Here are a cou­ple of com­mon ques­tions I get from dig­i­tal mar­keters: “Are my paid search key­words can­ni­bal­iz­ing my nat­ural search key­words?” And “How much should I really be spend­ing on my paid search bids?”  Believe it or not, using a pro­gram called “R”, you can actu­ally fig­ure this out using some sta­tis­ti­cal num­ber crunch­ing and Report Builder.  Here’s how to do it:

First, you need to first down­load how much you spent on your paid key­words each day over a long time period (prefer­ably a year or more).  The data should be read­ily avail­able through a dig­i­tal ad man­age­ment plat­form like Adobe Media Opti­mizer.

The dataset we need should look some­thing like this:


Notice that the total spend is bro­ken down by each day, and I’ve even split out branded vs. non-branded.

Next, using Adobe Report­Builder, down­load the total num­ber of vis­its, vis­i­tors, con­ver­sions, and any other poten­tially rel­e­vant met­rics, each subto­taled by day, and add it to the dataset we just cre­ated.  This will give us a mas­ter dataset that we can start mod­el­ing off of.

Here’s where the fun starts.  If you’re not famil­iar with R, I’d highly rec­om­mend becom­ing so.  It’s a use­ful and pow­er­ful open source sta­tis­ti­cal engine.  A detailed expla­na­tion of R is out­side the scope of this arti­cle, but Adobe Con­sult­ing is always ready to help if you need it.

So, let’s get started.  First we need to load the data into R.  You can save what you’ve down­loaded from Report­Builder as a CSV file and eas­ily import that file into R using the read.csv func­tion.  We’ll now use R to cre­ate a mul­ti­ple regres­sion model (bor­rowed from some com­mon econo­met­ric prin­ci­ples) that will esti­mate how much your paid search is can­ni­bal­iz­ing your nat­ural search:

The trick here is to include the total vis­its and other traf­fic related vari­ables in your mul­ti­ple regres­sion to account for the ebbs and flows of traf­fic and sea­son­al­ity.  You can imple­ment a regres­sion model with the R func­tion “lm”.  You’ll want to set nat­ural searches as the depen­dent vari­able and set vis­its, paid searches, and other traf­fic related vari­ables as your inde­pen­dent variables.

Once R spits out the results, you’ll want to take a look at the esti­mated coef­fi­cient for paid searches.  The coef­fi­cient rep­re­sents the model’s esti­mate of how your paid searches are affect­ing nat­ural search.  If the coef­fi­cient is neg­a­tive, that means that paid search is pulling away from your nat­ural search to some extent.  The actual value of the coef­fi­cient will give you an idea of just how much this is hap­pen­ing.  I’ve seen this value range any­where from –0.3 to –0.7, mean­ing can­ni­bal­iza­tion between 30% and 70%.

To find your opti­mal spend amount in paid search, we’re going to use a mar­ket­ing model called “ADBUDG”.  This model will cre­ate esti­mates of zero paid search spend return and paid search spend sat­u­ra­tion (read: how much return will I get if I spend noth­ing on paid search, and how much can I expect if money was no object).

Fit­ting this model is a lit­tle com­plex, so I won’t get into the gritty details here, but the out­put will look some­thing like this:

adbudg paid search

The fit­ted line rep­re­sents how many con­ver­sions you might expect (on aver­age) at a given paid search spend level.  It also gives you an idea of how fast you’ll reach your sat­u­ra­tion point.

Finally, with this curve, you can now esti­mate how much paid search is really worth to you.  If you have an idea how much your site’s con­ver­sion event is worth in dol­lars, you can cre­ate a profit curve using the model above:

profit curve paid search

These curves show you what your break-even points are for dif­fer­ent con­ver­sion profit mar­gins.  You can see that if my con­ver­sion is worth $150, I’d be will­ing to spend more on paid search com­pared to if my con­ver­sion was only worth $50.

So what have we learned here?

  • Site­Cat­a­lyst data can be used to deter­mine nat­ural search cannibalization
  • You can also fig­ure out how much con­ver­sion you might expect with­out spend­ing any­thing on paid search, and con­versely, the most you could ever expect.
  • If your con­ver­sion is worth a dol­lar amount, you can cal­cu­late exactly how much paid search is worth to you.

If this analy­sis seems too tough to tackle on your own, enlist the help of Adobe Con­sult­ing who can guide you through the process and have you back on track to paid search profitability.