The Election in Data and Software

Nick Timothy so on Conservative Home:

Ironically, the Prime Minister is the one political leader who understands this division, and who has been working to address it since she became Prime Minister last July. The Conservative election campaign, however, failed to get this and Theresa’s positive plan for the future across. It also failed to notice the surge in Labour support, because modern campaigning techniques require ever-narrower targeting of specific voters, and we were not talking to the people who decided to vote for Labour.

Do modern campaigning techniques require ever-narrower targeting of specific voters? Or is it more that the technology makes it possible, and given the hammer, the world fills up with nails? The advertising trade has a fundamental concept of a trade-off between reach – the number of people you can show a message to – and richness – the quality or elaboration of the message you deliver. Usually, the more elaborate the message, the harder it is to deliver. Imagine a spectrum from cinema advertising on one end to e-mail spam on the other. The era of high TV was also the era of high advertising, if you will, for good reason – TV ads could have both near-cinematic richness and an enormous mass audience, to some extent bucking the trade-off.

Richness is also a form of targeting. The more sophistication you build in, the more people you turn off. This is often deliberate – the iconic Tango Man ads of the 90s, for example, were designed to get much of their effect from baffling your parents and trolling moralising authority figures into getting outraged. Also, richness and targeting both serve the same purpose. It is hoped that a higher degree of engagement and a resulting higher response rate will more than compensate for the reduction in reach. This is crucial, because targeting by definition means sacrificing reach. Delivering ads to targeted individuals means not delivering them to non-targeted individuals.

So how did that work out for you? WhoTargetsMe built a browser extension to instrument what Facebook ads were being served to who.

Geographically, then, Labour’s Facebook campaign was achieving 2.2x the reach of the Tories’. British parliamentary constituencies are designed to be of roughly similar population, so this should be a reasonable proxy for reach.

On Election Day itself, Buzzfeed ran this highly significant story using their own FB instrumentation. They concluded that the most shared stories on Facebook in the immediate run-in were overwhelmingly pro-Labour.

Facebook advertising basically gives you two things – targeted delivery to a big audience, and the potential for viral spread. It’s worth understanding here that the key development in its history as an ads company, which took it from being a loss-maker to being the cash machine it is now, was when it started serving fewer ads. This decision to ration ad space was taken in order to make the app more mobile-friendly and also to rebalance the mixture of content (i.e. what you come for) and ads (i.e. what they come for). Doing so implied an increased emphasis on targeting and richness, because fewer ad slots were now available. It also permitted Facebook to charge more for advertising and to increase its profit margin, because the changes did not increase its costs (if anything, they reduced them).

This creates a complicated relationship between the two poles. Sure, you can drop exquisitely targeted ads and you can load them with HD video or even interactive gaming. But this comes at a price – a price in terms of cash, as you pay through the nose, and a price in terms of opportunity, as there will always be many fewer available ad slots than there are posts that could potentially be filled by shared messages. The optimal solution would be a targeted drop that then goes viral, but this begs the question. Everyone wants to have their cake and eat it. Another strategy, as a reader suggested to me, would be to run a lot of different targeted campaigns; but this has the problem that they might compete with each other.

The Tories designed a campaign around “strong and stable” and Brexit, and planned to execute it as a succession of targeted national ad drops and events, managed from the fourth floor of Central Office where a tiny command group consisting of Nick Timothy and Fiona Hill plus hired guns Mark Textor, Jim Messina, and Lynton Crosby held sway and controlled the keys to the ad-buying account. This was air-war politics par excellence. The central message would be delivered to people who voted UKIP in 2015 or the EU referendum, preferentially in Labour seats in the North and Midlands.

In the event, though, this didn’t work. For a start, the Tories failed to win the ex-kippers by the margin they hoped for. In many seats they broke 50-50. It was only in seats with very high Leave % where the Tory strategy worked. The problem may have been that ‘kippers in these seats were more likely to be ex-Tories, while elsewhere they were more likely to be generic protest voters. Labour canvassers reported significant numbers of people who voted Liberal Democrat in 2010, UKIP in 2015/2016, and Labour in 2017 (“S10, U15, L17“. This may be a case of being fooled by data – the data in question being Chris Hanretty’s celebrated mapping of Leave/Remain votes to parliamentary constituencies, which made the assumption that Leave votes were UKIP votes. The Tories added a corollary to this assumption – UKIP is a right-wing party, so are we, therefore Leave voters are Tories struggling to get out.

Secondly, the Tory strategy created a highly unfavourable electoral map. Much of the Tory effort was targeted on relatively safe Labour seats in the North and had the effect of shaving the big majorities there, while the Labour effort was targeted on marginal Tory seats they could flip. Interestingly, the parties’ strategies remained fixed through the campaign, but Labour’s turned out to be designed for a national swing quite close to the one that actually happened. Jeremy Corbyn campaigned in seats that Labour would flip on a swing of 1 to 5 per cent, and also campaigned in relatively safe Labour seats to work up the turnout and generate buzz in the regional media.#

The BBC’s Chris Cook doesn’t believe this was deliberate, but I wonder. The Tory strategy sounds very much like an effort to replicate Donald Trump’s campaign out of a can – a deployable package of tactics, techniques, and procedures that could be launched from anywhere with a Facebook ad-buyer account and a (platinum) credit card. Yet Trump’s campaign benefited hugely from earned media and virality. Famously, he didn’t even bother buying TV ads until very late in the day. Which brings me to this story.

two pieces of obscure software developed by Labour HQ are widely acknowledged to have played a significant role.

The first helped turn a swollen base of activists into proper campaigners. Called Chatter, it allowed Labour’s growing base of activists to have proper text exchanges with people they canvassed, rather than dispatching them blunt, campaigning messages. “It armed campaigners with the ability to actually make people feel like they were being listened to on a local level,” said a senior Labour figure.

The second was the closest thing Labour had to a secret weapon. Over the last year the party developed a tool called Promote. Its effect was to unleash the power of Facebook advertising to local parties across every constituency. The tool combined Facebook information with Labour’s voter data, but allowed senior activists and candidates to use it to send locally based messages to the right sections of their electorates. Labour is said to have spent heavily through the tool.

“People were seeing stories about their school and hospital, not just national messaging like the Tories were doing”

This might explain why the Facebook strategy worked so well for Labour and so poorly for the Tories. Promote offered decentralised control of the Facebook campaign, letting local candidates release their own messaging. This is also a way of getting the best effect from the rationed pool of ad space – you’re more likely to respond to something that’s actually relevant, so this both makes for virality, and avoids the problem of multiple targeted campaigns cannibalising each other.

Meanwhile, a critically important use of data is to inform your own expectations. Somehow, Labour went into the campaign knowing what kind of swing might be achievable and that it was possible to drive up turnout. This might be the consequence of Chatter, and of the huge activist base.

3 Comments on "The Election in Data and Software"

  1. Allegedly Labour HQ defunded some marginal – which is another datapoint to add to the analysis above – especially in light of the fact that there were very large swings in otherwise safe seats. My own constituency swung by 9% converting a very comfortable Tory safe seat into one where they won by a couple of thousand votes.


  2. Very interesting piece. Key I think is the interlocking between FB and other efforts, which is another area where the Tory centralisation had a theoretical logic that wasn’t completely wrong, but fell down in practice. I have some other thoughts to bring to the table, but I think I’ll have to blog them.


Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.