"At the 2008 and 2012 elections, the Obama campaign created fine-grained statistical models predicting both the likelihood of any single voter turning out to the polls, and their likelihood of casting a vote for their candidate.
As a ProPublica investigation found, these voter profiles were created using public voting records, consumer data, and canvassing by the campaign itself. Upwards of 1,000 different points of data may have been collected about a voter depending on their significance to the campaign, with the profiles naturally being more detailed in key battleground states.
This meant that campaign operatives knew, for example, if an individual Floridian or Ohioan had voted in 2008 but not 2004, where they shopped and what their purchasing habits were, and how they felt about a huge range of social and political issues.
Specially-created algorithms then trawled through this data looking for patterns that would identify voters as Republicans or Democrats, strong or weak party supporters, and voters or non-voters.
This allowed Obama's team to tailor their election pitch to individuals; in some states the campaign even went so far as to buy set-top box information from cable TV companies so that it could track the viewing habits of key voters and target its advertising spend accordingly."