A Penny Learned: How a Virtual Agent Gets Job Done
February 14, 2011
Penny only went to work in an investor center in October.
Back then, she could only answer one in four or one in five questions asked of her, correctly.
By the end of December, she was answering more than two out of every five questions, correctly.
"Which is pretty good," considering, said Brian Heffernan, product manager at Computershare, a transfer agent that operates customer service centers for shareholder servicing for public companies and closed-end mutual fund companies.
Considering that Penny is not human. She is a "virtual" agent given the graphic trappings of a human customer service representative.
She is a replacement for the scores of "frequently asked questions" and corresponding answers that Computershare put it in its online Investor Centre service.
The introduction of Penny was a means of increasing the automation of interactions with customers, but doing it in a more personal-and more precise-manner.
To get Penny to work, Computershare fed its questions and answers into a single database that now gets combed any time an inquiry comes in on screen, using a tool from noHold, a Milpitas, Calif., supplier of virtual agent technology that relies on natural language processing and inferences.
Users aren't forced to leaf through page after page of listings or hunt down answers. Instead, they simply type in their question, the tool then analyzes its contents, tries to deduce the information that is being sought and then responds.
When she started, Penny was not so effective. But the system captures each user query and system interaction. And members of Computershare's product team, contact center, operations and marketing department meet at the beginning of each month to evaluate the performance of the company's virtual agent.
If a question came in about "escheatment" or "abandoned property" and no satisfactory responses were logged, the team pretty much knew a new set of questions and answers-and definitions-would have to be fed into the database.
"There were a lot of questions that were asked that we had no content out there for," Heffernan said. But the tool constantly asks customers if they got the answers they wanted and, in so doing, can help guide the team into creating new material that fills holes.
"Those first two months of evaluation were pretty dramatic. We were able to put some content out there pretty quickly," Heffernan said.
This can get tricky. A person whose company offers a Dividend Reinvestment Plan might simply refer to it as a Drip, when typing out one of its widely used acronyms. The system has to know that the person is referring to a DRiP.
And the system has to know that companies use different acronyms for the same thing. Like DRP. Or even DSPP (Direct Stock Purchase Plan).
The ins-and-outs of complex subjects, like cost-basis accounting, have to be grouped and weighted, to help create more precise responses, automatically.
Over time, the tool gains experience and gets better.
And Penny will probably always be learning. Because funds change, services change, laws change, rules change. Which means Penny, who was named either after the coin or the wife of the company's founder, depending on whom you ask, will never completely replace human interaction in serving customers.
"We acknowledge it is not yet perfect," Heffernan said of the firm's virtual agent. "It's just one tool we're using."
Next up is online chat, which has been a low-cost answer to providing the human touch for many years. TradeKing, a discount brokerage with 250,000 customers and $1.4 billion in their accounts, has relied on live chat since its start five years ago. This has been a primary means of quickly answering questions, such as the meaning of difference classes of shares or the difference between closed and open-ended funds.
The company provides no investment advice, but answering such questions helps the investors decide how long, for instance, they might have to keep their money in a given fund. Which does affect the investment decision, directly.
In its investor center, TradeKing uses some social media techniques, to help customers help themselves. With forums, one investor can, in effect, advise another. Blog postings, from experts, can help, as can notes from traders in its Trader Network on the decisions they made and why. Users can advise each other how to best use the Mutual Fund Screener, as well.
Computershare is in the process of bringing online chat into its investor center. But it had to get over the "hesitancy hump," Heffernan said, caused by the fear that it would raise the company's costs in interacting with customers.
That hump was overcome, when it determined that customer service reps answering questions could and did interact with more than one customer at a time, in live chats.
Whether it's through Penny, the virtual agent, or her living kin conducting live chats, the overall effort, Heffernan said, is to eliminate or reduce electronic mail.
"We don't want e-mails because, as any transfer agent will tell you, responding to an e-mail is the most costly method with which you could interface with a customer," he said.
That's because email does not generate its own responses. That requires not just a human, but a human skilled enough at responding properly-in a manner that not only fulfills the customer's need, but also will stand up under scrutiny, in a compliance review.
Beyond that, responding by e-mail tends to invite ... yet another e-mail from the customer. The multiple contacts ratchet up costs, dramatically.
"Once it gets to the e-mail stage, it gets very expensive," Heffernan said.