Donald Trump's U.S. election victory threw polling experts -- from the New York Timesto Nate Silver's FiveThirtyEight -- into serious soul-searching after their forecast of a Hillary Clinton win turned out to be streaming melayu lucahwrong.

But for a little-known British company working for the Trump campaign it was confirmation that their data analysis -- based on a hyper-targeted psychological approach -- was working.

Just a few weeks before the vote, London-based Cambridge Analytica saw an increase in favour of Trump from 1% to 3% in Rust Belt states such as Ohio, Pennsylvania, Michigan and Wisconsin.

That was due to demographic shifts of older voters and those who usually don't turn out, which are thus not taken into consideration by many polling companies.

"We're not saying we predicted it right, but we certainly noticed some trends." Matt Oczkowski, director of product for the president-elect’s data team Cambridge Analytica, told Mashable.

"Donald Trump is not your typical Republican candidate so you can't apply the same demographic of people who voted in 2012 for Mitt Romney, for example," Oczkowski continued. "Using historic norm, people who are considered likely Republican voters, was never going to tell us who's going to win in this election."

For the past few weeks, the campaign saw a narrowing in its internal poll. When early votes started trickling in, the analytics team -- of 12 people -- divided between New York City and San Antonio noticed a decrease in the African American turnout, an increase in Hispanic turnout and those over 55.

Using a real-time dashboard of data, the team swiftly updated the models to include those hidden, unlikely Trump voters and gave him a much higher chance to perform well in the Rust Belt.

The company, which also assisted the Leave.EU group ahead of the EU referendum in the UK, uses a psychological approach to polling, harvesting billions of data from social media, credit card histories, voting records, consumer data, purchase history, supermarket loyalty schemes, phone calls, field operatives, Facebook surveys and TV watching habits.

The data is bought or licensed from brokers or sourced from social media.

"We were constantly in the field, polling from different media -- online and telephone -- 1,500 interviews per state per week," Oczkowski said.

All of that is combined into 15 to 20 models who are constantly updated in real-time as the database evolves.

Demographic shift

"What gave us advantage over the other polling companies is how quickly we were able to react by updating models to take into account where the demographic is shifting," Oczkowski said.

The marketing department then identifies key voter types -- extremely individualistic targets -- and creates targeted adverts tailored to particular issues, based on voter personality and how they see the world.

"Data influences marketing, and helps them decide whether the campaign should spend money on TV or change the candidate travel schedule," Oczkowski said.

Privacy concerns

Some privacy experts have expressed concern about Cambridge Analytica's use of big data.

Christopher Weatherhead, a technologist at Privacy International, told Mashable that despite the fact that some of the data is already in the public domain, people are probably unaware that it could be used for targeted ads.

"It is disturbing that people who share their stories online, who use the internet to educate themselves and explore, are being secretly monitored by companies who are profiling them, and selling the profiles, or digital doppelgängers, to a political party," said Weatherhead.

"Is it possible to opt-out? There must be more public discussion around these issues," Weatherhead continued. "It is troubling that such detailed pictures of peoples' lives are able to be created without their knowledge that it is even happening."

Those concerns were dismissed by Oczkowski, who said that Cambridge Analytica uses a "high quality of modelling" and the data is freely available to license or purchase.