Pay-per-Click Observer

It’s been a while, and so I’ll spew forth on the topic of forecasting just to break the silence.

The other day, a client said to me “we forecast with everything else, why not PPC?” That got me thinking. Why not? Surely, Google and Yahoo will have massive amounts of data from which to draw estimates? So I did some tests. I pulled forecasts from both places and, without getting into the details (mostly because I didn’t save them), they were completely useless.

Thinking about it, I came to understand the problem pretty quickly. How would they know how my ads will perform? Say I choose to bid $.50 on “foreclosure lawyers.” What kind of creative am I running? What happens if, as I almost always do, I change my bids every couple of days? What if a planeload of foreclosure lawyers crashes into Mount Rushmore?

There’s no way to know ahead of time all the myriad variables that can affect a PPC campaign dramatically. If my forecast is to be useful for anything, it has to be at least somewhat accurate. These vendor-supplied estimates weren’t even in the ballpark.

Typically, they overstated the results– drastically. I have to assume that this is done in the interest of creating, well, interest. They would like you to think that you have this massive well of inventory from which to draw for your campaign. The assumption that no one will come to your party if you throw it is a real wet blanket on anyone exploring the idea of a PPC campaign.

Still, I want to be able to give some kind of answer. The best results I’ve seen come from my own historical data. Since we keep all of our campaign data back to the beginning of time, we have our own, reasonably relevant, data from which to derive our forecasts. By performing a regression on that data (and adjusting it for known seasonal and budgetary non-linearities), we come up with numbers that, while still rough, are at least good “magnitude of order” estimates.

In most cases, that’s good enough. No one really expects us to be able to accurately predict the future.

In the case of new or test campaigns, though, I will stick to my guns and tell my clients that we just don’t know. We need to run them for a while and see what happens. There is just no substitute for truly relevant empirical data.