Voice recognition of keywords and the use of voice as a biometric for identification are becoming core features in interactive voice response (IVR) systems, delegates at the Voice Leadership Forum in Wellington this month were told.
First-hand reports of user experience of service through the automated voice channel came from New Zealand’s Inland Revenue (IRD) and Ministry of Social Development (MSD) as well as the BNZ and Westpac banks, were shared at the event.
IRD began to modernise its phone interface in 2008. The first offering was “virtual hold” – the customer faced with a long wait can hang up and be called back when a staffer is available to handle their call.
Now, with innovations introduced last year, the department’s system recognises key words in the caller’s speech and tries to route them to the person best able to handle the inquiry, customer-service group manager Heather Daly told the conference. If possible the customer is routed to a self-service option through voice-response or online.
“For example, if a customer calls about Working for Families close to a public holiday, we’ll play them a message telling them when the next payment will arrive in their account,” she said. At those periods, IRD can deal with 10 percent of its incoming traffic using such “banner messages”. A live person need not become involved at all, but the customer still gets served.
Keyword recognition has also been used to pick up customers inquiries about suspicious-looking emails and alert them to a phishing scam. They were directed to a message telling them what to do; most importantly not to click on the link in the message.
Customers who call the department only once a year to get forms to file a return are anticipated and sent an SMS message telling them when and where this year’s forms can be obtained.
The end purpose is to “make it easier for customers to comply” so IRD does not have to chase up late payments as often, Daly says.
A survey on the self-service system found that customers either regarded the interface as “very easy” or “very hard” to operate; the latter “shows we need to do more work on it,” she says.
Business intelligence tools are crucial to the analysis both of such “after-call surveys” and of actual call behaviour, Daly says. Two people are employed full time at IRD analysing customer behaviour on calls in real time, so the system can be tweaked for better performance.
Internal transfers when the customer is directed initially to the wrong person, have been reduced from 30 percent to seven percent of calls.
One failing in developing IRD’s voice systems was in not involving operational staff early in the developments. It is important to meet their needs and take their feedback as well as the customers’, Daly says.
IRD is piloting voice-print identification to answer completely automatically straightforward queries like “what is my IRD number” or “I’ve forgotten my password.” It has set a target of 800,000 customers “enrolled” into voice ID within the first year, concentrating on the customers that call most often.
MSD launched its voice biometrics service in June this year, with the aim of reducing to a minimum the time taken to become sure of a customer’s identity. With more than six million calls coming into the ministry each year, a second saved on each call is worth 1.5 full time staff members, says operations manager Tony Stenhouse.
Enrolment is not forced and, unlike IRD, there are no enrolment targets. A customer serviceperson will ask the customer, particularly one who is having trouble with the ordinary identification process of reading a client number “would you like to have an easier way of identifying yourself?”
Only three percent of customers decline the offer. Identification by voice is successful 80 percent of the time. Because mobile phones compress speech, it is not possible to do initial registration of a customer’s voice-print on mobile, Stenhouse says, but once the print is registered, with adjustment of tolerance it is possible for the customer to use a mobile for subsequent calls and be correctly identified.