Well, apparently Plenty of Fish did it!
So, for a first answer to your question, have to
'plan', that is, think of a Web site that can
attract that much traffic, at least from publicity,
viral effects, other network effects, happy users,
etc.
How to do that? Well one way is just to think of a
'business idea' (John Doerr said that business ideas
are easy and plentiful; bad business ideas are;
maybe good ones are more difficult and rare!),
develop a prototype, use 'lean' methodology and/or
the Steve Blank approach of continually 'iterating'
with the customers and revising the prototype to
achieve 'product/market fit', 'pivot' if this
doesn't work well, and keep trying.
I have another approach in mind borrowed from
project planning examples going back over 100 years.
A joke version is, a good recipe for rabbit stew
starts out, first catch a rabbit! More seriously
think of what one venture partner calls a "big ass"
problem. I would add detail: Want a problem that
is 'big' in the sense that we are sure that the
first good or a much better solution will be a very
valuable 'must have' and not merely a 'nice to
have'. To keep risk down, want no doubt about this.
If there is any doubt, pick another problem.
Another local, mobile, social, sharing app has too
much doubt. Similarly for another 'social graph'
recommendation site. From all I can see, mobile
payments also have too much risk from regulations
and need for 'critical mass'; if try this, then be
ready to jump quickly on the first good acquisition
offer.
The obvious example of such a "big ass" problem,
although not from information technology, is a safe,
effective, inexpensive, patentable one pill cure for
any cancer. Then don't have to worry about
'product/market fit'. Instead, a sad reality and
not at all a joke, have to hire security guards to
keep desperate customers from breaking down the
doors to the lab and trying to steal the pills. Why
don't we have such a pill? No one knows how to make
one. But the first person or group that figures out
how to make one and patents it will have a low risk,
first good or much better very valuable solution and
a very successful business.
So, right, we are getting a hint: For the success
we want, it might help to do some original research!
So, for step (1), think of a suitable "big ass"
problem.
Step(2). For this "big ass" problem, want to find
the first good or much better solution that will
clearly be a very valuable 'must have'. For the
very valuable part, in part want a barrier to entry.
For a cancer pill, could use patents. For
information technology might use Fred Wilson's
"large network of engaged users" (but apparently now
he is moving into mobile payments instead!), a
network effect (everyone uses it because everyone
else uses it, e.g., LinkedIn; nice to have; usually
tough to get started), a brand name, etc. Or have a
technological barrier to entry, that is, have a
solution too difficult to duplicate or equal. The
common claim that there is nothing new to permit
such is just not true. Uh, the technology might be
original and not on the shelves of the research
libraries -- right, that's commonly called
'research' and venture capital won't fund it,
evaluate it, review it, or even think about it and
instead will throw it into the bit bucket because
their backgrounds in bizdev, marketing, and selling
made them afraid and jealous of it; also they want
always to be "the smartest guys in the room" which
raises a risk of the other guys not being smart
enough to make money!
So, to get this solution, do something different for
recent information technology entrepreneurship but
nearly standard for much of engineering going back
over 100 years:
For our step (2), faithfully convert the real
problem into a mathematical problem, e.g., as in the
applied mathematics in each of most of the fields of
engineering.
E.g., notice that the wings don't fall off Boeing
747 airplanes. Why? A major part of the reason is
the applied mathematics of mechanical engineering.
So, from our step (2), we now have a clearly
stated mathematical problem.
Although we are in information technology, notice I
didn't say we have a computer science problem.
We're talking applied math, complete with theorems
and proofs and possibly with some advanced pure math
prerequisites, common in engineering going back over
100 years, e.g., to Maxwell's equations, but
recently rare in information technology startups.
Step (3). Get a solid mathematical solution to the
mathematical problem. Have two advantages here: (A)
Generally it is relatively easy to check such math
for math correctness. So, get lower project risk.
(B) The math approach can yield solution techniques
more powerful and too difficult to think of
otherwise. So, get a better shot at the first good
or much better valuable solution with a
technological barrier to entry.
Step (4). Write software to do the data
manipulations specified by the mathematical
solution. There are two advantages here: (A) The
math really should provide a
precise, usually succinct,
specification of
what the software needs to do. E.g., we
are not trying
to write an 'artificial intelligence' application
based on, say, 'rules' and Forgy's RETE algorithm
where have to keep 'tweaking' the rules until they
appear to work, i.e., keep throwing the software
against the wall to see if it appears to stick.
We
don't have to keep trying maximum likelihood
estimation of 'machine learning' to see if it
appears to stick. Similarly for neural networks.
Instead, if the software just does the data
manipulations specified by the prior math, then
likely the software is essentially correct, has done
its job. So we lower project risk. And such
software tends to be easier to write than the
common, elaborate user interface software. (B) Such
software is relatively easy to check for correctness
because of the fundamental advantage that we have a
precise, and usually succinct, specification of what
the software is supposed to do (the lack of such a
specification is the main problem in establishing
software correctness).
Step (5). Deploy the solution. Get publicity.
Likely have built into the problem specification
that want a lot of virality.
The steps (2) through (4) typically are
challenging and high risk. But there are two
advantages: (A) Typically these steps can be done
essentially just on paper, e.g., as an applied math
Ph.D. dissertation or engineering project planning
document. And, as for most applied math research,
the work usually needs just one person. So, the
cost, 'burn rate', is low. (B) The work, the
applied math and software, are relatively easy to
check for correctness thus lowering project risk.
If can't get through the fourth step, then return to
step (1) and another problem.
Back to step (1), picking the big ass problem:
Here definitely want no doubt. But 'social' is not
well understood so that for highly 'social'
applications we will usually have too much doubt.
More generally this 'way' of doing projects won't
work for all projects and would not have worked for
all successful projects in the past. Right. But
the intention that, once step (4) had been completed
successfully, the 'way' gives high financial return
at low risk for an appropriate project and enough,
broadly, to get the business progress we have in
mind.
So, that's the plan to get the page views!