The game of blackjack offers six important lessons for managers trying to make sense of marketing data, according to the man who crunches the numbers for Orbitz Worldwide.
Sameer Chopra, vice president of advanced analytics for the travel company, shared those tips Tuesday in a keynote address to about 250 data scientists and analysts attending Predictive Analytics World conference in San Francisco.
Sharing an old joke, he noted the best way to come home from Vegas with a small fortune is to start with a large one. Yet, blackjack is the one casino game that is winnable, legally, through card-counting and acting on the data, he said.
It's All About Numbers
The analogy of card-counting to predictive analytics might startle some in the industry, but both are based on minimizing risk by optimizing the use of available data to forecast outcomes. Chopra noted he is not endorsing gambling.
Data almost always beats intuition. You might think it is better to hold at 18 in blackjack, but you should hit. "I see this in the workplace," he said, noting that people are afraid to take a small risk, but the "really small changes" can yield to significant improvements, he said.
Wake up — Status quo and complacency is a losing strategy. Data strategists may be more comfortable with tried-and-true methods like SAS and SQL. But they need to continually evolve with the industry, getting familiar with alternatives like R and Python, open source, Hadoop and big data.
"Change before you have to," he said. "Wake up. Otherwise you're living with a losing strategy." He said with YouTube and free university lectures, there's no excuse not to learn. At Orbitz, for example, his team has learned it can generate a 40 percent lift by blending in text-based factors into its offers. It is also analyzing audio from call centers after it is converted to text and then mined. That has helped to address customer concerns up front.
Learn the power of ensembles vs. the individual. This rule applies to using data models, he said. Instead of trying to identify one correct model, it's better to combine models because that approach reduces variances. When building a data team, he said it's better to recruit employees with complementary skills rather than a "unicorn" with all the skills.
Think long term. In blackjack,even with a "perfect" strategy, you will experience long losing streaks. It's important, Chopra said, to have confidence in your long-term vision. Our society today is based on instant gratification, he noted. Today, the average length of time to own a stock is five days, but a half-century ago, it was eight years.
For Orbitz, this lesson translates to focusing in the long-term value of customers. "We're extremely focused on the customer," he said. "Some customers are more valuable than others." The result is that Orbitz now bases its offers on the perceived customer lifetime value, a strategy that has improved outcomes.
Carpe Diem — seize the day and the opportunity. When using data, it's important to play the odds and take risks rather than staying put. In cards, risk takers might double down — and they may lose more when they do. But when they win, they tend to win more, he noted.
Orbitz has adopted an uplift model that is based on focusing coupons on customers who are likely to spend more instead of those who were likely to make a purchase anyway. He called this group the "persuadeables."
Human judgments are essential to decision-making. "At times, we do need to deviate from the script," he said. When playing blackjack, for example, you may want to change from a card-counting strategy to avoid detection by cameras in the ceiling. Card-counting isn't illegal, but it is often unwelcome in many casinos.
At work, sometimes you need to look beyond the data to make the right decision. "I think it is very foolish to think that the human element will ever be out of the picture."
- SharePoint is Already Legacy
- Are You Too Old to Work in Tech? IT's Midlife Crisis
- Has Google Just Reinvented Gmail?
- What to Do When Yammer Adoption Stalls
- Faking Big Data #strataconf
- Is Your Information Architecture Ready for SharePoint 2013?
- Web Content is Obsolete