April 29, 2014 - Comments Off on Should crowd sourcing be the future of banking bonus decisions?

Should crowd sourcing be the future of banking bonus decisions?

On the day that the RBS group announced losses of £8bn, the news story was always going to make headlines. The announcement was the subject of nearly 20,000 tweets – the majority of which highlighted the bad deal that UK taxpayers got from the bank’s losses.

When we have computers analysing data quicker than you can say “algorithm”, it’s important that the analysis is put into context so we can properly understand the patterns. As data sets get bigger, so the patterns become more obscure and potentially misleading.

Therefore it’s crucial to consider context in the commercial world, where insight can drive value.

On a day when the RBS press team would have been nervously checking data streams to understand the impact of their announcements, they’d have been forgiven for thinking that things weren’t too bad. The first place to check would have been the automated sentiment of individual tweets.

If the RBS team were looking at how the raw sentiment was changing over time, they’d have seen a graph similar to the one below:

Analysing this graph would have led the PR teams to breathe a sigh of relief. It’s choppy out there in the sea of sentiment, but there aren’t any major waves of negative sentiment causing any lasting damage. Putting this into context though would lead us to a contradiction.

Looking at the conversation, this analysis just doesn’t match what we saw in practice. The UK population reacted negatively to the news in a way which only got stronger as the day wore on.

How to measure true sentiment more precisely

Bloom’s Social Intelligence Engine, Whisper, adds some context to the analysis by considering the importance of what’s being said in the conversation.

It’s generally accepted that automated sentiment analysis is never going to be perfect and is probably only 70% accurate at best. Whisper weights the sentiment of an individual piece of content with the influence of the account sharing the content. Doing this allows us to “rescore” the sentiment using advanced techniques so we can improve the accuracy of the sentiment analysis we do.

Whisper’s initial view of the sentiment of the conversation is shown below and provides a much clearer picture of the nation’s views:

This view is still flawed though as we’d be mistaken for thinking that the sentiment started to increase after the initial burst of news. At 0810, the conversation moved away from the initial factual statements and started to focus on the bonuses being paid to bankers:

“Bonus” is seen by the sentiment filters as a positive word. Whisper was able to respond to this piece of context and rescore the sentiment accordingly. On doing so, we see a very different trend:

Far from breathing a sigh of relief, the RBS PR team can only conclude one thing from this analysis, and that matches what would be evident from reading just a few tweets in the conversation. In recent days we’ve seen how the RBS bonuses have been rejected – something which is no surprise when you consider the negative sentiment surrounding the issue.

Whisper: providing context and deeper insight

Peter Laflin, who leads the development of Whisper for Bloom, has been using this example to talk in detail about the importance of context in data analysis. It highlights the dangers of relying on automated processes without setting the insight in a wider context before making business critical decisions. The talk recently won an award at the Institute of Mathematics and its Applications conference held in Leeds.

Whisper was also able to provide the following insight on the conversation:

  • The BBC had the most impact on spreading the story and were the most influential. Their key twitter accounts were the most influential, as were the accounts of the Today programme and Paul Lewis
  • Of the newspapers, the Telegraph had both the most influence and the best response to tweets sent to them, mainly due to the higher levels of engagement offered when compared with other newspapers
  • Individuals directed their frustrations at David Cameron and George Osborne, and there was no direct response to the conversation seen
  • A parody account of David Cameron was highly influential – reinforcing the bad feeling felt by the Twitter population towards RBS, but also the Conservative government. Parody accounts are becoming increasingly influential in spreading political stories.

In May, Peter will be talking at an event in the City of London for Wealth Managers and Traders. The talk will explain this case study in more detail and help illustrate how social data is becoming more important for organisations in their decision making.

Whisper is a social media analysis tool that can measure in real time which conversations have the most potential to drive change in how people value a brand. Using unique algorithms developed by leading mathematicians, it acts as an early warning sign for analysts and traders so they can react sooner to stock market activity and plan more precisely.

For more information on Whisper, and to discover how it can help your business, please contact Stuart Clarke on 07789 845799 or email [email protected]