|AI: The next frontier of CRM
|05 Sep 2001
intelligence isn't just for Haley Joel Osment. In the future,
automated agents might answer customer phone calls and carry on
conversations, much like human agents. Unlike Osment's film persona
in Steven Spielberg's "A.I.," these agents will only be programmed
to talk, not to love. Kevin Dowd, president of East Hartford,
Conn.-based Brainhat, a company that creates natural-language
interfaces, discussed AI in an interview with searchCRM Assistant
News Editor Christine M. Campbell. Dowd said artificial intelligence
can automate kiosks, switchboards and help desks, although he
recommended an enterprise not base its entire future on
|Kevin Dowd, president of Brainhat
|Can you explain a little
bit about artificial intelligence software?
|Ideally, what you want to do is get some
natural language ... and parse it into a data structure, which is
essentially like a diagrammed sentence. When you get it in that
shape, then you can manipulate it. You can turn a statement into a
question, you can compare something that you learn with something
you may have learned before and you can generate inferences. If you
do this with enough different sentence structures and with enough
different inferences, suddenly you have software that appears to be
intelligent and able to hold a dialogue with you.
|What are some of its
|If it can hold a dialogue with you over
some limited domain, like repairing something or ordering tickets,
then it can serve as a more natural front end for things that people
do. It's not going to be the same as hanging over the back fence
with your neighbor or talking to the barber. You can do quite a bit
with limited domains, where you tell the software what it is that
the conversation is going to be about, and you give it some goals,
and in a nonprocedural way it will find its way toward the answers
that you are looking for, like what's your name. That would be a CRM
sort of application.
|Can you elaborate on the
|The idea is that the software can be a
clearinghouse for language-based events. That means that English can
come in and English can go out. Other things can happen as a side
effect of things learned or things deduced. That could be that a
request is made, or a piece of mail is sent. One of the voice
interfaces (is) VoiceXML.
|It's Web markup for speech. There are a
lot of people pinning companies' futures on it. I'm not. I think
it's amusing, but I don't know if it's the end-all be-all.
The idea is that you can craft speech pages, and you can
essentially have telephony go through a server, which will do the
speech recognition generation part. On the back end, you can have
someone's Web server serving up these pages. You can replace the
more traditional ways for cobbling together CRM applications for
speech using some XML markup, which is more standardized.
|How did that lead to
|We said, "Here's this way to do speech
markup using your text editor. Why not generate those markup pages
dynamically so that you can have a dialogue?" We made the product
into a VoiceXML server. The idea would be that you could have speech
markup pages for ordering a ticket. If you ended up in some trouble,
you could have a more general conversation with something that knows
what it is you want to talk about. That would be Brainhat.
|How does Brainhat
integrate with an existing CRM system?
|CRM systems using VoiceXML are very much
like Web servers, where you can do a "get." The next link need not
take you to the same server. You can go somewhere else. It's really
not something that even has to be crowbared in.
|Where do you see the
|Artificial intelligence disciplines in
general, as soon as they become mainstream, they're no longer
artificial intelligence. Some of those are becoming mainstream, like
natural-language processing, (which is) very much an artificial
intelligence discipline, but people just put it as another feature
on a feature set.
With CRM applications, particularly the ones that are doing
speech or even type-text interfaces, you will see natural language
processing as a feature. In many cases, what passes for natural
language processing is really "word spotting" -- looking for cogent
stuff in a stream of text.
FOR MORE INFORMATION:
Best Web Links on customer service and support
to automate call centersAbout Brainhat
| Email a friend:
|In order to receive FREE CRM
related NEWS delivered to your e-mail, you first must