According to conversations with over two dozen staffers and advertising experts involved with the data and advertising operations of both presidential campaigns, the Democratic party's data machine that worked so well in 2008 and 2012 may have ended up working against Hillary Clinton in 2016. Here's what the campaigns did:
- Like Clinton, the Democratic campaign's digital strategy was disciplined and precise. They were methodical, using data to acutely target voters that were pegged as most likely to vote or most likely to be swayed.
- Like Trump, the Republican campaign's strategy was unpredictable and opportunistic. They were experimental, especially on Facebook, using data to identify wider sets of potential voters, and to target broader audiences that weren't necessarily pegged as definitive or persuadable voters.
Why it matters: The Clinton campaign's data-driven media and communication tactics may have been too precise amid an unusual media climate against a non-traditional candidate. Conversely, the Trump campaign's reliance on intuition over data drove their message to people and places they would've never otherwise reached, like white, male voters in Wisconsin and Michigan — the voters that delivered his victory.
Per a former Clinton party official:
The Clinton campaign was run like a management consulting firm. The Trump campaign was run like a family business.
- The Trump campaign's main data source came from the RNC, which worked with a third-party firm called Data Trust. The relationship between the RNC and an independent group, not a campaign, is what many from the Trump campaign credit for their electoral success. After their 2008 loss, Data Trust and the RNC began to create an infrastructure that could support any future nominee. "Our theory is Democrats created a machine designed to market one product — Obama. If we can design a system that can market any candidate, not one specific candidate, we think we can leapfrog the competition," Data Trust advisor (and former Reince Priebus chief of staff) Mike Shields said about their strategy.
- The Clinton campaign's main data hub was split between the DNC and the campaign itself, not a third-party that could cultivate data across cycles. Groups like Catalist, Civis, Organizing for Action and the DNC all housed different pieces of the data at different times that Obama used to target voters in 2012. Sources say the Clinton campaign worked with the DNC to cultivate and model a data pool that had been fragmented after Obama's 2012 win, making it harder for the Clinton campaign to compete with the GOP's long-term strategy.
Targeting strategy: One of the biggest differences between 2012 and 2016 was the increased domination of Facebook and Google, which made micro-targeting through paid advertising on both platforms a significant part of each campaign's strategy.
- The Clinton campaign used micro-targeting to focus on a specific group of persuadable voters, while the Trump campaign identified persuadable voters and advertised to broader, related groups.
One of biggest lessons of this campaign is you can't get by with small, targeted audiences.
— Google's Head of Industry - Elections, Lee Dunn
- Trump: Gary Coby, who led the Trump Campaign's advertising team alongside Brad Parscale, said that each day, the campaign tested 40,000-50,000 automated ad combinations on Facebook for $200,000-$300,000. From there, they found which messaging attracted audiences whose voter files weren't pegged as being likely to vote for Trump. Experimenting that quickly allowed them to build up enough historical data to very quickly identify trends of which ads worked and which didn't. Coby told Axios that campaign staff got so good at predicting effectiveness of certain messaging, that they could see what worked after only spending $20-$50 on a particular ad.
- The Clinton campaign took a more strategic approach, focusing on fewer, more targeted messages, that were more likely to appeal to certain voters and refining them through surveys on video ads across the web. The campaign used technology that was able to measure specifically how messages resonated within different voter files, and drill into subgroups to identify areas for future message optimization. The technology even included ways to measure how external events affected responder bias.