Twitter is a unique platform to discover what’s happening, anywhere in the world, in real-time. 

 The fast-paced nature of the platform poses an issue for our marketing strategy: how do we predict the big news stories of any given day and serve these to users, showing Twitter to be the best source for live content, no matter your interest?

To do this, we needed to work with Twitter to build a solution to make their marketing as real-time as Twitter itself. We needed to reach new users and drive up active users on the platform.

In 2018, we developed a proof of concept within Paid Search to tackle this problem. Using Twitter’s API and machine learning to create search ads in real-time based on Twitter trends. It performed extremely well, so we began developing the prototype into a full, multi-channel campaign management tool, using AI to drive personalised, coherent user experiences across the user journey, based on high-quality conversations happening on Twitter. We call this tool Magpie.

Magpie identifies the best trending content on Twitter and uses machine learning to analyse it and build creative and write messaging on the fly to serve to users interested in the content on other channels. For each user, it creates a curated experience by serving a collection of ads, across channels on topics that are of interest to them, based on their unique interests, behaviours and demographics.

All of this happens in near real-time, meaning Magpie is running an always-on, multi-channel campaign that creates and serves thousands of personalised ads every day, creating personalised user experiences at scale, with minimal human intervention. This is the first time AI and user-generated content has been used to run a real-time digital campaign before. 

At any given moment, the conversations happening on Twitter are powering a campaign that’s reaching new users off-platform and engaging with them.

Every hour, Magpie pulls the UK’s top trends and fetches the top Tweets. It analyses these conversations and uses machine learning. For every conversation that happens, Magpie identifies the sentiment being expressed, the topic being discussed, and uses this to turn user-generated data into structured data that can be manipulated at scale.

From here, Magpie then assesses whether a topic is brand safe, based on the signals it’s amassed. We’ve worked to fine-tune Magpie’s abilities over time, training our models to cope with thigs like Retweets, hashtags in comments, and even sarcasm.

Once a trend is identified as brand-safe, Magpie again uses historical data to identify which channels the trend will perform best, which audience to reach, and how to speak to them. It assembles channel specific creative, crafting natural sounding ad copy based signals like sentiment and category, and the audience being reached. Ads are automatically trafficked to the media buying platforms and mapped to appropriate audience targeting to reach relevant users. 

For each ad, Magpie identifies the optimum landing page that it predicts will drive the best experience for the user. All of this happens in real-time, 24/7.