Speak up, please!

Computers running natural language speech recognition programs are helping organisations like the Inland Revenue Department handle more than 40,000 calls a week.

The dreams of technologists for many years, or at least since the advent of sci-fi television programs, has been to engage in conversation with computers instead of typing at them. This has turned out to be harder to achieve than expected, and has been one of the things that prove just how smart we humans are, at least compared to our computers, which we often describe as high-speed morons. Software that can be trained to accurately recognise the speech of a single speaker is now well advanced but the business benefits of that application are limited. We need our computers to be less particular about their choice of conversation partner.

However, much like the fictional million monkeys typing at random to produce works of literature, the sheer force of processing power, along with some clever coding, has seen voice recognition systems evolve into useful business tools. Today it is no longer a source of wonder computers do a reasonable job of understanding what they’re told. Perhaps understanding is a little too generous. But these days, computers running natural language speech recognition (NLSR) programs can and do analyse the spoken word effectively and make rapid comparisons against their databases in order to make a decision.

Speak carefully

In Australia, Telstra’s White Pages phone number service has been using NLSR for some time. You either love or hate the service. Those whose speech happens to match the program’s parameters are able to get the information they want much more quickly than waiting for a human to punch the search button. The rest of us have learned using expletives in an attempt to avoid the computer interface leads to embarrassment when the call is rapidly switched to a live operator. Other NLSR systems are pushing past the boundaries of simple question and answer dialogue into the realm of genuine information collection.

A recent Frost & Sullivan report says speech technology has been gaining momentum and acceptance in markets where labour costs are higher, and these include New Zealand and Australia. In the Asia-Pacific region, the business consultancy firm says the speech market is still at its nascent stage, with a US$7.1 million worth in automatic speech recognition licence revenues in 2003, and compound annual growth rate of 34 per cent through to 2010.

For this special report, MIS focuses on the experiences of enterprises in government and financial sectors on both sides of the Tasman that have taken the voice recognition route.

IRD: Start simply

The Inland Revenue Department ranks number 13 in the top IT user organisations in New Zealand, according to the latest MIS100 report. The department collects approximately 80 per cent of the government’s annual $34.9 billion revenue. It has nearly 5000 staff in three processing centres, five call centres and 17 offices. The move to implement NLSR was prompted by the desire to continually improve customer service delivery and self-service channels.

Vaughan Crouch, national manager – social assistance, one of the business units of IRD, says the department handles around 3.6 million calls every year through its call centres. In addition, another 1.5 million calls go through a traditional IVR (interactive voice response) system called INFOexpress service. The department’s Child Support Agency deals with another 550,000 calls approximately.

When IRD rolled out an NLSR program this year, it opted to focus first on two services most often accessed through its automated telephone service. These are the applications servicing the requests for IRD stationery like forms and booklets, and personal tax summaries.

Crouch says both applications – rolled out in January and May respectively – have been “very successful”. The stationery line typically now handles up to 30,000 calls per month, which is great, says Crouch, because “that’s 30,000 calls that now don’t go through our call centres”.

The personal tax summary application is still relatively new, but Crouch says during the recent tax filing period, it handled 30,000 to 40,000 a week on top of the stationery calls. “We are obviously very happy with what we’ve got given that it has exceeded our business expectations. Our critical success factors have been achieved.”

Value added service

The new system tackles queries that can be met by the self-service system, and enables the customer service representatives to manage the more complex customer queries. “We take out the low value add calls where our customer service representatives don’t do much other than push a button,” says Crouch. “That means our customer service representatives are really doing ones where they really use their tax knowledge to help the customer.”

He says one of the key pointers for rolling out NLSR is to “start simply” as IRD did. “We have picked an application which was relatively a straightforward one for us.” Another advice is to understand what the customers are really doing, and to take time out to listen to customer enquiries. Crouch says the implementing team observed how customers are interacting with the staff at the call centres, and reviewed the regular reports on activities on the call centres. “I think that is the key.”

IRD worked with ScanSoft, which has already implemented NLSR applications at the Ministry of Social Development. “That’s one of the benefits of using an experienced company,” he says, “they have already done a couple of implementations in New Zealand so they are developing the New Zealand peculiarities of the language.”

IRD is currently looking at “more complicated areas” where NLSR could be implemented, says Crouch. “NLSR can handle greater scenarios than the push button IVR, and has the ability to deal with a broader range of inquiries and potential answers from the customers. We certainly hope we would be able to leverage that further into the future.”

Fielding calls at Centrelink

Centrelink is the Australian federal government’s service delivery agency, and provides access to a range of government services and information. When the federal government announced the working credit incentive, designed to encourage people to take up casual and longer term employment opportunities, the reporting requirements generated 200,000 additional calls every fortnight into Centrelink’s nationwide call centre network.

Centrelink chose to implement an NLSR system as a self-service option for people to report their incomes.

As of December 2003, Centrelink had an average of 68,000 working credit and student customers reporting their income through the natural language speech application each fortnight. In addition to completed calls, approximately 6 per cent of callers who complete the majority of their business using the service then transfer to a customer service officer in order to complete some additional business. These transfers are streamlined using Centrelink’s integrated CTI infrastructure and are presented to the customer service officers using ‘screen pops’.

Initially, Centrelink anticipated calls through the NLSR application would take an average of 180 seconds to complete while the expected call handling time for an agent was 300 seconds. Based on these figures, agent salary costs alone are almost four times those of the total cost of a self-service call, and Centrelink had anticipated achieving a return on investment within 18 months of full deployment. However, the unprecedented level of uptake of the speech service has led Centrelink to expect to achieve this in nine to 12 months.

Thanks to the project’s success, there is strong support within the organisation for additional speech applications and the potential for self-service within Centrelink. Employees now actively promote the self-service options and customer service managers indicate there is a potential to reduce customer traffic through face-to-face service centres and call centres by up to 50 per cent.

“To be honest, we are still amazed at how popular it has been,” says John Wadeson, general manager of Centrelink’s new business solutions. “The major transaction at this stage is we allow people to update their earnings. We’re looking at what else we can do with this technology. Most customers seem quite content to just use the system and get on with it.”

The human interface

Centrelink has found regular customers can rapidly adopt a new system and become comfortable with it, but infrequent callers often expect far more assistance than is available from a computerised interface. “People who might only contact us every three months are not as likely to use the self-help system,” says Wadeson. “If you don’t need to use your PIN regularly, then you don’t remember it or it gets lost or you think, ‘Well, I’ll just call up the call centre, I can get the same service and talk to a person’.”

Identifying customers who could use the NLSR service the next time they call is one way of increasing the availability of call centre staff for those who really need a human interface.

“Our staff in the call centres have been very good at promoting the service,” says Wadeson. “They take a very active line because our staff generally have the view they want to deal with the more high value work. They don’t see it as a trade-off for jobs; they see it as being better able to cope with people who we can really make a difference for.”

Because the application is focused on a particular business problem, misdirects and call dropouts are quite low.

“We seem to be running at fairly low percentages of people who just give up on the transaction,” says Wadeson. “When we first looked at the figures, we thought the numbers were too high but what we didn’t realise was that people ring up on the wrong day. If a person isn’t due to report on that day but they ring up anyway, the system shows the transaction isn’t complete. But if that person had come into the office or spoken to the call centre, they would’ve also been told that the legislation says to report on a certain date.”

Centrelink isn’t trying to use NLSR to replace every point of contact with its customers but sees opportunities for the technology to assist with other areas of the business.

“People are still able to talk to a call centre operator and are able to come into an office and many still choose to do that,” says Wadeson. “We’re not like a bank where there are different charges for different levels of service. For people who we think can use the service, we try and get them to use it so we don’t get large numbers of people who get rejected because the system couldn’t recognise them. Formal screening takes place.”

Speech technology applications have now become a key part of Centrelink’s customer service strategy, rather than an ‘emerging technology’ adjunct. As a result of this success, Centrelink has already embarked on the development of five additional speech-enabled self-service applications and is investigating the adoption of biometrics for access and authentication services, including speaker verification technologies.

Suncorp’s skilled response

Suncorp is one of Australia’s top 30 largest companies, comprised of Australia’s sixth largest bank and second largest general insurance group. Suncorp’s call centres handle 4.5 million calls per annum, assisting customers with products and services including banking, general insurance, lending, investment and claims. Suncorp customers’ enquiries therefore span the complete spectrum from basic information such as the location of a branch, to more in-depth service and sales queries about insurance policy benefits and policy information, claims advice and lodgement, personal and home loan applications, and wealth management products.

Suncorp decided on an NLSR program to replace the menu-driven telephone keypad push button conundrum its customers and prospective clients inevitably faced. The company aimed to provide a far more personable first point of contact for its clients and to deliver them more quickly and efficiently to their relevant customer service point.

“Speech recognition allowed Suncorp to provide its customers with an easier, faster and more efficient method of delivering the customer call to the best skilled, available agent the first time,” says Andrew Mulvogue, general manager, personal customer sales and service for Suncorp. “Speech allows our customers to just ‘say’ the enquiry type they have and be delivered to the skilled specialist agent for that call-type.”

After 18 months’ operation, the system is performing according to expectations, and the future possibilities of the technology are tantalisingly close.

“It recognises voices and it gets people to the right person the first time,” says Mulvogue. “The true benefit lies in being able to use the system to ask the customer a little bit more about who they are and understand who they are, integrate that with your customer information, and then present the information on that customer to the consultant who has the skills and also the performance record to manage that valued customer at the right level.”

Managing complex queries

Sending the call to the right consultant also keeps those highly skilled staff members free of basic enquiries and available to customers who can benefit from their skills.

“Getting to the right person with the right level of skills to manage the more complex transactions is the key to reducing repeat calls,” says Mulvogue. “The true business benefit is to get that information from the speech engine to our consultants. Initially, call centres were an efficiency-based solution whereas now that’s not what they’re about. They’re very much about business outputs and business drivers and customer satisfaction and customer retention.”

NLSR is one piece of the technology puzzle helping Suncorp’s staff spend more time with the customers who need their attention, while the simple enquiries are dealt with by the system without ever reaching the call centre. This has helped changed the perception of call centres from both sides of the telephone. Customers expect intelligent assistance from the moment the call is picked up, and call centre staff want jobs providing that level of service.

“Call centres were transaction houses but they’re not now,” says Mulvogue. “Now, they look holistically at the customer they’re talking to and they’re spending the time. And it’s not at all about speed and efficiency; it’s about business output and service delivery and meeting the customer needs.”

The ability to listen to whatever a customer says and then act on that information is the major advantage of NLSR over menu-driven prompts, but really getting to know who is on the other end of the line will allow Suncorp to take its service to the next level. “The great thing about speech recognition is that the customer will only say what they want. Our ability to understand what our customers are going to say and respond to that is what the solution is all about.

“So the reason you get such a benefit is because customers get it right. They don’t press another button thinking they’re going to get through faster. That opportunity is not there. They just say what they want. And it works. And it will be enhanced even more if we can understand who they are. And that’s the next evolution.”