“Internet time.” There’s a phrase you don’t hear much anymore.
Remember before Y2K when the Information Superhighway was leveraging the New Economy outside the box toward a tipping point that would synergize a paradigm shift in which dot-com companies could gain enough traction to get eyeballs and harness mindshare?
Yeah, unfortunately, I remember that too. Though much else seems to have fallen by the wayside, the attitude that we’re operating on “internet time” – that there is something unique about our current pace of change – has endured. But amid all the Sturm und Drang, one thing seems to remain stable in pay-per-click advertising for most keywords: the ‘Bid vs. CPC’ relationship.
The stability of this relationship can be verified, both with actual performance data and with Google’s Bid Simulator (GBS). For each word that gets sufficient traffic (about 20 clicks per week), Google gives estimates for 3 to 7 possible bids of weekly Cost and weekly Click volume. By dividing these values, we can obtain an estimate of the CPC at each of those bids. These numbers are updated daily based on all of the previous 7 days’ searches related to that keyword, regardless of whether or not your ad was shown.
The graph below shows all of GBS’s estimates as orange squares of ‘CPC vs. bid’ for a single two-word broad-match keyword from August 1 through December 31, 2010 (153 days, for a total of almost 800 bid-CPC pairs). This period includes the launch of Google Instant (September 8, 2010) as well as Thanksgiving (November 25) and Christmas (December 25). It also shows the actual CPCs, as blue circles, for each day. (In cases where the bid changed during the course of the day, the bids have been averaged accounting for the fact that traffic volumes are higher at certain times of the day compared to others.)
(The gray line marks where the CPC is equal to the bid. Due to the second-price mechanism of Google’s ad placement system, CPCs must fall below this line.)
Notice that the actual CPCs tend to fall along the range of CPCs estimated for various bids by the GBS. Only about a dozen days were exceptions, including Christmas Day, although most of these “exceptional” days were fairly ordinary: Friday December 3rd is among them, as is Thursday August 12th, for example. Thanksgiving and the launch of Google Instant were not.
Fitting regression lines to each data set, we find that they are nearly identical.
The fact that Google happened to provide multiple estimates at a bid of $3.60 (which has created a vertical feature on the graph) allows us to easily gauge the range of CPCs possible at that bid, with the lowest being about $2.37/click and the highest being about $2.74. So, were we to bid $3.60 for this keyword and use the regression as a guideline, we’d likely know our CPC to within 10%. (In fact, the higher-end of this range occurred in August, the middle in September, and the lower end after that. So, if we took the current date into account as well, we’d be able to estimate our CPC even more accurately.)
Since the GBS’s estimates are derived from the previous 7 days’ search behavior, it might not be surprising that they would resemble actual performance so closely. A more interesting investigation would take Bid Simulator estimates from earlier in time and compare them to actual performance much later in time.
In the graph below, GBS estimates were taken on August 8 (covering the period from August 1st though the 7th) and a regression line fit through them. These GBS estimates covered a suggested bid range from just under $2/click to about $5/click (not shown). The keyword was a six-letter term ending in ‘s’, broad match, and could be interpreted as either a plural noun or a verb (like “dances” or “pilots”). Actual performance data is also plotted for the months of November and December, and a regression curve was fit through these points. Again, the two regressions are in close agreement. (In this case, Christmas and Thanksgiving Days were not exceptions.)
Please note that there is nothing special about the date in August I have selected nor the months of November and December. When all of the GBS estimates are plotted along with all of the performance data, the results look much like in the first graph. The lesson is: essentially any single set of GBS estimates of ‘CPC vs. bid’ from, say, the month of August is still a reasonable set several months later, despite major holidays and the launch of Google Instant.
In this case, there was a period of several days throughout the August to December period in which the bid was set at $2.50, which allows us to gauge the variability in CPCs for a constant bid. This bid resulted in CPCs from $1.61 in late October to $1.90 in early December. So again, when the regression relationship is known, selecting a given bid specifies the CPC to within 10% (and even closer when the specific date is taken into account).
Above I have only shown two cases, but I have looked at literally hundreds and the results are essentially the same in each one (though there are cases where the CPC variability is larger). So, when it comes to the ‘bid-CPC’ relationship for bidding on keywords, “internet time” simply means that things are very stable over the period of weeks and months, despite holidays and major changes to Google’s algorithms and user interface.