Analytics: Art and Science of Better

Originally published 6 January 2010

Unleashing the Potential for Improvement

Select the best answer to complete this statement: Analytics is...
  1. Story telling.
  2. Treasure hunting. 
  3. Trial and error. 
  4. Applying science. 
  5. All of the above.
If you picked “E,” you understand that analytics is applying both art and science to make things better. The challenge in extracting truths from data is evident from this M2003 quote from Professor David Hand, Imperial College of London: “Data are facts. Facts cannot lie, but they can often mislead.” This generalized version of the scientific method – observe, define, measure, experiment, learn and act – can and should include creativity and innovation along the way.

Analytics is the art and science of affecting better – making the invisible visible, aiding judgment under uncertainty, overcoming cognitive blind spots, allowing us to continuously learn and improve. Though to many, analytics is an abstract concept; it’s really about people – applying science, advancing science, being creative, innovating and making things better.

Analytics allows us to achieve better results, allocate scarce resources more efficiently and effectively, achieve objectives given various constraints, assess next best actions, anticipate the future to be prepared for any set of scenarios and to make more informed decisions. Analytics supports achieving full potential – we humans are very much about exploring, discovering and pursuing new and better ways of doing things. Einstein said, “The important thing is not to stop questioning. Curiosity has its own reason for existing.”

Leveraging the Power of Analytics

The significant and growing interest in analytics reflects the ever-increasing number of problems to be solved and opportunities to be identified. The Lady Tasting Tea by David Salsburg highlights the contributions people have made to significantly advance the science of science. At the end of the day, it is about people – advancing science, applying science and creativity, pursuing better. One of my favorite quotes is from statistician John Tukey: “The best thing about being a statistician is that you get to play in everyone’s backyard.” [To extend this quote, we could substitute modeler for statistician since many numerate people don’t feel comfortable calling themselves statisticians; however, they have extensive quantitative talent (operations researchers, econometricians, psychometricians, etc.).] Analytics is relevant everywhere and now organizations have begun asking questions of their data, which, in turn, leads to more questions.

Many organizations recognize, as Tom Davenport pointed out in Competing on Analytics, that data and analytics should be treated as strategic assets. If you ask, “What does it mean to treat analytics as a strategic asset?” you will see that it’s about the whole analytic infrastructure – people, process, culture and technology. Companies are very interested in investing in their analytic infrastructure because it is a strategic investment critical to ongoing success. In times of rapid change, you need to learn rapidly. Trying new things is important (especially in the face of so much change) so you can continuously learn and improve. Getting people to try new things is critical and requires an analytic infrastructure conducive to experimentation.

If we were in a laboratory environment, we would start with observing some phenomena, measuring relevant attributes and then formulating the problem(s) to solve. While many of the problems we need to address don’t have a lab setting, we can still be more scientific about how we approach decisions.

The scientific method provides a means to an end. People make the tools, processes and methods (and the politics), and people bring creativity and innovation (and biases) to what they do and how they do it.

This is a generalized view of the steps involved in extracting truths from data and one that supports continuous learning. Part of treating analytics as a strategic asset means addressing process improvement opportunities along the way – viewing analytics itself as a process as well as relevant to and applicable within other processes. The point is, an analytic perspective pays off and it starts with observation.

Observe: Getting Started with Analytics

There is value in simple observation – in seeing the data you have and determining what additional data you may see the need to collect. The hand-washing example has appeared in several places (a favorite: Better by Atul Gawande). A physician, Ignaz Semmelweis, noted that at Vienna General Hospital maternity patients were much more likely to die when treated by doctors who had recently left the autopsy room. He concluded that there was a connection even though germs were then unknown. He did a statistical study, which indicated that mortality rates declined if doctors and nurses washed their hands in chlorinated lime before seeing the next patient (more on this story in a moment). The more modern-day example of hand washing involves the chief of staff at Cedars-Sinai Medical Center. After going on a cruise where the staff so diligently supplied hand sanitizer to passengers, he wondered if the hands of the people on the cruise were cleaner than his staff. The cultures were taken (measures) after the staff said their hands were clean. You can guess the results. Cedars-Sinai launched a hand-washing campaign with visual reminders, including the screen saver on the PCs at Cedars-Sinai. It did change hand-washing behavior and resulted in lower rates of infection.

The Importance of Definitions

Defining issues, events and how they are measured can pose real challenges. There are opportunities to innovate here as well. Pondering definitions and metrics can even lead to breakthrough, new problem formulations. The example of survival analysis applied to customer data more than a decade ago is a great example of innovatively reformulating the common customer relationship management (CRM) problem of predicting customer attrition. Defining and measuring customer events of interest can be tricky as well – such as when is a customer past due on a payment. Once the problem is set up as a survival problem, you can then answer the question of when not just if an event of interest will happen.

Measurement Costs and Benefits

I am a big fan of advocating data as “measures” and data collection as “measurement.” We are either actively or passively measuring things when we collect data. Actively, we are measuring with a purpose –typically to send a bill – but there are often valuable secondary uses for the data regardless of primary purpose(s). Passively, we often collect data because we can and/or might need to refer back to the data – customer calls, e-mails, etc. Measurement itself can and should be viewed as a whole separate process unto itself.

There is a cost to measure and we can’t always get ideal measurements for any given analysis. There is a good example of measurement challenges and costs in the book How to Measure Anything by Douglas Hubbard. In the 1970s, the Environmental Protection Agency (EPA) was challenged to estimate the effect of continued use of leaded gas on air pollution. Even though new cars were to take unleaded gas starting in 1975, people were still putting leaded gas in their new cars by simply removing the larger nozzle restrictor. How often was this happening and to what extent would it worsen air pollution? The EPA devised a way to “observe and measure” that proved effective. They randomly selected gas stations throughout the country and through binoculars observed what was happening at the pump, recording leaded or unleaded use, and used DMV licenses to learn the vehicle types. Through observation and measurement, the EPA learned that 8 percent of drivers who should be using unleaded were using leaded gas, and this helped them reduce the error and uncertainty in estimating pollution. In addition to the obvious cost of sending staff out to gas stations around the country, the EPA also incurred the cost of some bad PR as they were perceived as “spying” on people at the pump.

We can all benefit from a more analytical perspective when implementing and evolving our data collection processes. Much more attention should be paid to what we are actively and passively measuring, what primary and secondary uses we can make from data we are gathering, and if we are measuring what matters most.

Challenging “Business as Usual”

If we don’t try new things, we won’t know if we could be doing better and won’t learn as efficiently as we could. This often means cultural changes for many organizations by providing permission to fail; sometimes, the learning is more valuable. Encouraging people to try new things can be challenging. Influencing behaviors is hard, but often required to advance our understanding and to affect change for the better.

Suppose we’ve done our best; we’ve measured new things and measured more things, and conducted experiments to augment our learning. For many analyses, arriving at “the answer” is not the end goal. How we communicate the answer matters immensely – remember poor Semmelweis’ delivery, tantamount to “You’re killing people!” When he proposed the practice of washing hands, he was ridiculed by his peers. They refused to believe they had any role in causing patient deaths and felt washing hands several times a day was a waste of their time. These surgeons proudly wore blood-stained gowns to show their experience in surgery. Doctors just didn’t believe something they couldn’t see or prove was a factor or a problem. The controversy resulted in Semmelweis being fired. After a nervous breakdown, he ended up in a mental hospital and died at 47. His study ultimately resulted in confirming germ disease. How he chose to inform others could have been better. Years later, Joseph Lister told the story in a way that people were able to appreciate, understand and react appropriately to the information.

Challenging business as usual is about much more than arriving at “an answer.” It’s not simply a matter of saying “here are the results” and walking away. It’s more a matter of helping the data tell the story. Depending on what’s at stake, convincing and persuading may be needed as part of sharing results.

One way to more efficiently and effectively tell the story is through visualization. How we share results and how we challenge “business as usual” is important. The 2006 hand-washing story is a success. It did change behavior at Cedars-Sinai, resulting in cleaner hands and lower rates of infection. This is a case where visualization helped tell the story in way that compelled others to action. In contrast, the way Dr. Semmelweis shared his results (admittedly an extreme) is sadly a failure; it did not change behavior at Vienna General Hospital and could have helped others recognize the truth about germs much sooner.

If analytic results are going to challenge old ways of doing things, we are back to trying to influence behaviors. What worked yesterday may not work as well (or at all) today. There are ways of challenging business as usual that may make it easier to overcome some obstacles to improvement. Through best-practices sharing and various ways of fostering an analytic community within organizations, you can cultivate a learning culture that is more open to change. One way to foster innovation is to establish an analytic center of excellence. There are many starting points, depending on your organization’s culture and structure. The fundamental question is, “Are both data and analytics regarded and managed as strategic assets in your organization?”

An analytic center of excellence is a central point to promote best practices and facilitate the adoption of analytics across the organization. So often, analytics are siloed in departments. While you need analytic talent close to business issues like credit risk, marketing optimization, etc., business processes span departments. By supporting an analytic center of excellence and allowing it to have the whole process view, you have a greater chance of whole process optimization and not just localized, subprocess optimization.

We started with a quote from David Hand, “Data are facts. Facts cannot lie, but they can often mislead.” Similarly, Friedrich Nietzsche said, “There are no facts, only interpretations.”

At the end of the day, analytics centers on people. Enabling analytics to have the greatest impact requires dealing with the social dimension. We all have unique interpretations of our world; we all have biases and cognitive blind spots. Analytics can help us be more objective, but the context in which we act and compel others to act requires more than just getting the answer. One of the main lessons Dan Ariely gleaned from the research in Predictably Irrational is, “Although irrationality is commonplace … once we understand when and where we make erroneous decisions, we can try to be more vigilant, force ourselves to think differently about these decisions or use technology to overcome our current shortcomings.”

If we are ready to treat data and analytics as strategic assets so we can achieve greater impact, we need to consider how we: share results, make the invisible visible in a compelling way, aid judgment under uncertainty and challenge business as usual, enabling continuous learning and improvement in our pursuit of better.

SOURCE: Analytics: Art and Science of Better

  • Anne MilleyAnne Milley
    Anne, Senior Director of Analytic Strategy, Worldwide Marketing, at SAS, directs analytic strategy for global SAS product marketing. Her ties to SAS began with bank failure prediction at Federal Home Loan Bank Dallas and continued at 7-Eleven Inc. She has authored papers and served on committees for F2006, KDD, SIAM, A2010 and several years of SAS’ annual data mining conference. In 2008, Milley fulfilled a five-month SAS assignment at a UK bank. Anne is a contributing faculty member for the International Institute of Analytics.


Want to post a comment? Login or become a member today!

Be the first to comment!