You may have seen the live blog we produced for the morning plenary sessions on Thursday and Friday , Sept. 25 and 26, at the Innovators' Summit. The afternoon concurrent sessions at the Summit showcased all sorts of fascinating uses of analytics -- from understanding what consumers want in products to providing early warning of earthquakes. Anne Bullard and I split up and covered all 10 sessions over two days. Here is a look at each of the afternoon presentations.
THURSDAY, SEPT. 25
Rob Reul, Founder & Managing Director, Isometric Solutions
Trading Performance for Profit
If you were designing a new laptop computer, you’d probably want to know which features are most important to potential buyers. Would they most value the size of the memory or the hard drive? Performance speed? Battery life? Price?
One way to find out would be to survey prospective customers using conjoint analysis research, which derives the relative importance of various product attributes using statistical analysis. Rob Reul, founder and Managing Director of Isometric Solutions, an international marketing firm, says conjoint analysis offers “crisp, competitive, sharp marketing analysis.”
Until now, surveys have provided meaningful results only when respondents completed the entire survey. But Reul thinks choice marketing analysis, which will be possible in soon-to-be-released JMP 8 software, will make it possible to get good results by dividing survey questions between different groups. So surveys that now require nine to 12 minutes to complete can be reduced to 4 to 5 minutes.
“It’s a very iterative way to get to your customers and get some very helpful feedback and knowledge without wearing out respondents, which is important for the validity of analysis results,” Reul said.
Manuel Uy, Principal Professional Chemist, Johns Hopkins Applied Physics Laboratory
Solving Practical Problems from Earth to Mercury
Manuel Uy presented two fascinating examples of how the Johns Hopkins Applied Physics Laboratory used JMP: facilitating the launch of the MESSENGER spacecraft, which is now on its way to orbit Mercury, and evaluating detection systems for protecting first responders and troops in the event of a chemical or biological attack.
Uy explained that MESSENGER had a potential power failure problem related to the chip resistors used in the spacecraft and that he led the team at APL tasked with resolving the problem. If the chip resistors were to fail, the spacecraft would be out of power before it even reached Mercury, he said.
Design engineers, part manufacturers and fabrication engineers all pointed fingers at different things as the possible causes of the failure of the chip resistors. He and his team used JMP's design of experiments (DOE) capabilities to study five factors and found that humidity and the type of chip resistors were the likeliest culprits in the chip resistor failure. Thanks to his team's quick analysis of the situation, the spacecraft was able to launch after only a two-week delay.
In his second example, Uy showed how he used JMP's Custom Design feature to determine very quickly which of the seven systems he was asked to evaluate would be best for detection in biological and chemical attacks. He found that immunoassay systems were better than spectroscopic-based systems. He also found that immunoassay with optical detection was a more effective technology than immunoassay alone; however, such combination systems are larger and more cumbersome, and would be less portable for first responders on the move, he said.
Uy extolled the virtues of DOE, noting that "younger engineers are a lot more open to DOE than older engineers. It should be required for all science graduate students."
Emmanuel Roche, Vice President of Research and Development, Teragram
Statistics in Natural Language Processing
Analysis of text can happen in a number of ways, said Emmanuel Roche, VP of R&D for Teragram, a company recently acquired by SAS. Among the types of text analysis are binary classification, speech recognition and machine translation.
An example of binary classification is sentiment analysis. Software can classify mountains of documents, for instance, as either "positive" or "negative," which may be useful for a company that wants to quickly digest thousands of articles that relate to its products.
In an analysis of 10,000 product reviews available on Amazon.com, Roche found that certain words were predictors of positive reviews, such "must-have" and "must-read." However, the word "good" was too ambiguous to be an accurate predictor of sentiment. Words with negative connotation included "refund," "disappointing" and "waste."
Roche showed the results of two "round-trip translations" using Google's language tools. He translated a news headline about the US stock market from English to Chinese and back to English and from English to French and back. Both round-trips resulted in mistranslations of the verb "shore up."
He described the size of a language as containing 80,000 simple (though very ambiguous) words, between 1 and 10 million compound nouns, and 100,000 verbal expressions with 10 to 100 structures each.
He pointed out that people now have access to more content than ever before, and that the content was not unbounded. In addition, he said that both structure and linguistics were keys to effective text analysis.
"There's a lot of data to be extracted from text," Roche said.
Scott Lasater, Director of Lean Six Sigma Enterprise Institute, TQM Network
How Systematic Innovation Beats the Lone Genius
Six Sigma expert Scott Lasater quoted Will Rogers, who once said, “It’s not what you don’t know that hurts you. It’s the stuff you do know that ain’t so.”
Six Sigma, with its continuous improvement process, is designed to eliminate defects in processes and make the more efficient, all while using feedback at the end of the cycle to introduce new refinements.
“Organizations need a way to combat the process of employees becoming prisoners of their procedures and rules,” he said. “They need to avoid presuppositions and belief systems that aren’t true. People aren’t trained to look for innovative ways to improve systems.”
Thomas Little, President of Thomas A. Little Consulting
Financial Analysis and Return on Investment Modeling
Thomas Little described a project where a business hired him to devise a strategy that would cut total production costs by 30 percent, a goal that even Little thought might not be achievable when he began. Today, he said, the goal is within sight, and he credited JMP software and Six Sigma processes for the promising results.
Little focused on cutting the unit cost of production, and he used JMP for the Six Sigma analysis that let him examine his processes.
“I always use JMP because it has the analytical horsepower to drive the kind of projects we’re dealing with,” he explained. “The power of JMP is really an essential component of that strategy.”
Clark Abrahams, Chief Financial Architect, SAS
A Comprehensive Credit Assessment Framework (CCAF) – Merging the Best of Art & Science to Optimize Credit Granting
Noting that his presentation was essentially about problem-solving in the business world, Clark Abrahams opened his talk with a quote from Professor Robert E.D. Woolsey: "People would rather live with a problem that they cannot solve than adopt a solution they cannot understand."
The top characteristics of the ideal credit-granting solution, as identified by SAS' financial services customers, are that it is accurate, fast, cost-effective, flexible, consistent, reliable, easy to understand, based on proven lending principles, able to rate the borrower's ability to replay the loan, able to be effectively monitored, able to provide adequate controls to limit risk, and adaptive.
Abrahams' presentation focused on the details of the CCAF, Comprehensive Credit Assessment Framework, which he argued was a better approach to underwriting because it systematically integrates sensible judgment into a statistical process; can incorporate scoring models as dimensions in a more general model framework; is geared to dynamically handle multiple business contexts via generic primary-factor ratings; handles missing, incomplete and non-traditional data; and is an adaptive solution with an efficient, accurate and intuitive model validation process.
He used the JMP Tree Map visualization to show how it quickly helps financial services organizations see where they have overstated and understated risk.
FRIDAY, SEPT. 26
John Cunningham, R&D Manager, G3 Enterprises
In the Wine Country -- Changing Tradition
In John Cunningham’s profession, “closure” isn’t an emotional term. It’s a cork or cap for a bottle of wine or champagne. Getting just the right design is essential to preserving taste and quality. Cunningham, Senior Manager of R&D and Process Improvement at G3 Enterprises in Modesto, CA, spoke Friday at the Innovators’ Summit outside San Francisco. G3, founded in 2003, designs corks, caps and stoppers, as well as labels, bottle designs and other industry-related products.
“I brought JMP with me from the pharmaceutical industry to the wine industry,” Cunningham told attendees. “That was one of the things I told them in the interview process, that I wanted to have access to the software.” In pharmaceuticals, he had used JMP to design coatings for medications. In his new job, he would design coatings for corks to make them easier to remove and more effective at sealing bottles.
But introducing technology and hands-off processes into an industry known for craftsmanship and romance did not come easily. Wine makers felt their status as experts threatened by computer experiments designed to fine-tune processes that had been part of their craft.
G3 has successfully combined tradition (a wine maker is part of Cunningham’s R&D team) and statistical analysis to determine the cork length, diameter, aspect ratio and density needed to assure optimum “cork pull-force,” the amount of force it takes to remove a cork from a wine bottle. It has used designed experiments to disprove long held but incorrect theories that were contributing to other challenges. Efficiency has improved profits.
Customers are now asking G3 for help with their own challenges. Wine makers are optimizing fermentation processes with designed experiments. And the marriage of science with the art of wine making now promises a happy ending.
Dick De Veaux, Professor of Mathematics and Statistics, Williams College
Exploratory Data Modeling: The First Step for Successful Data Mining
In Dick De Veaux's standing-room-only presentation, exploratory data mining came alive in problems that ranged from how emergency room doctors quickly decide whether a patient is having a heart attack to how a nonprofit organization can optimize donations.
He referred to several definitions of data mining, including one from Usama Fayyad: "the nontrivial process of identifying valid, novel, potentially useful, and ultimately understandable patterns in data."
Attendees learned about different types of data mining models, as well as visualization methods, such as tree diagrams and mosaic plots.
In one example, De Veaux used the Prediction Profiler in JMP to find interactions between five variables from a data set containing 200 continuous variables and 50 categorical variables.
"Data mining is exploratory. It's great for developing new hypotheses and for reducing dimensions in data," De Veaux said.
Strong Huang, Six Sigma Master Black Belt for Worldwide Six Sigma Program Implementation, Beckman Coulter Inc.
Business Breakthrough Using Traditional Six Sigma
Earlier in his career, Strong Huang was part of a team of engineers hired to identify a major problem with a product and fix it quickly. The team implemented Six Sigma methods designed to improve quality and efficiency in manufacturing. The Six Sigma process is a loop that includes optimizing production and quality, eliminating defects and using customer feedback for continuous process improvement.
“I have to use all my knowledge to solve these kinds of problems,” Huang told his Innovators’ Summit audience. He is convinced of Six Sigma’s value for problem-solving and innovation. Huang, a Six Sigma Master Black Belt, works for the medical devices company Beckman Coulter, where he leads worldwide implementation of Six Sigma processes.
Richard M. Allen, Associate Professor, Seismological Laboratory, UC Berkeley
Warning for the Next Earthquake: Rapid Data Analysis Before the Ground Shakes
The goal of early warning of earthquakes is to provide residents a few seconds to a few tens of seconds of warning, Richard Allen said. That may be enough time to exit from a bridge, to hide under a desk or shut off sensitive machinery in a manufacturing facility.
In a highly interactive session, Allen described California's system of detecting an earthquake in progress and of predicting ground shaking and notifying residents. All available communication technologies, including broadcast media, the Internet and mobile phones, would be used for notification, he said.
He also discussed the issue of when to warn residents and ask them to take action. If the alarm threshold is set higher, organizations will have to deal with the possible consequences of more missed alarms, including damage to facilities and equipment, and harm to people. If the threshold is lower, organizations will face more false alarms, and that may harm productivity. School systems might prefer a lower threshold, whereas a chip manufacturer might like a higher threshold.