Data mining saved HP $20 million
Successful data mining strategy requires company-wide engagement
Computer giant HP added $20 million to its bottom line through data mining activities, demonstrating that businesses can see a huge return on their analytics investment.
According to Dr John Elder, chief scientist at data mining and predictive analytics consultancy Elder Research, data mining can help companies in three main ways - by eliminating the bad, discovering the good, and streamlining existing processes.
In HP's case, the company used data mining to 'eliminate the bad', that is, identify fraud activity.
"One of their problems was service fraud. People were tasked with fixing a machine, but they would fill in a form saying they fixed a machine but didn't. The only way HP knew was when a disgruntled former employee would call in with a tip," Elder told the Predictive Analytics World conference in London today.
HP therefore built a mining model based on the known cases, to find the unknown ones.
"[Initially] they had a few part-time people discovering the fraud, but then [with the predictive analytics] they recovered $20 million in nine months, and more people were hired to do it, and it became a profit centre for the company," said Elder.
The company also achieved hidden returns in the process of collecting the data and gathering metrics for the fraud detection model. For example, HP discovered that a large consultancy firm in the US was continually returning laptops before the warranty year-end, even though no problems were found with the devices, so that they would be replaced with a newer model.
"They [HP] had been consistently abused by this partner, and cutting them off saved them millions of dollars that was not looked at," Elder said.
Although using predictive analytics to streamline processes can mean that organisations can reduce staff or increase productivity of staff, Elder said that in every example he has seen in the real world, staff numbers are not cut. Furthermore, as with the case of HP, the profitability of the analytics division means that staffing levels can actually grow.
"Every single time, staff increases because the group becomes successful, they become a profit centre, and they get other work to do," he said.
Meanwhile for pharmaceutical company Pharmacia and Upjohn, which merged with Pfizer in 2003, data mining enabled it to continue developing a drug that became one of its most successful products.
The company simply needed to find out if a new drug was effective enough to warrant major investment for development.
"They were about to give up on the drug and not spend a billion dollars, so we came up with this new plot [mining model] that compared the placebo and the drug, and it showed that the large majority of people got better with the drug, and no one got worse.
"It [the model] was appropriate for their decision at that moment - 'should we move forward?' They went on to invest in this drug and it became one of their three blockbuster successes of the decade," Elder said.
But Elder said that one of the main lessons from real-world applications of analytic projects is the importance of company-wide engagement.
"It is good to have the technology, but you need to arm everybody up the chain to see the value and to see the connections," he said.
Although data analysts typically might not be strong at communication with other people, according to Elder, communicating with others in their organisation will allow analysts to learn more about their project, as well as enable their company to understand what they are doing and why.
"You need to reward that trust [from business allies] by giving them as much material and results as possible so that they can feed the beast - the client, the boss, and so on."
This is why Elder recommended that pilot projects should be done early, with short timescales.
"Try to get a result within three months that is useful and valuable, that has a small return that can give reward," he suggested.
Yesterday at Predictive Analytics World, business intelligence consultant Jos van Dongen advised that businesses should carefully evaluate open source software for data mining as they would any other solution.