Report on E-business
New tool answers e-mail more intelligently
Better auto-response software integrates
Friday, May 25, 2001
Special to The Globe and Mail
Sending a question via e-mail to a company's Web site can be like casting a fishing line into Lake Ontario -- you never know what you'll get back.
Many Web sites use automated-system responses to reply to customers' e-mail inquiries. Such systems scan e-mails for keywords that trigger standard replies, explains David Daniels, a senior analyst at Jupiter Media Metrix.
The problem is that many keywords are ambiguous and may have double meanings. And most auto-response software can't decipher such language nuances.
That can lead to responses that have nothing really to do with the query. One frequent culprit in the United States is the word "check," the American spelling of the word cheque. An e-mail query such as "Can you check on my order?" can trigger an absurd reply like "Your last check was found: status, insufficient funds."
That, clearly, is aggravating to the customer making the query. But a new kind of auto-response software -- one that some industry observers say is poised to become the next hot customer relationship management (CRM) application -- is expected to solve many of the current frustrations.
The more intelligent software integrates natural-language-processing (NLP) technologies.
This more advanced auto-response software doesn't read e-mail queries for keywords, as do earlier systems, but rather understands the context and intent of customer e-mails to gauge the nuances of human language.
What's more, programs that use NLP offer another bonus: they can cut staffing needs by 40 per cent at organizations with large volumes of electronic mail, Jupiter says.
"Within 12 months, I believe NLP-based software will have a deep penetration, especially with financial services companies and on-line retail," Mr. Daniels predicts.
Organizations that continue to rely on earlier software run the risk of alienating and annoying customers -- many of whom contact organizations through e-mail, he says.
The research company surveyed 225 Web sites and found that 26 per cent use e-mail auto-response systems. But fewer than 2 per cent of them successfully answered an inquiry with an automated response, according to a report that Mr. Daniels wrote last month.
"The faults of auto-response systems are spurring customer cynicism regarding e-mail as a service channel -- that it is one more insignificant interaction in an already frustrating waiting game," the report said.
In this era of real-time, diehard customer service, where on-line organizations are desperate to learn how to attract and retain customers, quick and correct e-mail replies are becoming a major expectation.
Most older automated e-mail response systems aren't up to the task. And relying strictly on real people is not a wise strategy either.
Customer-service representatives are more likely to write logical and correct responses, but they take a long time to reply to e-mail. In fact, the same Jupiter report found that, based on an average telephone call time of four minutes, and an after-call processing time of one minute, a customer service representative can deal with 12 customers by phone in one hour.
But because typing and reading take more time than speaking, the same CSR could respond to only 10 customers an hour by e-mail, based on a six-minute processing time.
The solution for companies is either to type faster or find more intelligent software that answers e-mails quickly and correctly. More companies are opting for the latter solution.
Royal Bank Financial Group is one firm in the process of implementing NLP-based auto-response software developed by Banter Inc., a San Francisco-based provider of intelligent-communication technologies, which outfitted Wells Fargo, the largest on-line bank, with the same solution.
"With e-mail volume growing, we really needed to look at improving our turnaround time to meet customers' needs," says Bruce Green, senior manager of enterprises Web services at Royal Bank.
He says the bank looked at about a dozen companies with NLP solutions but chose Banter's software because the program offers several applications in addition to replying to e-mail, such as training people to answer e-mail queries.
Banter's Relationship Modeling Engine reads an incoming e-mail and, using NLP, the engine attempts to decipher what exactly the customer wants to know.
The system learns what to say by watching real customer-service representatives answer queries. It learns how certain questions are answered and stores the information in its memory bank, constantly updating answers as new questions come in.
"The software keeps learning. It knows better than a new person on the job. The software can get as smart as the best customer-service representative, but not better. It needs someone to learn from," says Tom Aden, Banter's chief executive officer.
He says the software can reply to queries that contain misspellings, incomplete sentences and grammatical errors. And the software understands obscenities, too. For instance, those who want to "unsubscribe" to a newsletter rarely use that word, Mr. Aden says. "We notice when people use swear words, it often means they want to be off the list. The software can learn that and respond to language uses in the real world."
Mr. Green says that Royal Bank receives about 1,000 customer e-mails each day that will soon be answered with the help of Banter's software and 45 customer-service reps located at two call centres in Moncton and Mississauga.
Although he declined to reveal the cost, he says Royal Bank expects to recoup the investment in the NLP-based system within two years. He says the bank will not have to hire any more people to answer e-mail in the future, which is a major cost saving.
"The software definitely improves productivity. It prepares answers and does the analysis. Obviously we will need fewer agents in the future," he says.
However, Mr. Green says he does not foresee a time when software alone will be used to answer queries.
"When you use artificial intelligence, the goal is to work toward having no people doing the work. But our plan is to use both for the best quality control."
Other companies go one step further than Banter in anthropomorphizing software.
San Francisco-based NativeMinds, a provider of natural-language customer-service software for the Internet, has developed automated virtual representatives called vReps for e-business customer service.
Billed as a way to cut down on labour costs, vReps are on-line personalities that emulate real human customer service on the Internet. Oracle Corp., for instance, uses an internal help-desk vRep named Allen to provide technical support and training for internal staff. The vReps are designed to answer customer questions by understanding conversational context.
With intelligent NLP-based software capable of understanding human language, and responding to questions in real time, are humans becoming unnecessary?
Not quite. Intelligent software won't reduce the need for customer-service representatives, but it will change their roles, says Sean Seaton, director of consulting for CRM at Cap Gemini Ernst & Young in Toronto.
"The auto-response technology will let contact-centre agents behave more like consultants with customers. The technology will answer the mundane questions, while agents will do a lot more problem solving," he says.
Although NLP-based systems do not require as many people, the smartest auto-response systems still need a few people who can review complicated e-mail questions and route messages to the correct people within an organization, Mr. Daniels says.
The problem, he says, is persuading on-line organizations to trust software. "Many businesses have experienced problems with previous e-mail automation solutions. There is a real problem of trust. They wonder, 'Can the technology really do this?' "