Oct 28, 2011 (05:10 AM EDT)
HP Mines Data To Predict Consumer Behavior

Read the Original Article at InformationWeek

10 Cool Social Media Monitoring Tools
Slideshow: 10 Cool Social Media Monitoring Tools
(click image for larger view and for slideshow)
The combination of social media analytics with company data predicted customer behavior with up to 90% accuracy in a pilot project completed by HP Labs and HP Global Customer Intelligence, the company said.

Though HP has not announced any specific product plans based on these findings, it is expanding the use of the technology within its own operations and exploring similar pilot projects with some of its large customers. HP called its pilot Project Fusion because it combines two different kinds of data. One is unstructured data, including Amazon.com reviews, customer surveys, and other textual data. The other is structured data, such as customer support tickets, sales transactions, and customer demographics.

Using HP Labs' new text analytics technology, the team first converted the unstructured data into a structured format. Analysts then used standard data-mining and statistical tools to analyze the combined data sets. In one case, social signals predicted support tickets with 90% accuracy. HP also uncovered high correlations between social signals and product sales.

[Even Twitter can reveal consumer intentions. Read more at Do Tweets Predict the Future?]

Prasanna Dhore, vice president of customer intelligence at HP, said the kind of analysis his group is doing goes beyond the general measurement of "buzz," which has been the primary market research application of social media monitoring to date. "The results are significant enough to act on," he said. For example, by correlating social media complaints about a product feature that is not working correctly with customer support trouble tickets, an organization could intelligently modify its call center staffing plans to ensure enough of the right people to handle those complaints are on duty, he said.

One way the HP team made social data analysis more meaningful was to drive the analysis to a deeper level of granularity, Dhore said in an interview. As an example, he points to this comment culled from customer reviews: "I don't know what people are complaining about regarding the software but it installed seamlessly and is intuitive in its operations. Even though I am dissatisfied with the paper tray, all together I am happy with this printer." There are several sentiments expressed in that one comment--about the software, the paper tray, and the customer's overall impression.

Rather than doing sentiment analysis on the statement as a whole, HP designed its analysis to score statements about each aspect of the product separately. To accomplish this, HP went through the process of identifying key aspects of the printer analyzed during the pilot project so it could see what people were saying about the ink, the print head, the setup, the paper tray, the print quality, and so on. Another way of applying this data is as feedback to product development, Dhore said. For example, if people are generally happy with the print quality, it might be time to divert more resources to something they are less satisfied with, such as the setup process.

Similarly, HP can apply the same type of analysis to competitive products, so if one that consumers are comparing HP's offering against has dramatically better print quality, the company can investigate to see what that competitor is doing right, he said.

Zach Hofer-Shall, an analyst at Forrester Research focused on "social intelligence" and practical applications of big data analytics, said HP's efforts dovetail with his predictions in a report on integrating social and customer data.

"Today, there are the listening parts of the organization and the data parts of the organization--that won't be the case in the future," Hofer-Shall said in an interview. Social media monitoring is typically owned by the marketing organization, while company data is controlled by IT, but the two camps need to bring the data together to use it to maximum effect, he said.

Though matching social media profiles to specific customer records can be difficult, HP's examples show how social data about general sentiments and trends can be useful even in the absence of that matching, he said.

Hofer-Shall said he will be interested to see how HP's recent acquisition of Autonomy plays into its social media analytics plans, because Autonomy brings many relevant search and text analytics technologies to the table. Dhore's pilot project predated that acquisition and was based on HP's own software development.