Hilton Inc. has always relied on the information it received from guests to steer its services and improve the customer experience. So when it was time to invest in text mining and content analysis technology to boost its ability to find valuable information, the company chose a structured approach to shopping for the right product.
Christine Hight, Hilton's vice president of market research and customer insight, assembled a steering committee to evaluate and choose text analytics software for the enterprise.
“We knew we were sitting on a gold mine of information,” Hight said.
The committee consisted of two people from her department, one IT person with proposal-writing skills and another information technologist with application development expertise to ensure Hilton’s systems would integrate with the final choice. To get there, the core committee put together a cross-functional team with representation from across the enterprise.
“We knew that this was a need for the entire company, potentially,” Hight said.
Start with a steering committee
According to analysts and consultants, that approach was the right one to take.
“A steering committee encompassing both business and technical concerns is essential for any adoption beyond individual use,” said Seth Grimes, founder of consulting firm Alta Plana Corp., in Takoma Park, Md. When it comes to evaluating a text analytics product, he added it was important to involve end users as well.
The general rule of thumb is, if somebody can say no [to the purchase], put them on the committee. And then anyone you want to say yes, you want them on the committee as well.
Gareth Herschel, research director at Gartner Inc.
“It’s important to make sure you pick people from all levels of the company … and make sure you include people who are getting their hands dirty using the software,” confirmed Theresa Regli, principal analyst at Real Story Group, of Olney, Md. But, she pointed out, executive sponsorship is likewise important -- for two reasons: Someone has to sign the purchase contract, and it’s useful to have someone on the team to remind others of the ultimate business goals of deploying text analysis software.
“The general rule of thumb is, if somebody can say no [to the purchase], put them on the committee,” said Gareth Herschel, research director at Gartner Inc., of Stamford, Conn. “And then anyone you want to say yes, you want them on the committee as well.”
Pick the right people to evaluate
With an eye toward scalability, the committee identified all Hilton departments that collected or used unstructured data to do things like analyze guest comments and make changes to improve service. That’s no small feat for a company with 3,600 hotels operating under an umbrella of nine hotel brands. Adding to the complexity, Hilton has properties in 82 countries with employees and guests who speak a multitude of languages.
Typically, each hotel surveys guests independently, and clerks collect comments over the phone and by e-mail. Others review comments on social media and collect others directly from Hilton websites. The company wanted employees who collect that information on its evaluation team.
“The makeup of the group depends on what you’re buying the text analytics package for,” Regli said, noting that customer service personnel made sense in a service industry like Hilton’s. “In e-discovery you want the people who actually run the documentation group or mine the information. The people who need to be in the selection process need to be the people who are doing the manual work now.”
Herschel pointed out that in certain cases it made sense to include both end users and managers. “You want [managers] involved in the evaluation process, because they are responsible for the process changes” that will come about once text analytics is deployed.
Make every score count
After a seminar on text analytics, each person on the evaluation team took two weeks going through 17 returned vendor proposals, rating each on 10 criteria. After counting all scores returned by the evaluation team, “it was pretty clear who the top four contenders were,” Hight said. An averaging schema established the choice for a text analytics pilot project.
While Grimes said an initial evaluation and implementation shouldn't be too broad, “attempting to cover too many goals, business functions, information sources or users,” he said that would vary by organization. In Hilton’s case, a certain expansiveness was built into its project based on the various channels it wanted to dig into with text analytics and natural language processing.
While Hight said taking a cross-disciplinary approach to evaluating software was difficult, it served to immerse a group from many different departments in the language of text analytics and planted the “seeds for people to think about how their department can use it.” On top of that, giving each task force member’s vote equal weight gave executives confidence in the team’s recommendation. “Even though it’s hard to do,” Hight said, “it was the right thing to do.”