Back in 2010, after the Republicans won control of the U.S. House of Representatives, many of the pundits looking ahead to the 2012 presidential election were predicting a loss for President Obama. The president’s campaign managers knew that even in a best-case scenario, Obama probably wouldn’t be able to win the election with his existing base of supporters alone. The campaign needed a way to increase support for the president while making the most effective use of its available resources. Enter uplift modeling, a form of predictive analytics that aims to identify individuals who are likely to be positively influenced by ads, mailings, phone calls and other outreach efforts.
When an election is up for grabs, “the critical component is to make sure that the campaign is reaching out to the right voters,” said Daniel Porter, director of statistical modeling for the Obama for America 2012 campaign, in a video interview recorded at the 2013 Predictive Analytics World conference in Boston. Porter, who is now a partner at analytics services provider BlueLabs in Washington, D.C., led a team of eight data analysts who were charged with determining which voters the campaign should focus on. They turned to uplift modeling, also known as persuasion modeling. Its goal is to pinpoint the “persuadables” — in this case, voters who were leaning toward Republican nominee Mitt Romney but might decide to vote for Obama if they were contacted. With that information in hand, the Obama campaign could avoid spending money and volunteer time contacting people who the analytical models showed were already committed to voting for the president or for Romney.
“The old adage in advertising is, ‘I know half of my advertising isn’t working; I just don’t know which half,'” Porter said. “With uplift modeling, you can identify which half is working and which half isn’t — or more specifically, what customers are most receptive to advertising and what customers aren’t.”
To accomplish that for the Obama campaign, Porter’s team first set up an experiment in which some people were called and others weren’t; in a presentation at the conference, Porter said the two groups were then polled, and support for Obama was five percentage points higher in the “treatment group” than among the voters who weren’t contacted. From there, he said, the analysts used a matrix of political, demographic and household data to develop a set of predictive models that “applied a score to every voter in all the battleground states.”
After field testing and revisions, the final versions of the models were put into use in October 2012. Porter said they “guided every door knock and every phone call” in the final weeks of the campaign. The Obama campaign also used the models to target persuadable voters through direct mail, social media advertising, Facebook messages and TV ads; for the latter, he said, an optimizer was used to develop a list of “good buys” based on what channels and programs large numbers of the desired voters typically watched. The models enabled the campaign to reach out to voters at an individual level, based on what message they were most likely to be receptive to and what form of contact they were most likely to be persuaded by, he added.
And uplift modeling isn’t only for political campaigns, according to Porter. He said it could be used in a variety of marketing applications — for example, to identify potential customers who might be persuaded to buy particular products. BlueLabs also is working with a coalition of hospitals in New Jersey on an initiative to reduce patient readmissions through analytics and uplift modeling.
In this interview with SearchBusinessAnalytics Executive Editor Craig Stedman, Porter further discusses how the Obama campaign used uplift models to help win the election and possible commercial uses of uplift modeling tools and techniques. Viewers of this seven-minute video will: