[andrewfelix]

>but I'm sure that Visa has run the numbers and has a sound business reason for doing this.

As someone who worked for an advertising Agency with a big bank as a client, I think this is probably not the case. A marketing ploy usually starts with a reasonable idea and then slowly morphs into something horrible as it makes it's way through various departments; market research, legal, management etc.

This campaign probably started as something entirely different, was rejected by legal and then manipulated at the last minute by someone client side who realised they could force consumers to use their cards.

Some companies have tight, well researched, single minded marketing(eg. Nike). I don't think many banks do, and from what I've seen of Visa's marketing I doubt they do either.

[msellout]

As someone who gets paid to 'run the numbers,' I can tell you that many large businesses make their decisions based on internal bureaucratic politics rather than sound business reasons. The executive sponsoring of the analysis project can easily bully the analysts into agreeing or simply restrict information from them such that the available evidence matches the sponsor's opinion.

And sometimes there is simply no data. We can't estimate the value of a year of college to the average American. There are a vast number of variables involved in the flavor of Olympic sponsorship. There are a tiny number of observations. Sure, you could count other large events, but few if any are truly on the scale of the Olympics. And there will be differences according to the host city and the technology situation. The harder the problem is, the more your assumptions (or 'a priori knowledge' if you're confident) affect the answers. It's like Congress trying to estimate the effect of legislation on tax returns over the next 30 years.

[msellout]

I'm past my edit time limit, but I should add context to the assertion that we can't estimate the value of a year of college. There are some good methods for doing so, but none that stand out as the correct way. It's a classic hard problem in econometrics.