From Resource, April 2003 
Copyright by LOMA

Is Speech Recognition Technology Right for Your Company?

Speech recognition technology is becoming more popular as companies strive to find ways to cut costs and improve customer service.

By Tammy J. McInturff

Speech recognition technology is helping financial services companies save money and improve customer service. At LOMA’s Emerging Technology conference, Bruce Pollock, director of Strategic Technology Solutions at West Corporation, discussed several things a company needs to know as it begins researching speech recognition. He also shared some lessons learned in rolling out these types of applications.

Financial services has been a popular stage for speech recognition technology. There are also a number of life insurers that are rolling out speech applications, as well as many that are actively looking at it.

Natural Language Technology

Natural language speech recognition technology is different than what most people have encountered before in their workplace or home. You may have encountered the type of speech recognition that sits on a desktop PC, where you speak and your computer then types the words. This is quite different than telephone-based speech recognition technology, which is what many financial institutions are using today.

Natural language speech recognition is an evolutionary step from a single word or digit recognition. Single word or digit recognition is when the automated system prompts you to "press or say one." Natural language technology can recognize a phrase or sentence. So essentially, it allows callers to speak naturally into any phone. There is no training required, and it is easy to use and great for information and simple transaction calls. "A lot of Fortune 1000 companies are using speech recognition technology today," Pollock said. Airlines are also using it for flight arrival and departure information and speech is being used in virtually every other vertical market in some way or another.

Speaker Verification and TTS

Speech recognition technology is also different than speaker verification. Verification is using an individual’s unique voiceprint as an identifier. "Speaker verification is ready for prime time but not quite being implemented yet in full production in the marketplace—but it will be in the next 18 months," Pollock said. Your voice is more unique than your fingerprint, so some companies are examining this technology very closely as a means of authenticating callers.

Text to speech (TTS) is another type of voice related technology. This is when a caller can hear a typed sentence spoken back to them by a synthesized voice—which is actually based on a real human’s voice recordings.

The Guiding Force

What is influencing the growth of speech recognition? A trend to customer self-service is a key driver. It started with toll free numbers. Now the Internet is continuing to drive the self-service trend and as technology advances, it brings with it cheaper computing power. Smaller telephones have also increased the need for speech recognition technology since, it has become more difficult for individuals to press the numbers on keypads. The final reasons are that corporations need to cut their costs, improve customer service and get higher self-service throughput rates compared to touch tone, which speech is capable of doing depending on the circumstance.

Do people like speech recognition technology? "About two years ago, I was involved with an extensive focus group of utility industry customers," said Pollock. "We asked them their perceptions of speech recognition technology and showed them demos. Then we asked them how speech technology compares to touch-tone, which is what they were used to using. The focus group perceived speech recognition as fast and user friendly. The results were the same across ages and genders. Focus group participants also said they liked speech recognition because they didn’t need to associate their intent with a keypad. People get cognitive overload and they lose track of all the numbers listed to get to a particular function. So they press zero, which is why they want to talk to all of your operators and that costs a lot of money. Speech can also enhance a caller’s perception of a company. We asked the focus group participants if it really mattered whether this particular utility has speech or not and interestingly, they said it enhanced their perception of the company. They felt that the company was trying to make things better for the callers."

Speech recognition technology helps to deal with self-service transactions and keeps some customers from ever needing to talk to an operator. So, the company saves money. "Speech costs about one-fifth of a live operator call on a per minute basis," Pollock explained.

How many of your customers are really using the self-service touch-tone technology? "The bottom line is probably only between five to ten percent of all the callers are actually using touch-tone when they enter a company’s self-service system," Pollock said. "The rest of them are punching zero, which is why customer care costs continue to rise." The fact is, you can build good touch-tone applications and good speech applications as well, but it takes time, effort, expertise and financial resources.

Mini Case Study

Pollock discussed the challenges of implementing a speech recognition application for a financial services company. "The challenge that we had with one particular customer is that they had a joint partnership with a mailing house. During different times of the year, the company (and mailing partner) would send out letters to prospective customers saying, ‘You John Doe are eligible to be one of our credit card customers, dial the toll free number on this page to find out more and apply for the card.’ The company is—and has been—very successful and receives a lot of responses from these letters. The problem is that the transactions were costing them a lot of money to handle via operators. They wanted a scaleable model that was going to help grow their credit card business without incurring a large increase in costs. They didn’t want to have to build a new call center, nor did they have the money to do so," Pollock said.

The solution to the company’s dilemma was to front-end the first half of all of their credit card application calls with speech recognition. "We asked ten to twelve different questions, which normally would have been asked by a live operator. We asked a whole range of questions, from the customer’s employment status, to their home and business phone numbers, and social security number," Pollock said. "In this particular scenario, when the calls are transferred to an agent, the call data from the self-service session is attached. Many automated systems do not have this kind of data integration today—the data from the call is not passed on to the agent, which means the customer may have gone through the system and punched nine buttons, but when he gets to the call center, the agent has no idea why he is calling."

The Implementation Process

When implementing a speech recognition application there should be a pilot phase at the beginning where about 10 percent of the calls are skimmed off to the speech application for analysis and tuning purposes, before going live across the entire caller base. The application is tuned by monitoring it and examining how it is performing. In this stage, the calls are examined to see if the system understood what the caller was saying or if words need to be added to the system’s dictionary, which is something that can be done in very little time by skilled speech recognition grammar developers.

Once the major testing and tuning is done , the speech application is ready to go live across the caller base. Callers go into the application and are asked a series of questions. "Questions are asked in a well timed fashion, in the same pleasant voice that is moving them around the application," Pollock said. "In some cases, speech recognition can get the same job done as an operator, at a fraction of the cost. A key benefit of speech is that it offers the same, consistent caller interface for every caller, in every part of an application, every time. A well designed speech system can keep callers engaged in self -service and help a company save meaningful amounts of money." Indeed, well-designed touch-tone systems can often achieve similar results.

Additionally, speech systems can be designed to offer upsells—and the system never forgets to do so. "The savings average was about 22 percent per credit card application. There was a six percent reduction in the pilot phase, and an additional 16 percent savings following post-pilot improvements," Pollock explained for the example he gave.

The Pilot Program

The pilot findings were important. They discovered that callers do not enjoy hearing promotional messages in front of things. But, they do like advanced notice for specific information requests. For example, they liked being told at the beginning of the application that the system will be asking for a number on the top right hand corner of the application form. This way, when asked to recite that number they knew where to look on the page as opposed to having to fumble around while on the telephone. When not told the information needed beforehand callers often couldn’t find the page or location of the information on the page, so they would just go to the operator, which costs more.

In the case study Pollock discussed, they used several types of testing before going live with the system. "Our internal experts conducted call data analyses, internal focus groups, usability studies and Wizard of OZ (WOZ) testing. For usability studies, you get a series of people in the room and ask the caller to try calling the application to see how well it works," Pollock explained. "We did that actually before it came out live." WOZ testing in speech recognition is where you set up the mock application and have people call it. However, it is not the application on the other end that answers, it is a person that hears what the caller says and transfers them to the next part of the application. "So if the caller says something that normally speech wouldn’t recognize, the simulated WOZ operator would say, ‘I could not understand you, please say this or this.’ Afterwards we record and analyze that data," Pollock said. When considering speech recognition, you should make sure that the technology doesn’t overwhelm the whole process. In the case study Pollock discussed, a good portion of the analysis came from talking to callers about how they wanted the application to work. With speech, it is essential that you get caller feedback before rolling a system out and indeed, the same principal applies to touch-tone systems as well.

In the case study referenced here, several changes were made to the application based on results from testing. Script changes were made to help the caller better understand his or her choices. In some cases single words were swapped out in favor of other words that callers would understand better.

Issues to Consider

Many vendors claim that speech recognition can improve customer service and save you money, but is it right for your company? Touch-tone might be a good solution as well...and whether you should select one or the other depends on a range of factors. There are several things to consider when deciding whether to move calls from live operators to automated systems, or to make the move from touch-tone to speech. First, you should consider your business priorities. Are your customers satisfied with your service right now? Take a look at how your automated system is running and if it is actually working effectively. Do internal stakeholders support automation? Not everyone does. What is the end vision? Why and how would speech be incorporated?

It is important to consider these things as you look at speech recognition technology for your organization. Another thing to consider is whether to insource or outsource. For on-premise systems, it can cost around a million dollars to get started with a small system. So you have to have the capital budget it takes to implement the system, plus all the ongoing maintenance and tuning costs. Alternatively, you can turn to outsourcers for help.

Speech recognition is much different technically than touch-tone, the interface is different, and the skill set required to design, build and maintain a system is different. So this may be something that your company will have to outsource, not all IT departments have the skill set to implement a speech recognition application. If you plan to implement speech recognition, focus on the caller experience. Develop your prototype and support your pilot testing. Allow plenty of time for a pilot and to talk to you customers.

Communicating for Success

Communication with both your customers and your staff is essential to having a smooth speech recognition rollout. "Communications has two components and most of the time organizations forget about both of them. First is internal communication," Pollock said. "As you consider speech recognition remember to make sure that everyone in your organization is made aware of the upcoming change, especially those involved in the customer care side of the organization. Speech recognition is a steep change."

External communications is very important as well. "The key audience with external communications is obviously customers and the media. Use the Internet, in-IVR system voice messaging and written communications to inform customers and others of the change," Pollock said. "One company that implemented a biometric voice verification application failed to inform its customers of the change in advance and when the application launched customers had no idea what they had encountered. The end result was that the company eventually had to roll back its automation plans because it did such a poor job of communicating to customers about the new system. A lesson learned here is that as long as you inform customers about the speech system, they will be more effective conducting their business in self-service than if they don’t know what to expect at all when they call."

As you consider speech technology, it is really important to scrutinize all of the vendors that sell or rent speech applications. Ask if they have done mission critical speech applications, such as health care transactions, secure banking transactions or insurance transactions. You may also want to ask if they have phone numbers for those that you can call. Sometimes you can’t get these numbers because they may be internal numbers for validated customers only. Also, ask for customer references. You want to deal with vendors that can show you what they have done. Ask the vendor if they have built speech applications that also involve complex back end integrations and Computer Telephony Integration (CTI). "CTI is complicated but it is very worthwhile. A speech front-end combined with CTI ensures that when the call goes from the speech system to an agent for completion, the agent does not have to repeat the questions all over, they can see what the caller has done in self-service and pick up the call from that point," Pollock said.

You also will want to ask vendors about their procedures for ensuring that they build caller-centric applications. Have them explain the types of testing they do, including WOZ testing, focus groups, etc. and what they do with the data they collect. The data collected from testing should be used to improve or tune the system. You also may want to ask if they plan to assist you with training your call center agents and other employees on what speech recognition is and how it is being used. Whether you are dealing with your internal IT group, or an external vendor, you want to make sure you have all this support, including the ability to monitor and tune the speech applications. "You also want to be sure that the vendor is financially stable," Pollock explained. "You need to ensure that you have stable vendors in this difficult business environment—do some strong due diligence work and select vendors or outsourcers whom you are certain will be around in the next five years and well beyond."

Above all, focus on the caller experience. "It is not just the technology; it is about the caller experience, which is actually the single most important thing. The mistake some vendors have made in the past is that they designed automated systems that focus 100 percent on the technology as opposed to the voice user interface. If you build the speech application around the caller, it will work well. As long as you continue to tune and maintain the system, you are going to find you have a great combination of self-service technology and live operator service," Pollock said. "Generally, the calls that speech can handle in a call center environment are information and transaction calls. Speech can’t do complex problem solving (the third major call type in a call center); that is best left to operators."

There is some active research going on right now to see if claims information and coverages can be automated. There are consumer applications, but there are also business applications. "Business to business applications are equally, if not more powerful, because you have frequent callers who know how to use the system efficiently because they are interacting with it frequently," Pollock said.

So is speech recognition for your company? If you have large call volumes and are looking to cut expenses and improve service, it might be the answer you’ve been looking for. Speech recognition is becoming increasingly popular and is well received with customers. There are some insurance companies using speech recognition technology today, as well as airlines, brokerage houses, wireless carriers and other companies. Speech recognition can be used for anything from simple transaction calls to claims, and can also be used to "front-end" or triage more complex calls that eventually get sent to operators. It can also be used, when combined with reverse directory lookup functionality, to allow customers to advise you of a change of address and to allow you to send marketing materials to callers by simply having them speak their home telephone number to the system. The bottom line is speech has a lot of benefits, including cost savings, revenue generation and improved customer service. Whether speech or touch-tone end up being the best way to automate calls, insurers, financial institutions and other companies will save money either way.