Like any historic big system implementation — think ERP — predictive analytics implementations offer lots of exposure for error. John Elder, CEO of data mining firm Elder Research, told ComputerWorld’s Robert L. Mitchell that while 90 per cent of the company’s projects are technical success, only 65 per cent of those ever get customer rollout.
While Mitchell’s story in ComputerWorld lists 12 predictive analytics screw-ups — based on interviews with expertts from Elder, Abbott Analytics and Prediction Impact — most can be categories in a few broad categories.
* Data issues: Trying to create the perfect data set — on, conversely, not scrubbing enough garbage data — can both slow down or derail a predictive analytics project.
* Scoping issues: A firm grasp of the end result and sound project management principles, particularly with regard to timelines, are critical to a predictive analytics success. Starting with a massive, high-profile project is not the best way to go.
* Staff issues: There is a potential minefield here of competing agenda among the owners of the data, the IT department and subject-matter experts. Everyone has to be on the same page, so executive support vital.
The bot threat
Some of the most serious threats networks face today are "bots," remotely controlled robotic programs that strike in many different ways and deliver destructive payloads, self propagating to infect more and more systems and eventually forming a "botnet."