It was late afternoon on Oct. 5, two days before a leaked "Access Hollywood" recording of Donald Trump boasting about his sexual exploits would erupt into scandal. Key Trump campaign players Kellyanne Conway, Jared Kushner, Steve Bannon, Dave Bossie and others sat in a conference room at Trump Tower overlooking Fifth Avenue. The mood was calm, the discussion orderly and methodical. Trump staff including Digital Director Brad Parscale had brought in a handful of media vendors and data analytics wonks hired by the Republican National Committee to talk campaign data and media strategy. Yet, beneath the business-like veneer of the meeting was a palpable sense of urgency. Really, this media strategy huddle should have happened months ago.

As it happened, it may have been precisely the Trump forces' unpreparedness to wage a data-driven marketing war that let data help them win on election day.

When the dozen or so attendees emerged from Oct. 5's remarkably argument-free discussion back into the campaign's 14th-floor office space, where empty pizza boxes and Trump/Pence paraphernalia seemed out of place in the gilded tower, they had agreed on a process for using regularly-updated voter data to determine media plans and voter contact in the final stretch.

There were three especially important voter segments they'd need to swing towards their candidate: unallocated voters, defined as people who were predisposed to listen to the Trump campaign and the GOP's message; "DJT Underperform" voters, or Republicans still unconvinced about supporting Mr. Trump; and "HRC Change" voters, defined as people leaning toward Hillary Clinton but also craving change in government.

Unlike the groundbreaking Obama 2008 and 2012 campaigns, there wasn't a big team of people on the Trump campaign crunching numbers in some mysterious data lab. In this case, the people doing most of the data modeling and voter scoring – especially for field operations, voter contact and television advertising -- were from a collective of three data firms hired by the RNC: TargetPoint Consulting, Causeway Solutions, and Deep Root Analytics, which officially worked with the RNC through a new subsidiary called Needle Drop.

A fourth data firm, Cambridge Analytica had been brought in by the Trump camp in August and mainly handled digital persuasion advertising and targeting. The significance of the company's involvement in overall data strategy has become a point of contention, but it's clear that Cambridge Analytica, a slick London-based outfit that by then had rubbed a number of people in the beltway the wrong way, was not welcomed into the inner circle of RNC leadership. According to insiders, the British data firm was championed by Mr. Bannon, the executive chairman of anti-establishment website Breitbart who joined Mr. Trump's campaign as CEO and is now a strategist for the future Trump administration.

The Trump camp had a few staffers hired to manage email, fundraising and supporter data it generated, but even in late summer, its capabilities were in question. It had only begun using its email list for fundraising in June, and its first-ever TV spots didn't run until mid-August. Internally, the campaign had no real data infrastructure -- nothing that could come close to what was required of a modern-day presidential campaign for targeted voter contact, anyway.

But as it turned out, that lack of sophistication proved advantageous. It set the stage for quick integration of a data foundation that the RNC had fortified since 2013. As autumn rolled around and the party stepped in to work closely with the Trump campaign, it arrived with its own far more cohesive media and field operation ready to go. Had Jeb Bush or Marco Rubio been nominated, for instance, their campaigns would have shown up with established systems, hardened into place during the primaries, and people at the helm more likely to butt heads with the RNC and its chosen team.

A necessary gamble

RNC Chief of Staff Katie Walsh, a go-getter in her early 30s, joined Mr. Parscale, an imposing figure with a reddish beard, in his office in Trump Tower during those final couple months. They would pore over state-by-state voter polling data tacked to the wall. Or each week, as Ms. Walsh nibbled on the Trump Tower Bar's chicken tenders and fries she ordered often, they'd listen in together as execs from TargetPoint and Causeway -- the ones constantly refreshing survey data and updating the RNC database voter scores -- called in with fresh data nuggets.

"He saw a machine that was pre-built over here at the RNC, and Brad and I took a bet on each other," said Ms. Walsh, a rising star in the GOP whom several insiders interviewed for this story credited with shepherding the party's data and analytics forces into a streamlined operation.

"The RNC had built a data and voter contact program that was really plug-and-play ready for the next candidate," said Mr. Parscale, the co-founder of Giles-Parscale, the San Antonio-based digital shop that the Trump camp chose to work with early on. Some of the campaign's cogs remained with his agency in Texas, where its production team cranked out digital videos and Facebook ads, and where digital staff from the RNC and Cambridge Analytica were embedded. Mr. Parscale occupied an office in the same Trump Tower suite as Mr. Bannon, Ms. Conway and Mr. Bossie.

(There was also a Cambridge Analytica representative of sorts based in Trump Tower: Laura Hilger, hired from the firm's parent company SCL Elections for a two-month stint by Giles-Parscale, served as a temporary data and research assistant to Mr. Parscale, helping him compile and evaluate data from multiple sources.)

With a mandate from its Chairman Reince Priebus in 2013, the RNC had been gearing up its technology and voter data infrastructure, which was roundly panned following the party's presidential defeats in 2008 and 2012 to Barack Obama's digital- and data-centric powerhouse. TargetPoint had played a recurring role in the voter data side of the process. By the time the Republican party convention was held in July, the firm was conducting 1,000 to 2,000 survey interviews per week with voters in about 22 states for the RNC. The company worked hand-in-hand with Causeway, which helped manage and audit the data, and updated the rating-like voter scores in the RNC's database, which are used to gauge likelihood to support the party and its candidates.

"We'd been building this massive database," said Brent Seaborn, managing partner at TargetPoint. "We kind of kept sitting on this data, waiting."

Bad blood

It's important to remember, especially as Republicans now court the president-elect ahead of his inauguration, that Mr. Trump's antagonistic approach during the primaries had left many party insiders and leaders with a sour taste. Full party support for him seemed wobbly, and it remained unclear whether the Trump campaign would trust the RNC enough to strap on its pre-fab data apparatus. Plus, the Trump camp had already begun working with Cambridge Analytica, which had worked with Ted Cruz's failed primary campaign and garnered a lot of media coverage early in the primaries for its psychographic approach to voter targeting.

But the party had little confidence that the firm, which lacked American presidential campaign experience, had the proper set-up to handle data operations for contacting voters for canvassing and get-out-the-vote, or for smart TV ad targeting.

"I think at some point it became like a financial carrot," said one exec from the RNC's core group of data firms. "If the RNC and the campaign were going to raise money jointly, the RNC was like, 'You've got to use the people we think are necessary and you've got to plug into our infrastructure."

Around September, Cambridge Analytica's TV planning and targeting capabilities were put to the test by the RNC and the Trump camp. Multiple insiders from the Trump campaign, RNC and other firms said the widely-reported $5 million spent with the company that month was not for data at all, but instead funded a TV buying trial. The results, insiders said, prompted the RNC to push the Trump camp to use its preferred data and buying firms instead of Cambrige Analytica for TV.

Some say the RNC's leadership was biased against Cambridge Analytica from the start, perceiving the U.K. firm, relatively new on the scene in the U.S., as a threat to the data systems and processes the party had been prepping for 2016. Whether the role of Cambridge Analytica was merely ignored by RNC leaders or aggressively suppressed is unclear. There were no Cambridge Analytica representatives at either the Trump Tower meeting in October or an earlier strategy meeting at RNC headquarters in late August attended by TargetPoint, Causeway and senior RNC and Trump campaign staff.

According to the RNC and others involved closely with the data work supporting the Trump campaign, Cambridge Analytica staff embedded with Giles-Parscale in San Antonio focused on digital advertising for persuasion and fundraising, combining its data and the RNC's data -- both updated regularly based on voter surveys -- for ad targeting. According to Mr. Parscale, the two databases were combined, de-duplicated and brought online for custom audience targeting on Facebook and other sites, for example.

While the RNC data encompasses the entire electorate, Cambridge Analytica developed models only for 17 key battlegrounds. Some campaign insiders said Cambridge Analytica's data models were employed for other purposes, including informing where the candidate held rallies.

The company will continue selling its TV audience targeting services as part of its move into the corporate market, from its recently-opened Manhattan office on Fifth Avenue, just ten blocks down from Trump Tower. Some observers believe the firm's ultimate goal all along has been to pivot into the far-more-lucrative world of commercial brand marketing, once bathed in the fresh glow of a presidential win.

"We're looking to focus a lot of our attention on our commercial business this coming year," said Matthew Oczkowski, head of product at Cambridge Analytica, who is preparing his own move to New York. "Our intention in politics has always revolved around trying to elevate the data science offering and to be as helpful as possible. We plan on continuing to do so."

Targeting by TV shows, not households

In September, the RNC's Ms. Walsh was able to bring in her preferred TV data consultancy, Deep Root Analytics, completing the data firm triad the party and the Trump campaign relied on heavily in the final stretch. Working in concert with TargetPoint and Causeway, Deep Root updated its TV ad targets based on changes to the voter scores, assessing the number of votes expected in each county and translating those results into a media market framework, determining the amount of ad rotation necessary to meet predefined goals among those fluid voter segments.

That information was used when National Media bought TV ad time during programs watched by people sharing characteristics of those targeted voter segments. It was not, for instance, used for addressable TV ads aimed at specific households through set-top boxes. The Trump camp did not do addressable TV advertising, according to Mr. Parscale.

Deep Root also helped the Trump camp and RNC decide what ad messages would perform best for targeted voter segments in particular markets while National Media did the actual TV ad buying.

Based on TargetPoint's weekly surveys, for example, if 50-something males in Wisconsin were waning in their support for Mr. Trump, the firm would reconfigure its data models, affecting scores in the RNC data for voters who shared similar characteristics. That change might result in recategorization for that voter sub-segment, possibly affecting media buys and creative decisions.

At the same time, Cambridge Analytica was updating its own models to help determine the best messages to aim at voter segments for last-minute digital ad targeting.

Between January 2015 and November 2016, the RNC paid TargetPoint $4.2 million for data services, and gave Causeway around $500,000 in that time, according to Federal Election Commission reports. Deep Root, acting as Needle Drop, was paid $983,000 by the RNC.

Cambridge Analytica was paid a total of $5.6 million by the Trump campaign, the cost of the $5 million TV buy and $600,000 in data services, according to FEC reports and Mr. Parscale.

No matter what twists and turns the election took, even after the "Access Hollywood" tape scandal broke in early October, the campaign kept its ads up in rust belt states Michigan, Ohio, Pennsylvania and Wisconsin -- all states it won. In fact, the analytics-driven approach led the Trump camp to place ads in 13 markets that Mitt Romney's 2012 campaign never bought, including Flint, Mich.; Greenville, N.C.; and Tallahassee, Fla.

"It was out of necessity," said Deep Root Analytics CEO Brent McGoldrick, discussing the 270 electoral votes needed to win. "It was like, 'look if you're going to get to over 270 you're going to have to flip at least one of those states behind the blue wall, at least one of them," he said. "Any other campaign I've ever been with, there would have been a major reshuffling of the advertising and advertising budget at that point [after the Access Hollywood tape was revealed]. Someone in the campaign would have emerged as a nervous Nellie. That didn't happen in this case."

The data also informed development of ad creative and how it was aimed. In the final weeks, ads such as one called "Deals," with a tougher, masculine tone portraying Mr. Trump as a strong leader who would renegotiate "bad trade deals pushed by the Clintons" ran in smaller rust belt cities. In Toledo, "Deals" showed up during Nascar Camping World Truck Series, a pickup truck racing series, in the hopes of reaching male HRC Change voters.

To convince "HRC Change" women, the data indicated a softer tone was needed. An infographic-style ad called "Builder" -- one that looked very similar to those the Obama campaigns had used -- ran in markets including Milwaukee during The Ellen DeGeneres Show. "It takes a builder to rebuild the American Dream," said the ad, which promoted the candidate's promises to provide childcare tax reduction and paid maternity leave.

699,146 Florida voters

On Thursday, Nov. 3, a data set encompassing 699,146 undecided Florida voter profiles was provided by Causeway to the RNC for distribution to the party's field operation, the people guiding the labyrinthine door-to-door canvassing process to get out the vote, who would translate the data to walking maps for volunteers and, in some states, paid GOTV canvassers. The same data went to the Trump campaign and its direct mail, TV and other media vendors. Over the previous weeks, similar data sets for Florida and other key states, featuring updated targets for voter contact and all media, had been disseminated. The goal was not unlike that of omnichannel marketing in the corporate ad world: essentially to use the same playbook for all communications with individuals no matter what the medium.

There were five days to go till election day and the RNC's data coalition saw inklings of a Trump lead in Michigan and Ohio, with Florida closing in for him. Ultimately, he won all three states -- Michigan by only around 10,000 votes, Florida by approximately 113,000, and Ohio by a wider margin of around 445,000 votes.

Cambridge Analytica, which also used ongoing surveys to develop and update its own models, saw older Hispanics in Florida and North Carolina coming through for Mr. Trump in its own data.

In the last week or so before election day, Bill Skelly, partner and co-founder at Causeway, watched as the data showed accelerated voter migration into Trump territory. "I don't know that I've ever seen anything like it from a data perspective," he said.

So much of the election was driven by earned media coverage that it's important to step back and consider whether well-targeted paid media and door-knocks actually made the difference. No winning campaign could rely solely on data-driven paid media, which, amid the roar of earned-media messages especially in 2016, "is like a ceiling fan in a hurricane," said Deep Root's Mr. McGoldrick.

But good data is essential nonetheless, he said.

"Trump won by 100,000 in key blue wall states, so, yes, it helped," Mr. McGoldrick said. "It's required to win as part of an overall messaging strategy."