In the world of digital advertising, it often seems that algorithms are king. But to analyze social media mentions about a brand and its service requires something cleverer. BrandsEye cracked it by using crowdfunded analytics. Russell Southwood spoke to its Chief Marketing Officer Nic Ray about why its clients find the insights gained so useful.
The latest video clip interviews from Smart Monkey TV can be found at the bottom of this e-letter
Brandseye was started 10 years ago by one of South Africa's first digital agencies, Quirk (now called Mirum, a part of the J. Walter Thompson Company and WPP Network.):"When we were doing online campaigns in the new world of social media, clients wanted to know the efficacy of what they were doing and what people were saying,"
The easy part of the analysis for clients was to simply add up the number of mentions for the campaigns being run by the brands but it was hard to analyse topics and to give a more nuanced picture of what was happening. To overcome this challenge, Brandseye built its own software with algorithms to make sense of what the campaign data showed.
"This is what became known as online reputation management or social listening. Our techie is fairly obsessive about things but it was hard to make sense of sentiment. Algorithms don't understand sarcasm and irony (in comments)." They found they could get 50-60% accuracy on sentiment but that the customers (organisations like banks, insurers and telcos) were looking for a more finely tuned approach.
So Brandseye decided to solve this problem by using crowdsourcing. They would pay people per mention to categorise them:"We take a sample and put it through the crowd. The bigger the sample, the more granularity you get,"
Those analyzing the social mentions do three things: 1. Is the comment relevant? For example, all mentions of apples are not all about Apple Computers; 2. How does the comment score on a sentiment scale from -5 to +5; 3) Identify the topics the sentiment is aimed at. Using this approach, it was able to lift the confidence levels on samples from 95-99%. The analysis is also able to do gender and location (and race in South Africa).
The analysis process is undertaken using a gamified app with lots of checks and balances:"Every mention goes to more than one person until a consensus is reached and we reward accuracy. Those doing it make as much as a person would do for a retail job in their country but doing a lot fewer hours. It will also do as an additional income and people can do it in the gaps in their work."
So how do clients use this kind of social mentions analysis?:"We have a South African bank as a client. They look at the levels of customer service at the other five banks they compete with. They can then make resource decisions (about customer service) to get things right and then check the efficacy of what they've done." Telcos also find this kind of tracking of customer experience works really well.
Another example is work they've done for a European Pay TV provider who looked at their churn rate and knew that if they could do something about it, they could earn a lot more money. Those leaving often spoke badly about them on social media and by understanding their reasons more clearly, they could respond to them.
Another client is a Government that uses it as a tool to monitor sentiment towards their foreign policy, something which their embassies find very useful to have.
"Until they get this analysis, clients have no idea which topics are most important for their customers. They find it useful to understand what information will be most effective first."
The service is provided on a global basis from an office in Cape Town. The challenge in Africa is being able to source sufficient mentions across key languages. However, it has seen a fair amount of growth in Swahili and in Arabic across the Middle East.
The company has had a round of investment from both the leadership of the company and angel investors including its Chair Etienne de Villiers.