One of the defining features of the 2016 US election was the widescale use of digital ads. Whether it was through Russian money, bot accounts, or fake news, digital advertising undoubtedly played a significant role in determining how people voted.
There's no reason to think that digital advertising won't play the same sort of role in the upcoming 2020 election. But four years is a long time in the digital world, and plenty of things have changed since 2016. With that in mind, how can we expect political advertising to change in the run up to 2020?
For all the talk about the role of digital ads in US politics, digital spend still makes up a relatively modest percentage of most campaigns' budgets. Data from the 2018 Senate elections showed that the average Democrat candidate's campaign spent 7% of their budget on digital media.
Compare this to the huge 44% of the Trump 2016 campaign budget that was spent on digital ads, and 7% is practically nothing. As we approach 2020, politicians and their parties will likely look to the success of the Trump 2016 campaign as a reason to increase their investments in digital.
Most of this spend is likely to fall into the hands of Facebook, after they've remained steadfast under pressure to curb political advertisements on their platform. While Twitter have banned political advertising outright, and Google restricted the targeting options available to political advertisers, Facebook has made no changes to its policies.
The ability to freely target political ads on the Facebook Ads platform means that it'll likely take an even larger share of digital budgets than it has in recent elections.
Though I expect that that many campaigns will double down on Facebook, due to its accommodating approach to political advertising, the best campaigns will still use a diversified channel mix.
In particular, the most effective campaigns will also make use of Snapchat, which offers a more effective way to reach audiences under the age of 35. Snap's ad revenue per daily active user, which is a measure of how competitive a platform is, is several times lower than Facebook's. This is reflected in the generally much lower ad prices available on Snapchat.
All this means that advertisers looking to target younger voters will be able to purchase significantly more ad inventory by putting some of their budget into Snapchat Ads. They'll be helped by the fact that Snapchat doesn't impose any restrictions on the targeting of political ads.
One feature which Facebook have developed significantly since 2016 is what they call ad studies. Ad studies allow you to run a range of complex measurement tests on Facebook, which generally aren't available on most other platforms.
One example of an ad study is what's called a brand lift test. This is a test which is run as part of a particular advertising campaign, and which randomly splits your target audience into people who are excluded from seeing your ads (the control group) and those who are allowed to see your ads (the treatment group).
During the life of the advertising campaign, Facebook polls both groups with questions about your brand. If you were Coca Cola, the questions might look something like what do you think about coca cola? or how likely are you to purchase a coke?
Facebook is then able to compare the differences in responses at the end of the campaign, and attribute the difference to the ads that were run. For example, if people who saw Coca Cola's ads were 10% more likely to say that they'd purchase coke than those who didn't see the ads, then the ads were responsible for that 10% brand lift.
In political advertising, where the objective is to change opinions and beliefs rather than to generate clicks, such measurement strategies could prove invaluable. By learning how different messages affect voter perceptions, political campaigns can double down on the most effective ads.
Don't get me wrong, political ad testing is nothing new, and the Trump campaign did it at a huge scale in 2016. The difference is that back then, their ad buyers would have looked at metrics like click-through-rates, or likes, as a way to judge ads. These are both ultimately proxies for effectiveness; brand lift studies bypass the proxies and let you tap straight into user psychology.
It's not just the way that political advertisers are running tests that will change, but also how many tests that they're able to run. Many of Facebooks new ads features over the past year have been built with testing in mind.
Chief amongst these is a product which Facebook calls dynamic creatives, which lets advertisers test up to 6,250 ad variations in a single ad. Facebook will rapidly comb through the different combinations to find which are most effective.
The 5.9 million ad variants that the Trump campaign allegedly tested in 2016 may sound like a tough record to beat, but you'd only need less than 1,000 dynamic creatives to test that many ads. With this new tool in the hands of political advertisers, you'd expect the rate and scale of testing to continue increasing as we approach November 2020.
For all the drama and headlines following the Cambridge Analytica scandal, not much has changed.
Back in 2016, Cambridge Analytica used a dataset belonging to Aleksandr Kogan to psychographically profile Facebook users. Kogan, a research associate at Cambridge University, had created a personality quiz which users had to sign into via their Facebook accounts to receive their results.
By signing in with their Facebook account, users granted Kogan access to behavioural data about not just them, but also their friends. This included details on what pages they liked, and what groups they were members of; key pieces of personal data which allowed Cambridge Analytica to build out complex user profiles, which could then be used to target ads.
The most concerning privacy flaw that Kogan exploited was the ability to collect data on a user's friends, without those users consenting. This flaw is no longer exploitable, hindering the ability for researchers to collect this sort of data at scale, but that hasn't put an end to data mining.
Anyone can build an app which requires a user to log in via Facebook, and request access to their page likes or group memberships in the process. Once a campaign has access to that data, they can use it to create profiles of users, like Vote Leave did in the 2016 UK Brexit referendum.
Vote Leave accomplished this by running ads which offered users a chance to win £50 million if they could guess every result in the upcoming Champions League season, a task with near impossible odds. In order to enter, users had to supply some basic information such as their name and date of birth, and also how they intended to vote in the upcoming referendum.
This gave Vote Leave a huge trove of data on undecided voters, some of the hardest to find (but also most important) voters in any election. They could then model the data themselves to work out how to target more undecided voters, or upload the data to Facebook who would then do it for them.
This latter option uses what's known as a lookalike audience. If you upload a dataset of customer details to Facebook, Facebook can then model the data and find its users who most closely resemble the original users. In this way, Facebook can help campaigns find hard-to-reach groups, so long as they have some initial data to go off.
One key tool from the 2016 election was a Facebook feature known as audience insights. It allowed you to upload lists of user data, perhaps users who are undecided, or who represent a particular minority. You could then see everything that Facebook knows about these users, and how that compares to the general public.
For instance, if you had a list of undecided voters' personal information, you could upload this to Facebook and see what pages they're more likely to follow, or whether they're more likely to speak a particular language. These insights can then be used to inform a campaign's targeting and messaging strategy.
After the Cambridge Analytica scandal broke in 2018, Facebook remove the ability to use audience insights with lists of user data. No longer could you upload a list of people's information and see what Facebook knows about them.
This vastly reduced the power of the audience insights tool, but didn't render it useless; advertisers can still see insights on people who follow their pages.
Because of this, advertisers can treat Facebook pages as datasets of people, and build up their following with whoever they want to profile. If a malicious campaign wanted to profile people of a particular minority, they could run ads to that minority encouraging them to like a page owned by the campaign, and then use that page's followers as a data source in audience insights.
This isn't dissimilar to what some of the Russian ad accounts did in 2016. Data submitted to Congress showed that Russian accounts paid to promote meme pages appealing to specific minorities; in particular African Americans and LGBTQ people. If such ads were to re-appear in 2020, it's likely that they would be attempting to build page followings for audience insights.
2016 wasn't a one-off, digital political ads are here to stay. With the changes that different platforms have made regarding political ads in the last six months, it's likely that more and more ad spend will find its way to Facebook. And with all the new features that Facebook has released since 2016, the opportunities and dangers of political advertising are greater than ever.