Measuring search engine marketing

When Clickshift takes over a search campaign, the clients often want to see an immediate result on our dashboard. They do - and the immediate result is almost always worse than where they started! WTF?! Why do people hand off campaigns to us?

It’s a problem of observation, inherent any time you make major changes in your search campaign. You immediately see all of the costs of the campaign, but you don’t immediately see all of the revenue. Not everyone buys the same day they click on an ad. Adam Riff has a good piece on this sales cycle, and how to measure it. (If you’re not measuring it, you are way underestimating the effectiveness of your marketing campaigns and way overestimating the performance of your other online channels.)

The observation problem is especially true when you use marketing optimization software like ours. We report only the revenue you can attribute to us, and if you have a reasonable click-to-purchase latency, that revenue will be lower at first – even if we exactly match your previous performance on our first day. (With most software, there’s some learning time, so although the performance far exceeds manual methods, doing so on the very first day of a campaign is quite rare.) Overall, your business doesn’t perform worse, but the daily measurement that most people do with search campaigns misses the conversions you’ll get tomorrow from the ads you bought today. If you only track first session purchase, this doesn’t matter (but you also leave a lot of money on the table!)

Below, I’ve used some client data to show how, even if over 50% of users convert in the first session, it still takes at least 10 days for ROAS (return on ad spend) observed after a change to reach 80 percent of the original ROAS (and again, this is assuming performance is identical on day 1, and it usually isn’t).

• “Percent of sales each day” is a measure of the sales cycle curve. How many transactions happen within 1 day of click? 2 days? Here, the client saw 56% of sales in the first session, 6% the next day, etc. (Mathematically: “Sales on day N” / “Total sales during period”).


• “Purchases that are ‘unobserved’” shows the ongoing sales from the previous ads, not attributable to the new campaign or vendor. These are sales that are still happening, but usually aren’t being reported in your search analytics, especially after a campaign change.


• “Observed ROAS vs. baseline” – given these other curves, this is how the ROAS of the new campaign looks over time, even if, from day one, it exactly matches the performance of the previous campaign.

On the graph, I used number of sales instead of revenue – the consideration curve gets wider when you use revenue (higher priced items have longer consideration cycles), which makes the ROAS performance look even worse initially.

If you measure a campaign out of the gate without understanding these latencies, you just make the downward spiral of search engine marketing much steeper.

Sorry for all the math. At least there was no law!

Related Tags: search engine marketing, online advertising, consideration cycle