding or being fed by our Teradata system.

The point is that across the board we were working with proven products and with Teradata consultants who knew what they were doing, understood the tradeoffs and understood what we wanted to accomplish.

Q: The result of your "Mission Impossible" was, I assume, a successful rollout?
A: We're still rolling out! The initial phases did turn out well and on schedule. Now we're adding financial metrics, scorecards and other tools. The biggest bang for the buck is the targeted and event-driven e-mail campaigns as well as our ability to manage our business in real time vs. daily or weekly.

Q: Clearly, your effort required a shift in the way IT and business work together.
A: If you looked at how Overstock operated until this year, it would look something like this: Business would ask; IT would build. When we launched this process we made sure that business was involved up front. In fact, the data warehouse itself changed the mentality—getting business users involved in every aspect of the project and in every phase of the project. Additionally, our business users faced a large cultural change—they were not used to working with IT that closely or using an integrated data store, our Overstock EDW.

Q: How did you guard against any loss of momentum or enthusiasm for the project?
A: Early in the project we established an executive steering committee that included the VPs of auctions, travel, marketing, merchandising and books/video/music as well as the CFO. It helped enormously that they were able to see the project progress, see how money was spent, and, in later phases, make adjustments based on what they wanted to spend, based on what we'd learned and delivered.

Real-world insight

The company's approach to enterprise data warehousing was relentlessly inspired by what management could accomplish to drive the business with the technology—and with information drawn from multiple sources. Real-time loads were accomplished from Oracle, Oracle Financials and SQL Server but also from the click-stream and in multiple ways from the company's e-mail systems, including opens, bounces and clicks.

The approach was also relentlessly batchless. The goal was to provide 24/7/365 data access including sales data, merchandising and both site and e-mail clicks to over 300 users. Dashboards were developed for executive monitoring and management of revenue and profit, as well as for customer and product lines.

Equal attention is paid to access provided to analytical users, with tools for deep data mining; it is a platform designed to help the company's more than 25 power users and business analysts transform data into knowledge and knowledge into competitive advantage.

—Jack Garzella, Overstock.com

Q: What did you learn?
A: When we approached phase one we did very detailed report requirements and basic e-mail functionality. But no matter how much you do up front, you have to let the users tell you how they actually use the system and also be alert as to how business processes would change. We didn't involve the users closely enough in the process during the first phase to understand the business process changes, only the reporting requirements.

Phase two involved much closer alignment with users' day-to-day activities, actually breaking out steering committees by user groups within the company.

Q: Since you brought up day-to-day operations, let's take a look at how your business differs from brick-and-mortar businesses. How do different parts of the company use the enterprise data warehouse?
A: Eighty percent of our workload is real-time or near real-time. We load data from the operational system into the data warehouse in near real time; it's never more than a minute or two behind. Marketing wants real-time information, and we give it to them. The business side wants information every hour or every other hour; finance pulls some data on a daily basis. We're providing data for lots of different parts of the company.

The important thing is that wherever a number is used, whoever is looking at it, whatever the view, it's the same number.

Q: Can you talk about some of the advantages of having such a highly real-time enterprise? What can you do with this fresh data that you couldn't otherwise do?
A: For marketing, they will continuously look at how our e-mail campaigns, keywords, searches and other drivers are working on the site or getting people to the site. They can make instant changes to the site or marketing campaigns if needed.

For operations, we take our revenue, transaction load and other metrics such as bandwidth and Web session usage and try to predict events that will affect our Web site availability or performance and to foresee issues at our warehouses and partners.

The rest of the company enjoys seeing real-time revenue and metrics. Our reporting portal (Oreports) has over 70 drill down/interactive reports. We roll up the company P&L every hour for the executives. There are many "hourly" e-mail reports that are generated to the various business leads. It allows us to manage our business daily and intra-day if needed for sales, customer service and other departments.

Q: Where do you see Internet commerce headed?
A: In our case, we're going to get bigger, better, faster. Overstock is committed to offering the absolute lowest prices on the Net and matching that with the absolute best customer service. We take our own returns and staff our own call centers. Our customers are always dealing with Overstock, and we are working to become a one-stop shop for Internet commerce.

As far as IT, we're going to be spending more of our time working with business processes to enhance our execution vs. building tons of new functionality, but there will be a good amount of that.

Q: Finally, what advice would you offer companies looking at this sort of migration to an enterprise data warehouse—whether or not they're going to do it as quickly as you did?
A: Get good executive sponsorship and keep them on board. Deliver functionality in phases. Get it done and get it done right—and don't let politics get in the way. T