Data-driven decision-making, American Express style

Michael Vapenik, the enterprise data governance officer at American Express, expounds on what it means to be a data-driven organization.

The term data-driven decision-making has a prominent place in the annals of modern business jargon. Back in the...

'90s when I was covering the retail industry, it was common to see analysis pieces debating the merits of data-driven retailers versus the "merchant" types who operated on gut instincts. Mickey Drexler, the now 72-year-old CEO and chairman of J. Crew Group, was often held up as the epitome of retailing by gut instincts in his days as head of Gap Inc. But that was B.B.D. -- before big data.

With troves of data now available to companies and the power of analytics, the debate about whether a company -- or a CEO -- should practice data-driven decision-making would seem to be moot.

Except, it's not, and I know this because talks, such as the one delivered by Michael Vapenik at the recent Chief Data Officer Summit in New York, remain a staple of IT conferences. Vapenik, the enterprise data governance officer at American Express, made it clear there's still much to learn about the art and science of data-driven decision-making -- from ensuring data quality to building a data culture that pervades the enterprise.

Experience matters

Vapenik has been working with data for 30 years. After college, he joined the now-defunct discount retailer Caldor Inc., where a perennial goal was to maximize sales and minimize rain checks, a compensation or coupon that stores offer if an advertised sale item is not in stock. "I thought it was going to be easy," he said.

There was plenty of data to work with, he recounted -- except the data quality was flawed. "We were taking rain checks for items that were listed as [inventory] stock on hand."

His education in data governance and analytics continued during stints at two mammoth public-sector agencies, the New York City Department of Sanitation and New York City Transit. Projects he worked on ranged from predicting the participation rate for the city's fledging recycling program to reducing employee overtime at the nation's largest public transit agency.

During an 18-year career at American Express, Vapenik has spearheaded many data quality and information risk initiatives across the company's product lines and won acclaim for his analytical, quantitative and operational skills.

Here are some of his thoughts on the characteristics of data-driven organizations and on the kinds of people who can help data leaders build and sustain a data-driven culture at their companies.

Key elements: Data-driven decision-making

Critical data is organized and can be trusted.

The key here is critical data, Vapenik said, "because you don't want to boil the ocean." Companies that don't and can't prioritize data won't make progress. He cited studies showing that data scientists spend between 60% and 80% of their time organizing and cleaning the data; instead, they should be starting with the "end product," he said.

Trusted data is data that's been treated like an asset: "You need to know who is accountable for the data; you want to know its quality -- good or bad; and, finally, you want to be able to understand its value, whether that's value internally to the company or the value of monetizing it," he said.  

Data is centralized and accessible, but with appropriate guardrails.

American Express has centralized its big data environment. Doing so improves the quality of the data by eliminating multiple copies and ensuring the timeliness of the data, Vapenik said. But "people get nervous" when data they feel belongs to them is collected centrally. To ease anxieties about losing control of the data, data chiefs need to "maximize the availability of the data," but still maintain "all the appropriate controls."

It's also important that data presentation and data access tools are customized to users' skill levels. Experienced analytics pros may be OK with a dashboard, he said, but will almost certainly want to dig deeper into the data.

Creating a data-driven culture

Companies that strive to inculcate data-driven decision-making often believe that if leadership understands the value of data, it will all work out. But that is not the whole story; there needs to be buy-in from the rank and file. "We all know that this is a two-pronged approach," Vapenik said. "The rubber hits the road from the bottom up." 

Here are the traits he's found are important to look for in people when building a culture that promotes data-driven decision-making:

Right and left brain input

At American Express, the leaders in Vapenik's group were tested for right and left brain characteristics. "We had one person out on the creative end; everybody else was on the left, and I was in the middle," he said. But in building a data-driven culture, Vapenik said leaders will need to understand and connect with groups such as marketing, which depend on right-brain talent.

His advice on building the leadership group: "Your best bet is to have a mix of people, or have people who are toward the middle of the continuum of the analytic mindset," he said. 

Natural curiosity

"People who are really interested in how things work and how things get created, they are the folks who ask a lot of questions -- the who, what, where and why. They will get deeper into understanding what the data is and, ideally, will help in coming up with new solutions," Vapenik said.

Curious people will also gravitate to new technology and pull others along, he said, adding that he puts himself in this category. A recent curiosity-fueled foray he took was a deep dive into blockchain technology.

Seeing the forest, trees and roots

Vapenik has worked with numerous big-picture people who "are great at strategy," but aren't good on details; he's also worked with others who spend too much time focused on the minutiae -- the roots -- and can't see the proverbial forest or the trees. "When I think about a data-driven culture, it's about trying to find people who can descend from the forest to the trees to the roots and back [again]."

Experience with bad or nonexistent data

Vapenik said he came out of school believing that his biggest concern was how to build a data model. But what turned out to be even more critical was dealing with bad or missing data.

"Some of my colleagues on the purely modeling side [think]: 'I want to fix this data to make my model work,' versus, 'I want to fix this data to get a good answer.' Those are sometimes not the same thing," he said.

Enhancing the data or filling in missing information may improve the model's performance, but the use of the information that model generates "can cause problems," he said. He cited his team's work on the New York City Department of Sanitation's recycling program. The job was to ascertain the amount of garbage collected that could be recycled -- with only the total tonnage to work with.

"The consultant was doing the dirty work of picking through the garbage, but we set up a stratified sample across the boroughs [on the hunch] that garbage created in Manhattan will be different from that created in Brooklyn and so on," he said, in order to tease the data out.

People who can think through problems that have bad or missing data, "we have found to be very helpful," Vapenik said.

Willingness to experience the data

For a project at American Express aimed at reducing disputed card charges, team members were asked to look at their own credit card statements. When Vapenik, a New Jersey resident, reviewed his statement, he saw a charge listed as "Mountainside," the name of a borough, and not as the movie theater's company name.

"I could see where that is a problem for the average consumer. But then I tried to think about why a merchant would do that," he said. The company had theaters all over, so it identified theaters by the city they're located in, not by the brand name of the chain. When people analyze the data, they first need to think about the motivation behind the data, he said.

Storytellers and influencers

Look for people who "can make the complex simple," Vapenik said. "The easier you can make it for people to understand the data, the more success you will have." Data presentations -- the use of charts, bar graphs or pictures -- should be pegged to the groups you are trying to persuade.

Another tip: People who are skilled at influencing others often have a high degree of emotional intelligence, Vapenik said. Their empathy for people makes them effective communicators who are good at translating complicated data into useful information. "There's a connection between heart and brain," he said.

Team players and orchestrators

Vapenik said finding people who possess all the traits he described will be difficult, if not impossible. In reality, data-driven decision-making rides on many types of intellects. "Go for a collaborative, team players," he said. But like an orchestra, the group will need a conductor "who steps in, so all you hear is the beautiful music."

Next Steps

Gartner: Tips for achieving data analytics 'abundance'

Data transparency the lifeblood of digital economy

Analytics priorities in 2017

Dig Deeper on CIO career development and career paths