How does Google calculate ROAS?

"Here's the math: $5 in sales ÷ $1 in ad spend x 100% = 500% target ROAS" (Source:

Before I explain why ROAS kills profit, you need to understand exactly what ROAS means in Google AdWords. I have a complaint about the nomenclature as well as the math. Return on Ad Spend is a term the folks at Google can define however they want. They have defined it as “sales divided by ad spend”. But in finance, return is widely understood to mean the profit returned in addition to the initial investment. So if you invest $1, and you get back $5, that’s not a 500% return.  That’s a 400% return.  You got your initial $1 back and $4 additional dollars, for a 400% return.

If you’re not aware of this nuance in how AdWords redefines this finance convention, you will miscalculate and end up spending more on your advertising than is justified by the actual returns. I cannot help but question why they would have defined the term this way, and have come across several examples of Ecommerce sites being misled and overspending as a result.

On top of that, calling revenue a “return” ignores the simple reality that the true return on your ad spend would have to be calculated after accounting for your cost of goods sold and any transaction costs. In the above example, if you had a 40% gross profit margin on $5 in sales ($2 in GPM), then your financial “return” after accounting for those costs and your ad costs would actually be 100%, not 500%. That might change your decisions a bit.

Declining Marginal Returns

That out of the way, the main problem I have with Target ROAS bidding is the same problem I have with Target CPA bidding.  Both are based on averages and both are in the context of a system where there are declining marginal returns. Making decisions based on the averages of declining returns results in incorrect conclusions.

Here’s an example from Google’s Bid Simulator showing how raising bids to get more clicks (and conversions) will cost more than the average of the cost at the current level. Lowering the bids will reduce the average cost:

Bid Simulator

In this example, we currently bid $2 and get about 40 conversions per week at a cost of about $1340 per week. Our CPA averages $33.50.

If we were to drop our bid to $1.87, Google estimates that we’ll spend about $969 per week and get 33 conversions, for a CPA of about $29.36. But if we raise our bid to $2.52, we’ll get about 46 conversions at a cost of about $1890 per week, for a CPA of $41.09. This illustrates clearly how we get declining returns as we scale up. Our next conversion in online advertising almost always costs more than our last conversion, all other things being equal.

A Simple Example

Imagine a simple world, where there are only three people searching for “widgets” and they’re all going to convert.  Let’s say you sell widgets for $20, with a $10 marginal cost and a $10 marginal revenue. Suppose the three possible conversions will cost you $7, $9, and $11 respectively.

You can imagine the Google Bid Simulator for that auction.  It would look just like the picture above, showing how you could get about 1 conversion at the $7 price, about 2 conversions for $16 ($7 + $9), and about 3 conversions for $27 ($7 + $9 + $11).

It’s pretty clear that you can spend $7 and still net $3 in profit.  You can spend $9 to get the next conversion, and still net $1 in profit off of that transaction.  It’s worth it!  But it doesn’t make sense to spend $11 to get that final transaction. You’d only have a $10 gross profit on that sale, and after the advertising expense of $11, you’d lose $1.

What decision would Target ROAS make in this scenario?

"AdWords will set maximum cost-per-click (max. CPC) bids to maximize your conversion value, while trying to achieve an average return on ad spend (ROAS) equal to your target." (Source:

First off, you have to specify a ROAS setting.  Let’s say you set it to be 200% and you expect to get $1 in revenue for every $1 in ad spend (remember, a ROAS of 200% is actually just a 100% return).   You set it at this because 50% of your revenue is your profit margin on the product, so that’s the maximum you would want to spend.  But then ROAS would attempt to set your bids so it achieves an average ROAS equal to your target.

Google would happily spend $7 for that first conversion, $9 for the second conversion, and $11 for the third conversion, for a total of $27, getting that $60 in revenue ($30 in margin).  And it would want to spend even more trying to get a fourth conversion, since it hasn’t yet achieved the target ROAS. So clearly you can’t set your Target ROAS based on your gross margins. You would have to adjust it down somewhat.  But to what?  

The thing is, you cannot possibly know because the right level would depend on the rate that each incremental conversion would cost you.  And you cannot know what that is either without doing some mathematical modeling.  Even more, that line will vary by campaign, ad group, even by individual keywords or products in shopping campaigns.  

Any target ROAS shared bid strategy will result in bids that are too high for some keywords and too low for others. Target ROAS bidding actually seems designed to lure ecommerce merchants into overspending on their advertising. It is very easy to lose money with this bidding system.

Decisions Based on Averages

This is really the same issue that comes up with having Cost-Per-Acquisition goals and setting Target CPA bidding (see: Should You Target A Certain CPA in AdWords?).  You’re getting diminishing marginal returns that vary across your account at various rates.  You can’t set rational bids based on averages when you have declining marginal returns.

Target ROAS could be appropriate in some limited cases. It could get you in the right ballpark as a quick and dirty of moving towards optimum profitability.  But only if you really put the work into getting your math correct.  And even then, you’ll still lose out to competitors who put more work into setting more precise bids. Target ROAS bidding cannot get you all the way to optimal profit. It is more likely to result in sloppy analysis that will lose you money.

Next Post Previous Post