Last week Monday, the day before the US headed to the polls to vote, South African Jean Pierre Kloppers made a prediction to BBC reporters that defied what most analysts and polls were forecasting globally. He announced that US Republican presidential candidate Donald Trump would win the election. He was one of few to see it coming.
Kloppers is the CEO of BrandsEye, a Cape Town-based media-monitoring company that uses data to help clients better understand people’s sentiments about an issue or brand. One of the ways it does this is by collecting masses of social media posts. To ensure that the nuances and intentions behind words are accurately picked up and interpreted, BrandsEye crowdsources thousands of individuals across the world to go through the data. Because it relies on people, the system can detect expressions like sarcasm, which a machine might miss.
“We can get down into the nuances of how people feel. And in an election, that nuance matters,” says Kloppers.
The method was also used to successfully predict the UK’s vote to leave the European Union in the Brexit referendum, as well as the results of South Africa’s recent municipal elections, where opposition parties took control of key metropolises.
Succeeding where traditional polls failed
To successfully call Trump’s win, BrandsEye collected over 37.6 million conversations mentioning the presidential candidates and election (mostly via Twitter) from over four million authors in the US. The company focused on swing states, which are influential in the US’s electoral college system. Authors were sorted into states by a geolocation algorithm. These conversations were then sent to BrandsEye’s crowdsourced micro-analysts to understand general sentiment when it came to the election candidates.
Kloppers says there are a number of reasons why most polls got the election results wrong – mainly that people lied because they felt embarrassed about their views, didn’t trust the person conducting the poll, or didn’t want to share their true feelings with anyone.
“But that kind of question bias, we just don’t have it here on social media because we are not asking anyone any questions. We are using unsolicited opinions which are shared. There is no incentive for somebody to lie. They might, but generally if they are going onto social media to share an opinion, it’s because they have an opinion and they want to share it,” he notes.
“The second challenge with the polls is getting a representative sample. Because they typically deal with landline phone numbers, and they call a whole lot of people and only get through to some of them, they end up with a sample of data which they are not quite sure whether it is representative of everyone, or if it has been somehow skewed. And because that sample is small, typically around 1,000 people, any error that is introduced into the methodology can have a fairly significant impact on your end result.
“With social media the sample is so much bigger, so you don’t necessarily have that same challenge.”
BrandsEye’s clients include governments and advisory firms, as well as brands. Among them are Deloitte, Uber, McDonalds, Burger King and Adidas. The company is also providing insight into consumers for several businesses focused on expanding in Africa, and has already set up its crowdsourcing platform in Ghana, Kenya, Nigeria and South Africa.
“One of our clients is a fast-food company wanting to enter the Ghanaian market and they had a few stores [where they] weren’t gaining any market traction. They needed to understand why, so we tracked the conversation about eating, which is such a broad topic – but [specifically] in Ghana and [looking at] how they feel about fast food, and what they value. By casting that net broadly and then layering on top of that, we could figure out the things people valued… and make recommendations to the [client] on how to tweak their menus and to introduce products in collaboration with sporting events to appeal to local people. And it worked.”