I define "usability" as a perceived ratio between effort and benefit. It's
a ratio, not an absolute.

**** Long Explanation - Skip to Next Section if Desired ****

"Effort" is work, frustration, confusion, pain, errors, cognitive load or
dissonance, among other things. "Benefit" is pleasure, useful work
accomplished, entertainment, satisfaction.

When reasonable effort results in pleasing benefit, we think of an artifact
as "usable". When little effort produces a greater than expected benefit,
we smile. We smile also when a desired benefit is attainted with less
effort than expected. It's all about the favorability of the e:b ratio.
Infinite usability is getting something for nothing. Joy is our response to
extremely favorable e:b ratios. It seems almost biologically-based & has
thought-provoking analogs in nature, for example in foraging theory (which
is partly about optimizing energy input:output ratios).

Adding features to a product increases benefit. But also effort. Each
feature can be thought of as its own little "e:b" ratio. The problem is
that when we stack up many features, additional effort is required just to
deal with the resulting complexity of discriminating between them. And that
effort has no corresponding benefit. It's pure overhead. That's one reason
why benefit tends to grow linearly, while effort doesn't.

This can be demonstrated by scrambling the menus of a software application.
Or the instruments of an airplane.  The same features are there, but they'e
harder to get to & use, at least temporarily, sometimes permanently (if
you've done a good job of making things inconvenient). This additional
effort is an illustration of the additional overhead complexity creates all
on its own.

There are exceptions. In natural (and other) languages, effort tends to
grow linearly and then plateau with fluency. Fluent speakers, knowing the
vocabulary, grammar, and idioms of use, now have access to a well of
unlimited possibilities of expression that will serve them far into the
future. Software features which are synergistic tend to follow this model
because benefit grows non-linearly, due to the possibility of combining
features in new and flexible ways. (UNIX-heads love this.)

Another example is network effects. Metcalfe's Law states that the value of
a network grows exponentially with the number of nodes. Considering the
SPAM in my in box, I'm not sure this is always true. But it sounds good.
And I think there's something to it. If you can exploit that in your
product, go for it.

Multi-user applications have their own usability (e:b) characteristics.
Grudin wrote an excellent paper on this pointing out the failure of
groupware apps in cases where those individuals putting forth the effort
are not the same as those receiving the benefit. The total e:b budget may
be the same, but the distribution of e:b units is out of whack and the app
fails. Credit cards, weight loss programs, political systems, they all have
their own e:b curves by which the behavior of groups of individuals can be
predicted and to some extent controlled. Have you ever considered the
"usability" of a political system?

One key thing about the e:b ratio is that it fluctuates over time. We may
start out feeling a product is promising, only to sour on it after hours or
days. Or vice versa. We all understand the difference between
"learnability" and long-term usability. They're simply different e:b
curves. Trust, product loyalty, and other factors are largely time-based
because they are about our relationships with artifacts or companies, our
history of effort & benefit, and our subjective estimates of the future
costs, risks, and benefits likely to occur in those relationships.

Effort and benefit often don't accrue simultaneously. People are sometimes
willing to exert effort up front for anticipated benefit. (But often they
are not.) It depends on the person and the context. It's called "learning".
A "seductive" interface draws the user in with a tantalizing benefit, and
each subsequent expenditure of effort is rewarded with an increased
benefit, encouraging yet more effort, rewarded with yet more benefit, and
so on. We don't talk about seduction as much as we should in user interface

So for me, "usability" takes into account people's expectations, goals, and
whether they generally feel their interaction with a product or service was
(or is likely to be) worthwhile. Like buying a product, it's often not the
absolute price paid, but the perceived ratio between cost (effort) and
benefit. It's often gauged subjectively although there are ways to measure
certain aspects of it.

This explains why product re-designs can be dangerous. If you give 20% more
benefit, your users may still perceive the added effort required to learn
the new UI as "twice as hard". You have to manage those e:b curves so that
the effort always seems "worth it".

You can do an effort : benefit analysis of a product or UI design. It's
like doing function point analysis. You estimate/quantify the benefit units
and count & attempt to quantify the effort units required to obtain each
benefit. Some designs can reduce effort by an order of magnitude, while
still providing the same or similar benefit. That's cool when it happens.

**** Nielsen's Definition of "Usability" ****

In his book "Usability Engineering" Nielsen defines "usability" as part of
a tree of overall "system acceptability". I'm not sure whether  this
originally from Grudin, or Nielsen. Either way I like the definition. It's
straightforward, models the domain well, and lends itself to objective
measurement. I try to reconstruct Nielsen's tree from his book, below.
(Indentation in the tree below should read thus: "practical acceptability"
is made up of "usefulness," "cost," "compatibility," "reliability," and
"performance". "Usefulness" is in turn made up of "utility" and
"usability". And so on.)

system acceptability
     social acceptability
     practical acceptability
                 easy to learn (e.g. measured by time to learn a task)
                 easy to use (e.g. measured by time to compete repetitive
                 easy to remember
                 few errors (e.g. measure by error rate)
                 subjectively pleasing (e.g. measured by post-usability
interview / questionnaire)

In the book Nielsen talks about techniques for measuring the various
components of "usability". It's a good read.