You’re Spending Too Much Money With Google. Here’s Why.

I’m finally admitting to myself that I’m never going to land a cool job at Google. I’ve been sending singing telegrams to Larry for years (do you know how hard it is to come up with lyrics that rhyme with “Severts?”) and I’ve never received a single response. Can you believe that?

It’s okay, because I’m pretty sure I never had the right stuff for Google. First of all, I’m ancient by the company’s standards. I’m older than Sergey Brin. According to Silicon Valley lore, time didn’t start until Sergey was born, so I’m not sure what that makes me.

Second, I don’t live in the Bay Area. The Bay Area, if you haven’t heard, is where all the Smart People in the world live. Their success, they will be quick to tell you, is entirely a product of their merit. Luck plays no part. Just ask that PayPal guy whose name I can never remember—Peter something or other.   He travels the country telling us stupid people that we should try to be more like him.

As I have slowly clawed out of my deep depression over being old and stupid, I have realized that there’s an upside to this tragedy. I am now liberated! I can finally speak my mind about Google because I no longer have to worry about torpedoing my job prospects there.  So here it goes.

I don’t actually believe that all the Smart People live in the Bay Area, of course, but I can see why “Googlers” might think that. Why? Because the rest of us buy their Adwords keywords without using our brains, sending trainloads of unearned cash to Mountain View so that Googlers can continue to indulge in perks like ermine-lined commuting buses and break rooms full of free fois gras (from humanely-stuffed geese.)


ROAS is really GROAS

Here’s the problem.   When digital marketing people bid for keywords, they keep score for themselves with something called ROAS, which stands for Return On Advertising Spending. This sounds very precise and very smart, which everybody who uses it will assure you it is. And it elegantly requires that you input only two numbers: how much you spent on your advertising (s), and how much revenue the advertising generated for you (r.) Simply divide r by s and voila! You now know exactly how effective your advertising is.

Ah, if only it were that easy. Instead, I am going to argue that ROAS is not very helpful, and that in some cases it is dangerously deceptive. Allow me to explain why.

It’s always obvious how much you spent on your advertising (s), but the revenue that spending generates (r) is often difficult to determine. How can that be? After all, you know exactly how many people clicked on the ads, and it’s usually quite easy to determine what they bought from you (let’s keep things simple and assume you only sell goods online—physical stores create all kinds of new measurement problems.) So what’s difficult about it?

The challenge is in determining what part of that revenue was truly incremental, and what part was going to come to you anyway. To do this, you must estimate your base case, or what some people might call your counterfactual. By this I mean what would have happened in a parallel world where you did not purchase the ad.

When you mention this concept to Google’s salespeople (or, to be fair, any salesperson from any digital agency or network), they will look at you like you’re speaking Klingon. Part of their confusion is probably real, as it appears that they don’t get asked this question very often. And part of it, I suspect, is feigned, for even the salespeople at Google are supposed to be Smart People.

A Smart Person might find this analogy useful. Let’s assume that you and I own a donut shop. One day, let’s say it’s a Tuesday, we decide we want to grow our business, so we hire a neighborhood kid to walk around in front of our store dressed up like a Bavarian Cream. How would we determine if our money was well spent?

Well, we know what we paid the kid (s). That’s easy. But how much extra revenue (r) was generated?  We certainly wouldn’t count all our revenues for the day. Instead, we would take today’s revenue and subtract our base case.   What’s the base case? How about Monday’s sales? Or better yet, our sales from Tuesday of last week (donut sales probably vary depending on the day of the week.)

As stupefyingly obvious as this seems, the ROAS metric does not do this. Instead, it will use the equivalent of all of Tuesday’s donut sales.

If you are a digital marketing person, I can already hear you saying, “No it doesn’t! It only measures the revenue coming in directly from paid search! Organic search is tracked separately!” It is technically true that these sources of traffic are tracked separately, but the ROAS math implicitly assumes that all the paid search volume you receive is new, that none of it was stolen from organic search. This is completely false. Invariably some of that traffic was diverted, and in those cases, your marketing spending was wasted.

So ROAS is improperly named. I would propose that it should instead be called GROAS (pronounced “gross!”), for Gross Return on Advertising Spending.

Try ROAST instead

The metric digital marketers should be using is something I call ROAST, for Return On Advertising Spending–True dat!  I know this name is terrible, but the acronym sounds yummy.

ROAST will never settle for (G)ROAS’s distortion of the truth. It will always insist on estimating a base case so that we will know with more certainty what our true return on investment is.

The first step in calculating ROAST is to parse your keywords into similar buckets or to analyze them individually. I know this sounds insane, especially to those of you who are running five-million-keyword campaigns, but you’re going to see in a minute why this is necessary (and why your automated software platform could be leading you down a rat hole.)

The second step is to take that keyword (or tranche of very similar keywords) and estimate the likelihood that a customer would have clicked on your organic listing anyway (had you never placed the ad.) Let’s call this percentage (b) (a value somewhere between 0% and 100%.)

How do you estimate this probability? I can think of at least two options:

  1. Take an educated guess.  A good, honest effort will be much better than the (G)ROAS assumption of zero; so don’t be too hard on yourself. If you sell donuts online, believe you have a 20% share of that market, and your organic listing is in second place, then 20% seems like a fair estimate for a keyword like “online donuts.” If you want to be conservative, go with 10%.
  2. Experiment.  This is what the Smart People at Google would do! They would turn off a particular campaign for a defined period of time (for a week, a day, or even an hour) and measure how much of that paid traffic seeped over into organic search.

The third step is to calculate ROAST with this simple formula: (r(1-b))/s. Or, if you already have the (G)ROAS, just multiply it by (1-b). Using this math, an estimated b of 20% will reduce a $10 (G)ROAS to an $8 ROAST.

The superiority of ROAST: an example

So what, you say. Whether I use ROAS, GROAS, or ROAST, it all seems the same to me. Well, all these terms are indeed identical as long as you assume that b is the same across every keyword. But this is most certainly not true.

To illustrate, let’s use the example of the donut shop again, but let’s now stipulate that we only sell donuts online. Let’s assume that one of the keywords we have purchased is “Jeff’s Donut Shop.” Google, our platform software, and our digital agency will all tell us we should buy this “branded” keyword because its (G)ROAS is very high. The (G)ROAS is so high, by the way, because the people typing that term into Google are very inclined to buy donuts from us, so they generate a lot of revenue after they click on our ad.

But certainly you would admit that somebody who has just typed “Jeff’s Donut Shop” into Google is on a very specific mission. They know about our donuts and want to find our web site. So as long as we are in first place in the organic listings (where we should be if Google has done its job properly, which it almost always does), the customer is highly likely to find our listing and click on it.

In this case, I would argue we should estimate b at 95% or more. Again, it would be better (and reasonably easy) to determine this by suspending and restarting the associated Adwords campaign, but this guess will suffice for now.

If we’re right about b being 95% and we insert that number into our ROAST formula, we will find that a very high (G)ROAS of $20 for “Jeff’s Donut Shop” will translate into a ROAST of $1. Yikes!

In contrast, let’s explore a keyword like “online donuts.” This is not a branded keyword and as such will have a lower (G)ROAS. People who type this term into the search box are less certain about where they are going to buy their donuts. They will probably look at several sites, and they may be completely unfamiliar with Jeff’s Donut Shop, so even if they click on our ad, they are much less likely to buy from us than the first group, so their (G)ROAS will be much lower. For the same reasons, these people are much less likely to click on our organic listing.

Consequently, for “online donuts,” I would roughly estimate b at 10%.   This will reduce a $2 (G)ROAS to a $1.80 ROAST.

The table below shows you the wildly divergent answers (G)ROAS and ROAST will give you in this example.

Keyword                                  (G)ROAS                ROAST

“Jeff’s Donut Shop”               $20.00                    $1.00

“Online donuts”                      $2.00                      $1.80

If you use (G)ROAS, “Jeff’s Donut Shop” is ten times more productive than “online donuts.”   If you refer instead to ROAST, the roles are reversed, and “online donuts” is 80% more effective than “Jeff’s Donut Shop.” I’m sure you’ll agree this is a remarkably different picture.

Key takeaways

I would encourage you to take away two things from all this. First, ROAST will always be lower than (G)ROAS. Your advertising on Google (or any other digital platform) is always less productive than the (G)ROAS says it is (assuming all your return is generated online) because (G)ROAS doesn’t account for the base case.

Second, (G)ROAS in most cases severely overestimates the return on branded terms. This leads to some very bad decision-making. Software platforms that optimize for (G)ROAS will buy up branded terms first, sending money to Google that should have stayed in your pocket. At one client of mine, the software was buying up the entire market for branded terms because it was so enamored of the terms’ high (G)ROAS. Meanwhile, the platform passed up many juicy non-branded terms—because the budget had been drained in the pointless pursuit of branded terms.

Digital marketers who are managing their campaigns manually will usually make the same mistake. Even if they understand that branded keywords are incomparable to non-branded, they will often add them to their mix to “increase their ROAS,” often at the behest of their agency or salesperson. This is absurd, but very common!

Anticipated objections

Defenders of the status quo will probably object to these arguments in the following ways:

  1. They will say everybody knows that branded keywords are different from non-branded. This is generally true, but digital marketers don’t appear to truly understand the implications of the differences, leading to the bad behaviors I’ve described.  Critics who defend using ROAS reveal that they don’t understand the implications either.
  2. They will say that smart marketers knowingly overpay for branded terms so that they can block out evil competitors. This is the line I would use if I were a Google salesperson. And I’m sure in some cases this is a legitimate threat. But certainly one should wait until that threat actually materializes before reacting to it. And if one does react, one should use the proper ROAST math to determine the right level of investment. Someone typing your company’s name into Google is unlikely to be easily swayed by a small text ad from your competitor.

Most of these objections from critics will be swathed in layers of industry mumbo jumbo with a dollop of condescension. Don’t be intimidated. Just ask them how they’re accounting for the base case.

 A final exercise

If you buy these arguments and are starting to feel like a sucker, please don’t. Some of the largest, most sophisticated advertisers in the country are guilty of this (G)ROAS obsession. Take a quick look at the big wireless carriers. If you type their names into Google, you will discover that they all place paid ads right on top of their organic listings.

You’ll also notice that none of these companies are buying ads for their competitors’ branded keywords. If these viciously competitive spendthrifts aren’t trying to steal branded keyword traffic from their mortal enemies, do you really think you have a lot to worry about with your competitors?

One final thing. I can’t help but notice that as of August 29th, 2015, the keyword “YouTube” generates no paid ads on Google (see the picture at the head of this article.) The same goes for “Google maps” and “Gmail.”

It appears that some Smart People at Google understand ROAST after all.


Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s