Customer service isn’t the only application for these features; they can be equally valuable for employee self-service. BNSF Railway, for example, deployed an IVR system from Nuance  to gather arrival, departure, and freight information from train conductors in the field. The system currently processes approximately 600 calls per day, according to BNSF director of network support systems Shannon McGovern.

Speech technology allowed BNSF to deploy an automated system that would be immediately familiar to its conductors, who were accustomed to dealing with live agents. It also meant that BNSF did not need to invest in any new technology on the trains themselves beyond the cell phones and radios that were already there.

“In my opinion, we saved literally millions of dollars, between what we would have spent on equipment and also on implementation and training costs,” McGovern says.

Testing, testing...

Speech-enabled IVR applications can be highly effective at reducing call center costs, but they require continual monitoring by employees or contractors with expertise in the field. It’s a mistake to assume that someone who is proficient in building and monitoring Web apps is the going to be skilled at maintaining a viable VUI (voice user interface).

“Just like if you were building a Web site that would deliver customer service and be an expression of the brand for a company, you wouldn’t just have anyone do it,” observes Nuance’s Mahoney.

“If you go in thinking that an IVR is just like a Web page, you will lose, because VUIs are very different than GUIs,” says Jeff Fried, CTO of IVR testing and monitoring company Empirix. For one thing, according to Fried, speech is inherently more timing-sensitive than Web interactions.

“A slow Web page is much more tolerable than a slow IVR,” Fried says.

Another classic problem is what Fried calls an incomplete implementation. In these cases, customers might hear different voices at different prompts, or find themselves dumped into a certain menu level with no apparent way to escape. Still other systems might prompt users for an account number or other information, only to have them recite it all again to a live operator.

The key, according to Fried, is to thoroughly test any speech application before it’s deployed to catch these problem cases before the customer does. “Even a simple IVR -- with, let’s say, 20 different states -- to get good coverage you may need a thousand different tests,” he says.

The testing isn’t over when the application is deployed. Speech applications must be continually monitored and live calls should be analyzed to determine whether the software is meeting its stated objectives.

“That’s when you’re going to be able to increase adoption and increase automation in the application,” says TuVox’s Martin, “by understanding where people actually opt for automation, and particularly, understanding where they opt out and why.”

Call center in a box

Fortunately, not every speech application demands a costly development and testing cycle. Some are fairly straightforward; for example, internal phone directory and routing systems. “A directory is more of a packaged application,” says Nuance’s Mahoney. “That’s something that can be installed, maintained, and configured typically by the channel you purchased it from. There really is no application design to that.”

Even when faced with considerable custom development, however, companies have options beyond hiring expensive VUI-design specialists. Vendors such as Nuance and TuVox have begun offering sophisticated, higher-level development tools than were previously available. And recently, the industry has seen the emergence of more open speech-enabled systems, known as “voice portals,” which make speech applications more accessible to technically proficient customers.