The Trump campaign used data and analytics to capitalize on clues from early voting returns, including data on Latino voters in Florida, whom an undisclosed Hispanic agency helped the campaign reach.

That's according to Matt Oczkowski, who headed up the Trump campaign's team of embeds from U.K.-based data firm Cambridge Analytica. Mr. Oczkowski believes that polls failed to predict victory for the president-elect in key battlegrounds partly because many surveyed only people deemed to be likely voters because they had voted in previous elections, excluding non-voters being drawn to the polls by Mr. Trump.

"We've seen times over the course of the last month that pointed to a potential outcome like last night," Mr. Oczkowski said Wednesday.

Mr. Oczkowski, head of product at Cambridge Analytica and former chief digital officer for Wisconsin Governor Scott Walker's short-lived presidential primary campaign last year, said the Trump campaign looked at early voting data around ten days before the election and used it to update its data models for get-out-the-vote and last-minute persuasion efforts.

On Twitter earlier in the day, he suggested that the process and the Trump camp's overall approach to data was legitimized by the win that stunned pollsters, pundits and countless citizens in the U.S. and across the globe.

Data vindicated victory

— Matt Oczkowski (@MattOczkowski) November 9, 2016

For much of the election season, however, there wasn't much of a data strategy at the Trump campaign. It wasn't until July that the campaign hired Cambridge Analytica, the company that handled some data-related work for Ted Cruz's failed primary campaign and is funded by big Republican donor Robert Mercer. And its work even then was sufficiently low-key that its involvement was only speculation until it was confirmed by Federal Election Commission spending reports.

Around a dozen people from the company worked with the Trump campaign, alongside Republican National Committee staff and other vendors for a total of around 70 people. They first shared office space with Mr. Trump's digital agency, Giles-Parscale, then moved into their own offices paid for by the RNC, Mr. Oczkowski said.

Early on, Cambridge Analytica's data showed that Mr. Trump was garnering support among voters who were white, more rural, and a little younger than the typical Republican, and that issues such as wages, immigration and law and order were of key importance to them.

In the final days leading to the election Tuesday, the Trump campaign read the early voting tea leaves and isolated three key trends in the makeup of the voters: There were fewer African Americans, a slight increase in Hispanic voters, and a "big increase" in voters 65 and older, according to Mr. Oczkowski.

Working with Brad Parscale, the Trump camp's digital director and president of Giles-Parscale, the team worked "long nights" to re-weight its polling around 10 days before the election, he said.

While the campaign has only begun to look at the election results, "We suspect that the older Hispanics, particularly in Florida, there was a block of them who voted for Trump," said Mr. Oczkowski, suggesting that Latinos in Miami-Dade County, many of them older immigrants from Cuba or elsewhere who support strong restrictions against illegal immigration, played a role in Mr. Trump's narrow Florida win.

"We're going to be looking at different geographic trends to see how that played out," he added.

The campaign was helped by a Hispanic consultancy that was embedded with Giles-Parscale. Mr. Oczkowski would not name the company, and because it was paid by Giles-Parscale, rather than directly by the Trump camp, the identity of the company will not be apparent in FEC reports.

Mr. Oczkowski said the consultants helped the campaign determine how to apply its data to Spanish-language messaging and Latino voter models, distinguishing between Latino voters from Arizona and Florida, for example. "There has been a large Hispanic effort" in outreach by the Trump campaign, he said.

Cambridge Analytica provided data reports to the consultants on issues that Latinos care about, said Mr. Oczkowski.

There wasn't much of a data infrastructure to work with when the company joined the campaign. "We had to build everything from scratch," said Mr. Oczkowski. That included building voter data models, not unlike the ones that pretty much every sophisticated political campaign employs. Collaborating with the RNC's data program, the Cambridge Analytica team started out creating 15 different political models, determining important issues associated with each voter segment, and crafting Trump-specific turnout models in order to "try to understand how he differed from a typical Republican," said Mr. Oczkowski.

Though many widely cited polls don't seem to have accounted for a new crop of voters being attracted by Mr. Trump, Cambridge Analytica and a handful of others on the right tried to factor in the Trump effect.

Great America PAC, a super PAC that supported Mr. Trump, attempted to generate data on unlikely Trump voters through direct response TV advertising. Republican data firm Deep Root Analytics invested a six-figure sum to revamp its data models to reflect the mindsets of crossover voters, from free-trade Republicans who might have supported Hillary Clinton for president to Democrats seeking tariffs who could have voted for Mr. Trump.

The polling re-weighting that the Trump camp conducted based on early voting data drove adjustments to digital ad messaging, campaign messaging and even plans for rallies. Last-minute pushes in Florida and Michigan were "certainly informed" by those data alterations which showed that Mr. Trump could capitalize on appeal to voters in certain cities who were deemed highly persuadable, Mr. Oczkowski said.