Computer picks Boks to beat Scottish by 11
Moreover, five days before the Springboks played Japan using recent form in the absence of previous results between the sides, it predicted an upset.
Analysts found this unfathomable and crunched the numbers again a day before the game and the software took a softer, more palatable stance on the Boks.
“We all make subjective calls based on the informal processing of data at the back of the head. What we do is formalise that process,” explained Tom Maydon, the comp head of credit solutions.
“We look at previous results and recent form. We also look at what the bookies are saying as well as qualitative data. That includes the form of players and to get that is quite tricky in rugby. For that we tap into a Fantasy League and get a sense of what the public is doing in terms of selecting or dropping players. We see that as a proxy for determining the form of players.
“Once we’ve got all that information the machine will crunch the numbers and output the result of the game with a predicted margin. Everyday we grab more data and feed it in. We turn the wheel and out pops a result.”
The bedrock of the system is the data which dates back 20 years, incorporates 6 000 matches involving 99 teams.
In explaining how they arrive at their conclusions, the one thing Maydon neglected to mention was the influence of the most powerful man on the field.
“We can look at the referee and how he officiates matches involving a particular country as well as weather conditions.
“There are a few more things like players’ metres gained, tackles made etc which we don’t use yet. When we do we can make a call on a player’s form as opposed to his pedigree.
“If we were looking to do this exercise seriously we would look into using a lot more data. We are not doing this for any party but just as a navel-gazing PR exercise.”
Enough of the company speak. How will the numbers stack up when the Springboks play Scotland this weekend?
“The bookies are saying South Africa by 14. We’re saying South Africa by 11.”