Join Wayne Winston for an in-depth discussion in this video How do I determine what data I need to solve a problem?, part of Wayne Winston on Analytics.
This may be the most challenging aspect of analytics. In one, to be honest that we don't do that good a job in the classroom. because usually like if you use business school cases, they'll give you the data. So, let's talk about some examples of problems organizations needed to solve. And how you figure out what data you needed. So Hewlett-Packard was trying to reduce co, employee churn. In other words, they wanted less people to leave the company. So they said, what data do we need to try and figure out what makes people leave the company. So they tracked, for all their employees, the salaries, the job performance rating that HP gave them.
The raises that the employees got and the promotions that they got. And they tried to predict using a tool called logistic regression. Who was likely to turn and leave the company? And who was not? So they found some surprising things. You would think that if you got promoted, you wouldn't turn or leave the company. But it turned out it was more complicated than that. People who were promoted and didn't get a reasonable raise were much more likely to leave the company. Because a promotion entails working much harder. And so if you don't get paid for that, then that's promotion isn't worth that much to you.
So they realized they had to, in order to reduce churn, they needed to basically increase the raises they gave people with promotions. Another example of basically what data do we need. We want to evaluate defensive ability of an NBA player. So you would say, look at the box score. Well the box score in basketball is heavily weighted towards offense. We see blocked shots and steals, but we don't see much else. So it's hard to really measure from the box score how good a defender or player is. So a great defender throughout his NBA career's been Kevin Garnett. But in the box score it's not going to show up that well.
So what can you look at? Well, you can look at how many points does the team give up when Kevin Garnett is on the court versus when he's off the court. And historically, Kevin Garnett's sort of been a 10 point defender. Which means when he's on the court, after adjusting for other factors, the Celtics or Timberwolves would give up 10 points less than when an average NBA defender is on the court. So even though the box score doesn't indicate how good of a defender he is. We can figure out how good a defender he is by looking at how many points were given up when he's on and off the court.
United Healthcare had another interesting problem, where they were trying to keep health care costs flat. So what data did they need to keep health care costs flat? Well, basically, surprisingly, they needed to know whether employee is sick or not? So they paid their employees $450 to have a diabetes screening. They found out 30% of their employees had diabetes or pre-diabetes, and did not know that. So basically diabetes really contributes greatly to the bad Healthcare outcomes in this nation and around the world.
And also basically the cost of healthcare, so if you catch people early with diabetes, even if you're paying your employees $450 to be screened. You can have better healthcare outcomes and greatly reduce your healthcare cost. But you had to be smart enough to know, I need the datum, whether my employees have diabetes, so you had to test them.
- What is analytics?
- Predictive vs. prescriptive analytics
- What do you need to know to become an analytic professional?
- Looking at examples of good and bad metrics
- How mobile devices will shape analytics