PredictWise Audiences: Real world outcomes that can decide the election!

Those of you who follow PredictWise regularly will know that PredictWise Audiences achieved much better return metrics in Katie Porter's CA-45 district when put to the test against a rival Audience provided by a consulting group that shall remain unnamed. In short, our Audiences generated 2-4x lift in crucial return metrics such as video through plays and comments. But, these metrics are hardly indicative of actual persuasion. And, the same can be said about within-platform testing: While brand-lift is indicative of some change, it is unclear whether effects are an artifact of taking surveys in the same laboratory-like setting in which ads were consumed, and how much decay we should expect.

PredictWise now tested lasting attitudinal effects of ad campaigns targeted on the basis of our Audience Technology, together with Acronym. First, we identified targets that are persuadable around healthcare, i.e. soft Republicans with high levels of political tolerance who have a progressive stance on healthcare, on the basis of PredictWise baseline survey and device data gathered over the last three years from Millions of American cellphones. Second, we keyed this Audience by Mobbile Ad IDs (MAIDs). In short,  MAIDs are the most persistent, individual identifier of the digital age, ascribed by Google/Android and Apple/iOS directly to your smartphone. PredictWise baseline data – data on 250 Million Americans on 250+ pscyhometrics, policy preferences, economic and electoral attitudes – are directly tied to these MAIDs, meaning that our audience data can be onboarded into any digital platform at virtually zero loss – read more below!

Third, we used our Random Device Engagement sampling technology to survey the identified Audience on their cellphones, and a comparable control group, before treatment. Then, we onboarded 100K MAIDs into Facebook, blasted them with paid content over a treatment period of two weeks, and again surveyed respondents directly post treatment, and once month out, to gauge effects.

PredictWise Audiences make the difference

In short, the results are stunning:

  • Recall effects are big immediately after treatment, with 8 percentage points gain in Michigan, and 11 percentage points gain in Pennsylvania. As expected, recall completely decays one month post treatment
  • Knowledge effects on coverage of pre-existing conditions (creative #2) are large: We see a comparative 6 percentage point gain in both Michigan and Pennsylvania immediately after treatment, i.e. the effect of ad exposure on knowing that no Republican healthcare plan reliably covered pre-existing conditions was 6 points. And, effects do not decay at all: One month after treatment, we still see most of that lift (5 points in Pennsylvania, 6 points in Michigan). Creative #1, covering healthcare costs, was less effective. The effect of ad exposure on knowing that healthcare costs have gone up during the Trump administration was 1 point in both Michigan and Pennsylvania. Both effects decayed
  • We see large effects when it comes to drop in presidential approval: the effect of ad exposure on drop in presidential approval was 3 points in Michigan, and one point in Pennsylvania. Not surprisingly, both effects decay.
  • While we see no initial comparative drop in presidential issue approval in Michigan (treatment effect: 0 ppt) or Pennsylvania (1 ppt), effects grow over time: One month after treatment, the respective treatment effects are 2 percentage points in Michigan and 4 percentage points in Pennsylvania

In short, while ample scientific literature shows that randomly targeted digital ads do not move the needle much, ads targeted on the basis of PredictWise Audiences data can have a massive, lasting impact, and could very well be the difference between loosing and winning elections!

[Note: all results are presented as difference-in-difference Average Treatment Effects.]

Addendum: Using digital identifiers to increase match rates

As mentioned, PredictWise ID Resolution builds on our core data keyed to MAIDs. MAIDs can be directly integrated into most demand-side platforms (DSPs), governing ad real estate in the digital world. We compared the lift of using MAIDs as opposed to regular PII (personally identifiable information) across 4 states, showing that PredictWise ID Resolution can achieve a 25 points increase in match rates, on average!