Not that it was an overnight success. "Initially, we got a lot more tickets," she said, because the knowledge base was offered with the ability to escalate issues to be handled by a representative "and we were making it easier for them to find us. We were quite overwhelmed at first."
Part of the problem was the same Yola had encountered with other knowledge base searches, where the terms users entered when asking a question didn't match the keywords in the support articles. That changed as the system learned the words and phrases users entered and could match them with the responses from the agents. By the metric of an automatic answer rate--where users are sufficiently satisfied with the automated response that they don't feel the need to open a support ticket--performance steadily improved.
"We started off at about 67%, but now we're consistently seeing it's between 90 and 95%, which is unusually high for self-service," Viljoen-Platts said. The results have been notable enough that one review of Yola's service on Consumer-Rankings.com called out the quality of customer support for praise, even while noting the absence of live support options such as chat.
Although nanoRep can automatically add answers to the knowledge base as they are generated, most customers chose to have them go into a queue for review first, which is the way it works at Yola. Viljoen-Platts said one of the things her team has done to improve performance is fine-tune the responses before they are added to the knowledge base. In the beginning, they made the mistake of putting too much information into each support messages. That became a problem because there were so many places information needed to be updated if instructions changed. Instead, what worked better was to send back a few sentences, with links to tutorials and articles.
"We settled on short, pithy answers with links to information, and then made sure whatever we linked to was up to date," Viljoen-Platts said. "That way, we're updating the information in one place, and customers tend to tell us pretty quickly if ever the information is not correct."