Considered to be Donald Trump’s Digital Guru, he predicted Trump to win 305 electoral votes while Trump actually got 306 of the 538 electoral votes. He was one of the very few people allowed to “tweet” on behalf of Donald Trump.

How did he manage to predict that number?

Parscale says “the campaign’s ‘data operation’ ran everything from television ad buys to budget choices to where the campaign was on the ground, which he said was a first in American politics. He explained that they built models based on live calls, web tracking, web surveys and other sources, which gave the campaign the ability to zero in on undecided voters and persuadable targets. He pointed to states like Pennsylvania and Michigan, which the campaign focused on after the data revealed that Trump had a better chance there than many polls showed.”

“The data doesn’t lie, and that’s the beauty about data.” concluded Brad Scale in his interview to Fox News in November 2016.

This post is not to talk about Brad Parscale or Donald Trump but to highlight how critical it is to test such predictive models since Bad Data, process issues with data and misinterpretation of results can yield false positives and negatives, which can lead to inaccurate conclusions and wrong business decisions.

Here is our video webinar on how we approach predictive data model testing at Zuci Systems. Our framework ZUJYA is built using the same approach and helps in automated testing of predictive data models.