Economics of Ad Delivery

I would like to talk a little bit about the economics of ad (as in web advertising unit) delivery. Not because I am the biggest expert in the world, but because I have been wrongheaded about the issue in the past. I can (and have) been found guilty of advising people not to upgrade to additional servers (here), that there is more performance to be had out of their existing servers, that it is expensive, blah blah blah. That isn’t to say my advice isn’t good — its always good to get the most out of your machines, but the motivation is wrong; basically it comes down to that I am stingy. I didn’t want to upgrade & add new machines because who wants to spend an extra $100/mo on some marginal improvement in site speed?

Well, I’m here to tell you that is wrong-thinking. You should spend the extra money, and here’s why:

To make the math easy (i.e. not MY earnings numbers), let’s just suppose that you make about $5,000 per month from your website due to advertising. Even if you make no money at all, bear with me. Advertising depends upon page and ad delivery speeds. Being able to serve more pages to users faster equates to more ad delivery, which equates (almost linearly) with more money. So lets play the pretend game. Lets pretend that adding a server will make your site 10% faster to the visitor. Now it’s a stretch to claim that a faster site will result in 10% more visitors — that’s the result of long term marketing efforts. But it is entirely reasonable that a 10% increase in speed will result in 10% more page views. That is, 1 out of every 10 visitors just might look at 1 more page. At the current revenue rate of $5k/mo, that is an extra $500 per month. Your new $100/mo server just netted you $400 extra profit a month. In fact, at $5000/mo, your new server only has to be responsible for 2% more performance/ad delivery to break even.

Now think about your traffic growth rates. Over the past 8 years, SR has seen a pretty solid 20%/year growth rate, except for the one year we flatlined. Even though I had our server tweaked out of its mind, and it should theoretically been able to handle any added traffic, somehow, it didn’t. Then, I added a new machine and blamo! (who says “blamo” these days) the traffic growth returned. Turns out site latency and performance was holding users back.

Now just suppose that the slight sluggishness of your site costs you your 20% growth per year (you have different rates). If you also make $5000/mo, then the server performance is costing you .2*5000 = $1000/mo one year in the future. That means (not factoring in risk), you should be willing to pay almost $1000/mo in the present to insure your 1 year growth. If you put an 80% risk premium on your growth (not counting interest rates, which we’ll leave out to keep it simple) AND count that your growth is realized over n future years, you should be willing to pay (.8*1000 = 800) + (.64*1000 = 640) + (.51*1000 = 500).. ~= $1,950/mo in the present for performance. In reality though, you won’t be keeping the same server, and will do other optimizations each year, and will instead diversify your growth investment on other campaigns as well. We just want to illustrate the point that $100/mo for a server is trivial toy money. Sticking with our $800, we should buy 8 servers to insure our growth, and 5 at a minimum for the $500 10% gain. If you can manage it, by all means, go for 8 servers. How could I have possibly argued against upgrading?

Now with e-commerce sites, I wouldn’t go so far as to claim a linear relationship between site speed and sales, but I can bet you there is a relationship.

What am I not advising. I am not advising if you have no traffic and no earnings to go out and add a server. The administration, setup, and operating costs of replicated servers can be prohibitive for new webmasters — I am instead encouraging one to think of the relationship between site performance and earnings. How much performance can you eek out, and how much is your growth on traffic rate worth in present day dollars and present day performance?

*Note – this assumes that ad prices are paid uniformly and delivered at 100% fill based on impression/chain length. They of course aren’t, but this is simple illustration math.