Dan Kaminsky's Blog

DEAF AND DUMB

One of the core laws of information transfer theory is as follows: Humans
read faster than they can listen, but speak faster than they can write.
Voice mail vs. E-Mail provides an excellent illustration of this asymmetry:
It’s quite a bit faster to send twenty voice mails than it is to type out
twenty emails – but it’s much faster to read twenty emails than to listen to
twenty people’s voice messages, especially if there’s anything important in
that message that needs to be subsequently written down.

Visually organized systems, such as documents and web pages, end up being
more complex than a simple stream of spoken phrases to generate, but once
generated can be skimmed quickly for areas of relevance to the reader which
then may be focused on. Notably, the message is not time dependent-the
words stay where they were as the reader moves onto another segment.

By sharp contrast, the spoken word is nothing more than a stream of
vibrations in the air-the stream might be repeated, but it exists as an
understandable entity only because of our short-term memories. As soon as a
word is spoken, it disappears, and can only be retransmitted on demand, not
simply referred back to with a jump of the eye.

Thus the prime flaw with Interactive Voice Response, or IVR: It must speak
everything-slowly, linearly–and IVR is left struggling with a painfully
slow method for providing feedback to its users.

And yet, Since telephones are built to carry speech, one might assume. But
sound is only the most efficient input to information systems when our
spur-of-the-moment phrasings of a specific request, or even our standardized
vocalizations of a given demand, can be understood and interpreted among
many other possibilities.

In other words, speech gets its efficacy from being able to refer to
anything at anytime–in exactly the way the twelve touch-tones most IVR
systems use to receive input from a user can’t. Those twelve tones–zero
through nine, pound, and star–can only carry a limited meaning at any given
time, as the ephemeral nature of speech relies on little more than short
term memory of the user to remember which buttons will do what.

Indeed, more than a few IVR systems at places degrade to a simple yes/no
relationship: “Is this what you want? Press some key. Is this what you
want? Press some other key.” A user waiting for a final option may have
to wait minutes before the last possibility sprouts up — indeed, the
slowness of providing information to the user slows down the ability for the
user to provide information to the system!

A web-based system provides information visually; IVR requires it to be
spoken slowly. A web based system, though not able to recieve information
audibly, can at least provide a persistent visual list of messages (through
“links”) it can service in a compact and quickly parsed index known as a web
page. An IVR system, in contrast, can only depend on the memory of its user
to store which numbers link to which meanings — and users can’t remember
much of anything.

Bottom line: A web based system is silent but deadly in its efficiency–
IVR is just deaf and dumb.

DEEP IMPACT

There are no clearer signs of the limitation of IVR systems than the fact
that it is not a closed system. One cannot service their entire class needs
through the telephone system; among other things, one cannot query the
telephone system for available classes and their code numbers! Instead, one
must receive a class schedule for a given quarter — a difficult proposition
when one is out of state or even out of country — and using the numbers
from that, register for classes.

If people were to use T-Reg for DISCOVERING classes, it’d be so slow that
the entire system would bog down under the strain. Clearly, the carrying
capacity of the system is limited if making it a complete solution for
registration — something you’d expect Telephone Registration to be — would
destroy it.

One of the primary reasons using T-Reg for class discovery would damage it
so severely is that of limited resources. T-Reg uses circuit-based
telephone technology, which must dedicate significant resources –one of
sixteen telephone lines –, to any caller in the process of registering.
Already, with students providing little more than class ID numbers, these
sixteen lines are exhausted quickly, and access must be rationed out to
prevent system overload. This rationing is not entirely effective, and busy
signals from all circuits being busy are quite common.

In contrast, web based systems are packet switched and stateless. In plain
english, this means that unless somebody is actively sending or recieving
information, there is no live connection between them and the server. Even
as much more information is sent visually to the user to read through, as
the user is reading, the server is quite free to service another user using
the same resources. It is thus much more difficult to overload such a
system.

It is also much less expensive to increase or even maintain capacity for an
web-based system. Phone lines for businesses are generally charged by the
minute, on the concept that incoming calls generate revenue. Sixteen lines
under constant use are not likely to be inexpensive. While data links for
web services are not free either, the minimal amount of data flow inherent
to a web-based registration system means that the cost of such a system —
both to establish and to expand — is almost completely independent of its
communications medium.

Web bits are cheap. Web servers are too — but the custom hardware and
software used for most IVR systems are not widely sold, which means
economies of scale and competition have had little effect on them. Moving
to a web based system may indeed save significant amounts of money in terms
of simply limiting the need to upgrade the existing telephonic interface.

WEAKNESSES

There are some costs to a web based system that should be noted. Web
systems do require more advanced resources from the client; while simple
touch-tone systems are ubiquitous, fully web-enabled desktops are less
common. It is therefore advisable to keep both systems available, with IVR
being deprecated to a backup.

Security is also a threat: Web systems are much easier to manipulate than
the telephone system, and the number of individuals with the means to attack
a web-based registration system is far higher. This risk can be mitigated
using generally available encryption and authentication techniques, and
exposing servers to regular security examination. As with any expansion of
user access, there is indeed risk that a non-user will assume control. The
best one can do is mitigate that risk.