Outsourced IT development Differentiator

Just musing if I'm a CTO/Start-up or a business owner, looking to outsource some/all of my technical effort. How should I go about it? What is it that helps me in deciding where to go, whom to entrust such a critical piece of my business?
On a Weblog, They suggest what to do in contract to make it legally binding, but not much on how to decide. A lot of literature is also there, to tell you, when or what to outsource, but whom to? Forrester and other such market research organization continue to publish studies that tell you that this business is close to several billions, and still increasing. Still, very few guidelines on how to select a IT vendor. Now, vendor selection as process is beaten to death, but in this case the (new development) success rate is abysmally low. I mean what data I should seek from a vendor to evaluate their maturity, is it instincts based, is it subjective and only experienced veteran can make a good decision? Now, don't get me wrong, I understand that "it depends", essentially on ,what I'm looking for. But why should I choose TCS over Infosys or vice versa. is it only cost arbitration?
Ok, let me look from the vendor eyes, If I'm CEO of a outsourced company what should I do to differentiate my company from the others. A differentiation that is solid, covers a broader spectrum and is obvious. A differentiation that let's me charge a premium on my services.
It could be a niche, which is ok, but in these days of open source, Internet a long term niche is hard to maintain, more over if this is a new niche, it'll take a lot of time in educating the market.
Let me ask this again, what could be the next disruptive thing that can change this market.
If I'm a US based company then how do I compete with outsourcing destinations in India,China and others. What I could do that is hard to emulate and gives me long term scalability?
huf ...., too many questions or is it only one?

Tell me how u measure them and I'll tell u how do they behave.

Very Old saying, isn't it??
Essentially measurement is the first step in defining goals, and sometimes the most important one.
Think about it, we measure distances in miles/km and a lot of transportation is measured in 2 paradigms
1. No. of KM a thing need to travel.
2. Weight
if somebody adds the time as part of the measurement, wouldn't that be great. Actually a lot of courier companies just did that and they charge a premium for it.(i.e. they take less time).

Being a software professional, I was thinking how do we measure people. Let me ask a more generic question,why do we want a performance measurement.
I mean if I work in a software production factory(I don't believe its anything else), how to reward people for good work and how to define good work.
I think a measurement should do the following
1. Tell the shareholders how the company is doing.(I think that is related to earnings?)
2. Tell a customer how well people in a company do there job?
3. Tell a COO /CEO (whatever) how is various units/deptt. doing and what he needs to do to improve/fix things.
4.Tell managers how teams are doing.
5. Tell team leaders how members are doing.
6. Tell individuals how these guys are doing.
Essentially quiet a few of these questions are answered in what is called Performance Management Systems.
At the higher management a lot of time is spend in deciding what a PMS should do. And after 10 years in Software Industry I think at the developer/individual level this is what we measure.
1. Analytical Abilities
2. Leadership qualities
3. Knowledge
4. Performance
and some other things. Normally these are measured on the basis of indirect weightage being given as answers to different questions.
After measurement overall numbers are normalized and whole company is divided into groups of V.Good, Good, bad etc.
Now, if we look at the criteria most of those are subjective and based upon the perception of the appraiser. If the measurement is based upon the "Perception is reality" , I think this isn't hard to guess how people will behave.
A lot of people will make sure that they stay in good books of the appraisers.
So let me understand that, most of the normalized curves in IT industry are biased towards people being good. May be 10-20 % V. Good, 60-70% good and 10-20% bad. Wow, so, We work with good people.
Let's consider another axiom of Software Industry.
"60-70% projects fail", because of different reasons.
What !!, 60-70% people executing projects are good, but 60-70% projects fail.
Why would that be? Could measurement be one of the reasons?
Will dwell upon it in further posts.