Industrial Statistics Meet Data Overload
I walked into the conference room, referred to euphemistically as the “war room.” I’d been sent by my boss to see if I could help the marketing team with a data-analysis problem. My boss was a senior vice president from headquarters, meaning that I wasn’t exactly welcomed with open arms by the team of data analysts who had semi-permanently taken over the room. I wondered how I’d be greeted if I announced, “I’m from headquarters, and I’m here to help you.” After appraising the expressions on the faces of the analysts, I decided to keep my humor to myself.
The problem was that the company wasn’t going to make its quarterly numbers unless internet sales picked up considerably. The team of marketing analysts and designers was charged with developing a plan to accomplish this. Their proposed plan included a number of special offers; how to present these offers on the web was the topic under study. The large darkened conference room was awash in the glow of laptop screens, and the walls were papered with candidate web-page designs, charts, graphs, and tables of data. The analysts looked intense and concerned as they peered at their screens, ignoring me.
I’ve long been a proponent of mining historical data for information that can be used to improve product and process performance. There’s gold to be mined from the vast data warehouses accumulated in the internet age. In this case, however, the past could offer little guidance. Instead, we needed real-time information on the voice of the customer to guide the decisions of our leadership team. We also needed creative ideas from our web-design teams. We had plenty of ideas and data--too much, in fact, for the analysts. The problem wasn’t their intellectual capacity; these folks had plenty of brainpower. The problem was their statistical toolkit, which was lacking the most powerful tool of all for the task at hand: statistical design of experiments (DOE). As a result of their ad hoc experiments, the team was going in circles trying to make sense out of the millions of data points they’d gathered on the 19 offering-and-design-factors under test. You could almost smell the smoke from the overworked spreadsheets.
It turns out that the number of combinations that can be made from 19 two-level factors is 2 19. These 524,288 combinations is a number big enough to confuse just about any team of analysts with only spreadsheets and printouts at their disposal. There was simply no way the team would ever reach an actionable conclusion in time to give quarterly sales the bump they needed. However, using DOE, we could analyze the main effects of these 19 factors using only 20 web-page/offering combinations; 40 if we wanted to repeat the experiment. When I presented this information to the marketing vice president, he was incredulous. His team was equally dubious. The situation was a stalemate until, a couple of weeks later, the marketing vice president left for greener pastures. The new guy was more open to the idea, and we were able to conduct a screening experiment and a follow-up factorial DOE that revealed an interaction effect that more than doubled the response rate on the web in time for the company to make its numbers.
This is one of countless examples of how the power of industrial statistics is now being applied in new ways in the brave new world of the internet, where massive amounts of data are available in real time. Quality and Six Sigma professionals have much to offer, but you must venture outside of your comfort zone and enter into new areas of the company. You will, of course, encounter resistance and disbelief. This means that you’ll need the active support of your leadership to make it happen. Which means that your first order of business is to prove to the leader that you have something of value to offer.
One place you might begin is by digging into the literature to uncover successful Six Sigma project case studies in nontraditional areas. You’ll find this information in the archives of this magazine ( www.qualitydigest.com ) and elsewhere on the web. You might also look at books that describe new areas where these techniques are being applied. A couple of good books are Moneyball: The Art of Winning an Unfair Game , by Michael Lewis (W.W. Norton & Co., 2004), and Super Crunchers: Why Thinking-by-Numbers Is the New Way to Be Smart , by Ian Ayres (Bantam, 2007), which are entertaining as well as educational reads.
Ultimately, much of what Quality 1.0 was about was looking backward. We studied history to determine what failed, then used this feedback to determine what kind of corrective action should be taken to address the root causes of problems. Quality 2.0 demands that we focus our attention on the future. We must use new tools to look at data generated to test our new process and product design ideas in real time. We must focus on the future. Quality must become proactive rather than reactive. It will be a challenge to make this happen, but it will also be a lot of fun.
Thomas Pyzdek is the author of The Six Sigma Handbook (McGraw-Hill, 2003), Quality Engineering Handbook (CRC Press, 2003), and The Handbook for Quality Management (Quality Publishing, 2000). He is a consultant on process excellence. Learn more at www.pyzdek.com . Contact him at tom@pyzdek.com .
|
