Nexidia: A solution which could change the way we work

Written By:
Content Copyright © 2006 Bloor. All Rights Reserved.

Every year I see products that are quite neat and I like. Every
couple of years I see something which impresses me. But it is only
rarely that I see something that makes me think “Wow I need
to rethink things, just imagine what I could do with that!”
Nexidia is one of those handful.

We are all aware of the vast quantity of data that exists from
recording interactions in call centres and the like. We are also
aware that voice recognition is somewhat limited, you have only to
look at Sky News in any reception lobby to see that it is not that
fast or that accurate, although I have to admit that when Tony
Blair becomes Tony Blur it might actually be an improvement on real
life! But what Nexidia does is to open up all voice data into a
source of structured data that we can analyse at vastly faster than
real time.

The problem with voice recognition is that it attempts to
recognise words and as we know the Oxford dictionary is a very
large tome, and we also know that people have different dialects
and various impediments to perfectly clear speech which makes that
a very daunting tasks. With so many options it is extremely
difficult to structure voice recordings, they are all but
inevitably unstructured BLOBs, as far as analysis is concerned.
This makes analysis slow and manually intensive.

The people behind Nexidia are academics from Georgia Tech who
decided that there had to be a better way of thinking about the
problem. Their answer was to follow the lead shown by opera singers
and the Muppets. An opera singer who, for instance, does not know
German learns his part not by learning the words but by learning
the sounds. Likewise we all know that the Swedish chef in the
Muppets was not really speaking Swedish but it certainly sounded
like it. The basis of this approach is phonemes: the sounds from
which all language is constructed. The advantage of this approach
is that whilst the English dictionary is vast, the phonemes of
which English is constructed is limited—to be precise it is
44—and, indeed, nearly all languages are constructed by using
around 40 phonemes, from a pool of around 400, so the potential for
tackling all languages using this approach is clear.

Nexidia is therefore at the same time very sophisticated and at
once also very much a blunt instrument. It is clever and
sophisticated in that it can detect phonemes, but like the Muppets,
or our Opera singer, it actually does not know what lies behind the
sounds. But because of that essential simplicity it is possible to
structure the data: you are only storing strings of some 40 odd
possibilities, and you can analyse them many times—fifty or
more times—faster than real time. The analyst adds the
intelligence by asking the tool to find instances of a given word
or phrase; so you ask it to find where customers talk of
“rebate”, of “failure”, of
“dissatisfaction”, and it will go away and find them
really quickly. Instead of needles in haystacks you can now whittle
down the search and then listen to just the relevant recordings. So
just as when OLAP started we all learnt to slice and dice and
“furckle” around with our data, so with Nexidia we can
apply those same skills to voice data.

You will have already guessed that this tool has immediate
applicability for Security, searching voice recordings for the key
words that those charged with our national defence might want to
track down. But beyond the world of national security it has
unbelievable commercial potential. Today we can only listen to a
tiny percentage of the voice messages we tape, so we limit its use
to a bit of compliance and in extremis we may go back to it to try
and avoid a sticky problem, but once we can use all of it
meaningfully all of the time we can do just about anything. Think
of this: today we rely on a call centre agent to put down key facts
in a text box which we then analyse, but their notes are just
that—their notes; they edit and record what they want to be
the record but what if we can analyse the actual interaction? We
could start to learn how satisfied customers really are, we could
identify why some calls result in an up-sell and others fail, we
could spot trends emerging which are not even recorded in our
operational systems yet.

The only limits on what Nexidia could do for a business is the
limits that our lack of imagination imposed because we just have
not had such capability before. Remember when Xerox first invented
the copier? Analysts saw no future in it, because they could see no
one wanting more than 3 copies of anything. I believe, in a decade,
our use of voice analysis, through tools like Nexidia, will open up
ways to remain profitable and build loyalty which we cannot dream
of today. I shall be watching Nexidia progress with the keenest
interest and advising all of my clients to take a look at it.