'safety' in numbers for start-ups?

I was listening to an interview on Mixergy with Ryan Allis, an enterprising young man who built e-mail marketing company iContact into a company with >$3M in monthly sales, all by the tender age of 25. While this is a great accomplishment I couldn’t help but feel a little worried at some of the things Ryan was saying:

We very happily, intentionally burn capital every month. And we do that purposely because of what is called, ‘unit economics’. And we spend about 350, 400 dollars up front to get a customer that over their lifetime is going to spend about $2000 with us. So, if you think about that, that’s about 130% internal rate of return on marketing dollars spent, and we want to do that all day long.

a large part of the initial growth was a mathematical model around customer acquisition, figuring out how much it costs to acquire a customer, on average, figuring out how much a customer pays us per month, figuring how many months they stayed with us, figuring out their lifetime value, and then doing a calculation to determine how much we could afford in online marketing spend, to acquire the customer. And if we simply figure that out, which we did in 2005, right, when you’re at that 1.2 million in revenue level, then we could go raise venture capital all day long. Because the second you have a proven mathematical model that has inputs and outputs, it’s very easy to raise capital.(my emphasis added)

Maybe I’ve just been reading Nassim Nicholas Taleb’s books too much recently, but the mathematical models Ryan is talking about aren’t like V = IR or F = MA, they’re more like a predictive model based on the sales time series data iContact has gathered. Freak events or changes in market conditions (better/more agressive spam filters, the rise of other communication channels like FaceBook and Twitter) could change the models (and given that it’s 2010 when Ryan is saying this, and not “pre GFC” probably HAVE changed them). Since iContact has survived the recent financial troubles I’m guessing they’ve got this stuff all sorted out, I just don’t believe Ryan’s use of this predictive model is quite as risk free as it comes across in the interview. But then I’m a software developer, not an economist so what would I know…