If you think that your organization lacks the skills it needs to give birth to a successful predictive analytics initiative, take heart from the words of Dr. Spock, the famous baby doctor, say data mining experts. Trust yourself. You know more than you think you do.
You don’t need a data scientist to succeed with predictive analytics, and you don’t have to have a deep understanding of data mining. But you may need a little help to get started.
That’s the consensus of practitioners like Anthony Perez, director of business strategy at The Orlando Magic. “The folks that put together the models here all have MBAs. We don’t have anyone who’s a statistician,” he says. On the other hand, everyone understands how to do regression analysis and they are all Excel power users. And do get training, he advises. His team spent time learning the ins and outs of tools from SAS.
At the office of the CTO at the USPS Office of Inspector General, Bryan Jones’s team had data mining and statistical expertise on staff. That served as a foundation for predictive modeling, but he brought in a consultant to work on the first model and help bring his team up to speed. Now, he says, “We try to do some of the more simple models ourselves.”
Even Procter & Gamble, which has an advanced analytics practice, emphasizes a mix of skills. Sixty percent of analysts serve as business unit advisors who understand analytics but act more like business consultants. “Our recruiting profile is some technology undergrad with a business degree, operations research, and statistics specializations. That’s who we target for these roles,” says Guy Peri, director of business intelligence.
A deep technical background is a plus, but it’s not enough, says George Roumeliotis, data science team leader of data driven experiences at Intuit. Look for people who can understand the business strategies and processes first. “People run into trouble when they prioritize technical prowess over business insight.” But Roumeliotis does want analysts to have some technical chops as well. “I look for an understanding of some statistics, and the ability to use SQL and some mainstream programming experience with something like Java or Perl. In this way analysts may be able to get and prepare the data themselves, rather than waiting on someone else. An analyst, he says, needs to be “technically flexible” to understand and handle the data and modeling needs.
“You need to know statistics, and it helps if you’re a programmer to code up your own experiments,” says John Elder, principal at Elder Research Inc. “You want someone who is curious, motivated to find things, and who is humble because it’s easy to make mistakes, and they need to own up to those quickly. If they don’t, that can be disastrous.”
“You need some experience in either computational math or statistics,” but the best analyst prospects tend to tend to be people in the business who are already using Excel or Solver and can learn on the job, says Dean Abbot, president of data mining consultancy Abbott Analytics. Most don’t understand machine learning and predictive algorithms, so he teaches them what goes into a model, what should come out, and how to get a good result. It’s the qualitative aspects of the discipline that matter most, not the math, he says.
“A little math is nice,” Roumeliotis says. “But business savvy is priceless.”
Related Stories
- Putting predictive analytics to work
- Who owns your analytics group?
- IT-centric enterprise BI models unsustainable, says Forrester
- Strategic guide to big data analytics