The Customer Service Survey
VocaLabs' weblog providing news and commentary on the challenges of providing good customer service.
"Please say yes or no. Do you think this speech app sucks?"
Monday - October 17, 2005 04:35 PM in
Three articles popped up on my radar screen in the past few days about frustrations with typical speech-recognition applications today.
The first of these, an opinion piece in The Boston Globe called "Virtually Annoying," is a rant about how much the author hates speech recognition systems. This article came to my attention by way of the inescapable Walt Tetschner, publisher of ASR News, speech industry gadfly, and frequent critic of the use of "persona" in self-service systems.
Next up, a puff-piece in Inc. magazine about technology to improve speech applications; that is, new technology designed mainly for the purpose of fixing the problems often seen when deploying other new technology improperly. Kudos, by the way, to the PR agencies for ClickFox and Angel.com for placing their clients in this article.
Finally, a response to the Inc. article from the wonderfully-named (but too-infrequently updated) weblog, Mommy Why Do Speech Apps Suck? The gist of it is that the problems with speech applications don't need more new technology, they just need a better application of what we've already got.
Despite their different perspectives, all three articles are essentially saying the same thing: speech recognition technology isn't living up to its promise, not because of inherent shortcomings in the technology itself, but because of the way it is implemented. Even the rant in the Globe isn't complaining so much about the speech technology itself, but the way it is implemented with artificially perky personae and used to block callers from agents (which, as we all know by now, is a really bad idea).
In other words, speech recognition technology isn't being used to serve customers, and the end result is poor customer service.
This is not a novel observation, and it dates back to the first implementations of IVR technology. We can throw a few other observations into the mix from the research we've done on customer service automation:
The recipe for success in speech recognition isn't hard to figure out: allocate the time and expertise necessary to design a good interface, make it easy for customers to get to a live agent when they need to, and measure performance using appropriate metrics.
This recipe is exactly the same as for tone-based IVR. The difference with speech is that a badly-built speech interface is much worse than a badly-built tone-based interface (the flip side is that a well-built speech interface can be much better than a well-built tone-based interface). A bad tone interface will likely be awkward, inefficient, and annoying, but a bad speech interface simply won't work. Worse, misleading statistics can make a bad system look like it is performing better than a good system. For example, with a bad interface callers are likely to hang up and call back, which can look like multiple successful calls when calculating a containment rate.
Following the recipe isn't hard or expensive. It is simply a matter of priorities: if the priority is on providing good service cost-effectively, then hiring experienced designers and doing extensive testing will be easy to justify, and the speech application is likely to be a success. But if the only priority is saving money, then odds are the system will not be built by qualified experts using appropriate testing methods, and the resulting application won't work.
Posted by Peter Leppik
Next up, a puff-piece in Inc. magazine about technology to improve speech applications; that is, new technology designed mainly for the purpose of fixing the problems often seen when deploying other new technology improperly. Kudos, by the way, to the PR agencies for ClickFox and Angel.com for placing their clients in this article.
Finally, a response to the Inc. article from the wonderfully-named (but too-infrequently updated) weblog, Mommy Why Do Speech Apps Suck? The gist of it is that the problems with speech applications don't need more new technology, they just need a better application of what we've already got.
Despite their different perspectives, all three articles are essentially saying the same thing: speech recognition technology isn't living up to its promise, not because of inherent shortcomings in the technology itself, but because of the way it is implemented. Even the rant in the Globe isn't complaining so much about the speech technology itself, but the way it is implemented with artificially perky personae and used to block callers from agents (which, as we all know by now, is a really bad idea).
In other words, speech recognition technology isn't being used to serve customers, and the end result is poor customer service.
This is not a novel observation, and it dates back to the first implementations of IVR technology. We can throw a few other observations into the mix from the research we've done on customer service automation:
- Given a well-designed self-service option, customers prefer to use self-service when feasible.
- Many companies intentionally make it hard to reach a live agent, a technique which does little to increase automation, but tends to make callers frustrated and upset.
- Speech recognition systems can provide customer service just as good as live agents, but building an outstanding speech interface is technically much harder than a purely tone-based interface.
- Many common performance metrics, like IVR containment and call abandon rate, give incomplete or even misleading data on customer service quality.
The recipe for success in speech recognition isn't hard to figure out: allocate the time and expertise necessary to design a good interface, make it easy for customers to get to a live agent when they need to, and measure performance using appropriate metrics.
This recipe is exactly the same as for tone-based IVR. The difference with speech is that a badly-built speech interface is much worse than a badly-built tone-based interface (the flip side is that a well-built speech interface can be much better than a well-built tone-based interface). A bad tone interface will likely be awkward, inefficient, and annoying, but a bad speech interface simply won't work. Worse, misleading statistics can make a bad system look like it is performing better than a good system. For example, with a bad interface callers are likely to hang up and call back, which can look like multiple successful calls when calculating a containment rate.
Following the recipe isn't hard or expensive. It is simply a matter of priorities: if the priority is on providing good service cost-effectively, then hiring experienced designers and doing extensive testing will be easy to justify, and the speech application is likely to be a success. But if the only priority is saving money, then odds are the system will not be built by qualified experts using appropriate testing methods, and the resulting application won't work.
Posted by Peter Leppik
Posted at 04:35 PM by | | | |