Eric Siegel, Ph.D., founder of Predictive Analytics World and author of Predictive Analytics, kicked off the Marketing Research Association’s third annual Corporate Research Conference last week with a great introduction to predictive analytics.
Eric defined predictive analytics as: technology that learns from experience [data] to predict the future behavior of individuals…. in order to drive better decisions.
At its essence, a predictive-analytics system takes in profile data, some of which includes the attribute to be predicted, then builds a model and generates a predictive score for all other records in the data set. The predicted attribute can be whether a prospect or customer will click a link, respond to an offer, or cancel a subscription.
Now, the resulting model may not be particularly accurate. In fact, accuracy shouldn’t be expected. As Niels Bohr said, “Prediction is very difficult, especially if it’s about the future.” Eric said, “Predicting that is better than guessing is often very valuable and sufficient in itself.”
He gave the example of a mailing to 1 million consumers that produced a 1% response rate for a $220 price. If the mailing itself cost $2 per piece, the profit would be $200,000. Now imagine instead mailing the 250,000 consumers whose predictive score was in the top quartile. If that converted at 3%, then “this model stinks,” said Eric, “but the profit would be $1,150,000! That’s a skunk with bling.”
In fact, Eric said, “The prediction effect means a little prediction goes a long way.” It tips the odds in your favor. He quoted Barry Jennings of Dell, saying predictive analytics “is not perfect, but it can reduce risk and maximize the potential.”
Causation and Predictive Analytics
Don’t worry about causation. Correlation is enough. For instance, consumers who purchase diapers are more likely to purchase beer. “People tried to interpret it, but it simply turned out to be true: Daddy needs a beer.”
Just because items are predictively correlated “doesn’t tell us anything about the why”. For instance, for one firm, the domain name of the email address was highly predictive: an Earthlink.com email address converted at the rate of 5 times that of a Hotmail.com address. Again, no certain causation.
“Things are connected to one another and that manifests in the data, as predictive analytics find these relationships. The data speaks! Dental patients miss fewer payments and are better credit risks. People who purchase felt tips for the bottom of chairs are good risks. People who buy large dog collars are worse risks. Finding an explanation? That’s just entertainment.”
Items that are correlated may simply share a common cause. For instance, “If people who eat ice cream are attacked by a shark, a causal explanation could be that eating cream makes them taste better!”
Of course, a more likely explanation is simply seasonality: “in warm weather, more ice cream is eaten by humans, and more humans are eaten by sharks.” The two share a common causal element; one does not cause the other; they are correlated but not caused. Again, knowing whether it is causative is often an extracurricular activity.”
Causation is even more difficult to identify when predictive scoring combines multiple factors.
Predictive Analytics in Action
Eric shared some case studies of predictive analytics in action:
- First Tennessee lowered direct mail costs 20% and increased response rates 3.1%, for a 600% ROI.
- Target improved direct mail success by 15-20%.
- Premier Bankcard reduced mailing costs by $12 million.
- Life Line Screening increased response by 38%.
- Harbor Sweets, in an example of a small business using predictive analytics, earned a 40% response rate from lapsed buyers.
The three primary applications of predictive analytics today are:
- Response modeling for acquisition
- Churn modeling for retention
- Recommendations for ads or content
An ongoing expansion of predictive analytics is leveraging social media. “Social media is just more data: the row of data for each individual gets that much wider. We can predict behavior from who you know, from your social activity, and from what you like. Homophily is the tendency of individuals to associate with similar others. If you bought a product, your associates are more likely to buy that product.”
Some tips for putting predictive analytics in action for your organization:
- “The data preparation phase is arduous and takes 80% of the hands-on time of predictive analytics.”
- “You don’t need an initiative to collect additional data. Use all the data that is being naturally accumulated: sales activity and web activity is sufficient. Most organizations don’t have to buy third-party data.”
- Big Data doesn’t have to be huge. “You need a few thousand records who did the thing you want to predict and a lot more records that didn’t do the thing you are trying to predict. In B2B work, you often have less data, so you need plenty of examples. There’s no hard science for how many positive and negative examples of the behavior you need.”
In conclusion, Eric said, “Here’s my predictive analytics takeaways: leverage Big Data, predict per individual, and act on the predictions to fortify and bolster all kinds of organizational operations.”
Author Notes:
Jeffrey Henning
