Predicting a No 1 song used to be a mug's game. But today a team of data experts and computer geeks will generate a scientific model which promises to help EMI's star acts craft the perfect hit.
The mission for the data scientists, gathered by the record company behind Coldplay and Kylie Minogue is simple: devise an algorithm from a wealth of consumer research which can finally predict if a listener will love a new song.
Pizza and beer will be served at today's 24-hour "hackathon", staged in central London, until the data experts come up with a solution. The results will be used to help EMI executives decide which new acts the label should sign, or what single stars like Katy Perry and Tinie Tempah should release to reach new fans.
The project is being overseen by David Boyle, EMI's head of consumer insight, who helped build the data analysis infrastructure for President Obama's 2008 presidential campaign. In an era of declining recorded music sales, EMI needs to know more about how the potential audience will react to an artist or song before investing in an expensive release campaign.
The scientists will be given the results of EMI's One Million Interview Dataset, the largest body of detailed music research ever compiled, which asks consumers across 25 countries their opinion of EMI artists and rate songs they have just heard. A song from one new EMI pop artist was described as "catchy" by 36 per cent of 13 -15 year-olds and "depressing" by just 1 per cent, suggesting the act will hit its target young audience. A rock act, however, polled strongly as "aggressive" and "distinctive" by an older audience whilst still rating "cool" for 16-34 year-olds.
Artists need strong egos to learn from the research. "It's hard to accept negative feedback, but the insight can help shape the artist's thinking into where consumers are," said Boyle. "One household name went through the research in absolute detail for three hours. We've had major artists decide to re-sign with EMI because of the insight into the kind of projects they should pursue."
The 125 modellers, recruited from Data Science London, a non-profit collective, will be given research about the consumers' geographical location, whether they buy music or illegally download it as well as their passion for particular genres. A cash prize will be awarded to the algorithm which best helps EMI predict a song's success. But Boyle insists that the statistical modelling will not replace the old-fashioned "hunch" of a talent scout.
"Selling music is about persuasion and marketing," he said. "We want to find out what it is about a listener that makes them like a song."