Software firm Zone Labs recently dug itself out from under a mountain of unanswered e-mail by using an artificial intelligence system that is able to “read” text queries and suggest responses to customers' questions.
Now, Zone Labs, San Francisco, which markets a computer security solution available for download directly through the Internet, is turning e-mail service queries around in two days and automatically rerouting queries that should go to the sales department. The company also is preparing to use that same AI solution to migrate its customer service function from an e-mail-only environment to a self-service, Web-based format.
The situation was much different about eight weeks ago. With 10,000 unanswered e-mail messages in its service queue, the company was the poster child for poor customer service on the Internet. Just a few months after it began selling ZoneAlarm, a software product designed to protect the information stored on computers used within a network, customers had downloaded 3 million copies and had begun firing back service questions via e-mail.
After considering several options, the company decided to install e-mail management technology from Banter Inc., San Francisco. Rather than searching for keywords, as some e-mail management solutions do, Banter detects the intent of the e-mail message by looking at the patterns of words in the text and comparing them with previous messages that have been answered correctly. It then suggests responses that the agents can select from, and it displays those suggestions on the agents' screens.
Using that system, Zone Labs' two-person staff now is handling 200 to 500 incoming e-mail messages per day with no phone support.
“My estimation is that without an e-mail management system, an agent can handle 70 to 100 e-mails a day, some days more, some days less, depending on the complexity of the questions,” said Claus Tarstrup, director of product marketing at Zone Labs. “With Banter, on good days, agents are answering 250 to 300 e-mails [each], on bad days, 150.”
Zone Labs has the system set up to provide five suggested, canned responses for each e-mail inquiry. An agent then reads the question, selects the correct response from the five choices the system offers and responds. Agents can then augment those responses with additional text comments that they can type in themselves.
In addition to helping the agents stay on top of the e-mail traffic flow, the e-mail management system also is helping Zone Labs to distribute some queries directly to appropriate departments. For example, the system routes e-mail messages that it determines are from potential buyers directly into the sales queue.
“If someone says, 'I'm really interested in ZoneAlarm, and I want to buy 1,000 copies. Can you provide me with a quote?' it will actually take that e-mail, even though it was sent to support, and put it into the sales queue because it knows that the intent of the e-mail is that somebody wants to buy the product,” Tarstrup said. He receives about 15 e-mails per day that are routed to the sales queue from the customer service department.
He said it is fairly common for prospective customers to send a sales inquiry to the customer service department deliberately as a test of the vendor's reliability.
In addition, the system can detect if there is an unusually large number of questions about a particular technical issue and send up a red flag to the quality assurance department.
Like other AI solutions, the Banter system is designed to learn continually from the responses that are given to certain questions. When agents select a correct response from the field of chosen answers, the system remembers the pattern of words that appeared in the question and offers the agents the same answer again when it sees that word pattern appear in another question.
Tarstrup said the Banter system has even been able to distinguish what operating systems customers use based on their questions, even when customers don't specifically mention that they are trying to install ZoneAlarm on a Windows system, for example. Thus, Banter is able to suggest the correct response automatically without having an agent try to take the time to sort through the whole message.
Russ Rosen, director of marketing strategies at Kana Communications Inc., which markets a rival AI e-mail management product and is based in Redwood City, CA, said such systems are helpful in handling large volumes of e-mail inquiries in which many of the questions fall into the same category and can get the same prepared response. Less complicated forms of e-mail management, such as pattern-matching systems and Web forms in which users simply fill in the blanks, can also be highly effective customer service tools in many cases, he said.
Tarstrup concurred with Rosen's assessment of AI e-mail management, saying that the system has been particularly effective in sorting all of the queries his company gets about the availability of a ZoneAlarm system for the Macintosh.
“There's no value to us in having a pair of eyes evaluate 1,000 e-mails asking us if we have a Mac version,” he said. “We don't, but we know we will have [one] soon, so why waste time answering that?”
Tarstrup said Banter is able to identify correctly 90 percent of the queries about the availability of a Mac version, allowing agents to spend more time on the more complicated or multiple question e-mail queries. The system also has been able to identify 80 percent of the questions about how the system works with America Online, which is another common query.
In addition to helping the company sort through its pile of incoming e-mail, the Banter system also is helping Zone Labs build a self-service component for its Web site. The knowledge the system has gained by using e-mail responses will soon be placed on the Web so that customers will be able to type free-form text questions at the Web site and get an answer without the intervention of an agent. That system also will have the capacity to improve from previous correct answers.
Meanwhile, Zone Labs is preparing to introduce additional products that will increase the complexity of the system even further. This time, Tarstrup said, his company is ready.
