When it comes to high-level, potentially enterprise-changing strategic decisions, C-level executives and departmental managers usually rely on, or at least consult, mountains of structured data, often
However, they often miss out on equally important unstructured data, be it hidden away in emails, memos or even in people's heads.
Unlike tactical decisions, which are made on a daily basis by front-line workers, strategic decisions may occur only once or twice a year and involve high-level issues, such as whether to outsource production to India or launch a new product line, according to Gareth Herschel, research director at Stamford, Conn.-based Gartner Inc. Enterprises that don't take unstructured data into account when making strategic decisions, he said, "risk missing some key data points and key insights."
"The challenge that organizations have is they have lots of structured information – the data warehouse is full of structured data – and that's what tends to fill up PowerPoint presentations [that most strategic decision makers consult]," Herschel said. "What most organizations don't have is the anecdotal information or the data that isn't in the data warehouse."
That data, however, has traditionally been difficult to come by other than on an anecdotal basis.
Mining the depths of unstructured, text-based data
Text-mining tools offer a way for companies to tap into unstructured data, including emails and Word documents. The tools analyze patterns in text-based data and apply structure to it so it can be easily consumed and analyzed, much like structured data that sits in a data warehouse.
Such unstructured data, once unlocked by text mining, "can give you a much more visceral understanding of what's going on inside the organization," Herschel said. "It's capturing all the information that people have heard or said or read that is typically locked up in their brains and doesn't make it into the data warehouse."
Since January, Gaylord Hotels has been using text mining software from Reston, Va.-based Clarabridge to analyze guest satisfaction rates. At the end of their stay, guests at each of the Nashville, Tenn.-based chain's four hotels are asked to rate their satisfaction level on a scale of 1 through 5 and include comments on any specific concerns they had. Prior to implementing text-mining software, analyzing guest comments – the unstructured data – and tying them back to their corresponding ratings to gauge their importance was difficult and inexact.
"We previously had a manual process with another survey vendor where they were reading through these comments and manually calculating them," said Tony Bodoh, manager of operations analysis at the hotel. "It was a very laborious process, and last year we had about 80,000 surveys returned to us. So it was taking longer and longer to get those results back to us, and we frankly weren't able to get enough information beyond the number of comments."
With text-mining software, the hotel can now identify specific problems in guest comments faster and more efficiently and can connect them to the ratings scale for easier analysis. Those otherwise-unknown insights are then factored into strategic decisions around pricing structure and staff training.
Betting on the future with prediction markets
Another method to gain access to the ultimate unstructured data – employees' internally held and often-unexpressed opinions – is the use of prediction markets. Prediction markets typically take the form of an internal poll, posted on the company intranet, for example, asking employees for input on a soon-to-be-made strategic decision.
Employees might be given 1,000 virtual dollars, Gartner's Herschel explained, and asked to "bet" on the prospects of a potential new product enhancement. The prediction market technology tracks the bets like an internal stock market and forecasts the most likely outcome. Those employees who invest their virtual dollars for a product or service that is ultimately successful are then recognized for their foresight, while executives gain valuable new data upon which to base their decisions.
A prediction market might indicate that the marketing department overwhelmingly supports a particular new product, but the engineering team does not, Herschel hypothesized. "What that tells us is we've got a great idea for a product but we can't build it," he said. "With a prediction market, if 90% of engineering respondents say this is a dumb idea and it won't work, that's really a data point you're going to have to deal with when making the decision."
In mid-2003, Motorola Inc. developed an internal tool called "ThinkTank" -- open to all employees -- to gather new and innovative ideas. The Schaumburg, Ill.-based mobile phone maker eventually found itself overwhelmed with the number of ideas submitted – hundreds a month – and turned to Nashville, Tenn.-based Consensus Point to implement a prediction-market tool to harness them.
Now, when a new product idea or branding strategy is submitted to the prediction market, which has been operational since August, Motorola's employees are each allocated 100,000 virtual dollars to invest in, or against, it.
"The question that's posed to the market is: 'Here's an idea; do you think it's a valuable one for the company?' " said Rami Levy, a distinguished member of Motorola's technical staff. "And people basically buy more stock in that idea if they believe in it, or they can sell it short if they don't believe in it, or they can ignore it."
When executives and other strategic decision makers at Motorola mull their options, the sentiment revealed by its prediction market is another important data point they consider.
Text-mining, prediction-market technology still maturing
Gartner recently identified Clarabridge; San Francisco-based Autonomy Corp.; Palo Alto, Calif.-based Attensity Inc.; Jacksonville, Fla.-based IxReveal; and Chicago-based SPSS Inc. as text-mining vendors worth considering. There are fewer prediction-market vendors, according to the same report. These include Inkling Inc., with headquarters in Chicago; Xpree, based in Redwood City, Calif.; and Consensus Point.
Despite their promise of unfettered access to unstructured data, both prediction markets and, to a lesser degree, text-mining technologies are still fairly immature, Herschel said. Clarabridge, for example, has been around only since 2005, and Consensus Point since 1995.
Herschel warned that before making an investment, companies considering text mining and prediction markets should first identify their unstructured data needs and then make sure current technologies on the market meet those needs, rather than relying on future enhancements.
Have you invested in text mining or prediction market technologies to unlock your unstructured data? Email SearchBusinessAnalytics.com editors with your story!