Keywords

Never Ever Use These Match Types Together

To the tune of Taylor Swift’s song.

It has been some time that I wanted to sit down and write this post. I cringe every time I see incorrect use match type in a client’s account.

As of today you can choose from four type of match types:

  • Exact
  • Broad Modified Match (BMM)
  • Generic broad
  • Phrase

Care to guess which ones are ok and which are not?

Explaining this concept to my clients, I always like to do a small exercise of the imagination. It goes somewhat like this: imagine a car. This car is similar to any other car but the difference is each time you press on the gas pedal, the windshield wipers begin working. Every time you hit on the breaks the right doors open. Each time you signal left or right, your car seat reclines. Now imagine trying to drive this imaginary car on your daily commute from home to work. Yes, it will drive you crazy!

Managing and trying to optimize an account with most of the keywords are generic broad or phrase match is like driving that car.

So now you know that Exact and BMM match types are ok, the other two, not so much.

So the begging questions is why?

There are two major concepts on selecting keyword match type, coverage and linear independence.

Coverage is the ability of having a small set of keywords that cover an important amount of user search queries of the interested category you compete in. Coverage is lowest with Exact match as the probability to match a certain user search query is lower than other types of matches. In the other extreme, the Generic Broad has the highest coverage. However, this high coverage comes at a price. A very steep price indeed as you will see below, the price is low independence.

Linear independence is the ability for one keyword of behaving independently of other keywords in the same account. We can think of it as low interdependence or low cannibalization of searches between each keyword. For those of you that remember some algebra, it is similar to the concept of avoiding the co-linearity of vectors. Now, keywords and semantic objects are not vectors. However, organizing an account as clean as possible is the first step to been able to optimize the account in the future through keyword bidding.

Why then do broad generic keywords have low independence? And why should it matter? There two main problems with broad generic keywords:

1) Not all the keywords needed to be present in the query to activate the match. What we call the partial match.

2) Google allows the use of synonyms on broad generic.

This generates many search queries that may not be aligned to your business offerings or there is a lost opportunity of handling a generic keyword as a mixed bag of intentions. Let’s take one example from the auto industry. One might say that the adjective “used” car and “pre-owned” car means the same and Google might interpret them as synonyms. However, I can tell you from experience that keywords +used +car and +pre +owned +car do not behave the same. For a used car classifieds site, generally, the “pre-owned” term usually has a higher conversion rate. Because of this conversion rate difference, it is important to have these terms separated using BMM structure instead of a generic match type. This will allow to bid separately for each term optimizing the account better.

So remember, every time you use a generic broad match a SEM fairy dies. Joking aside, I recommend using a mix of 5%/10% exact match keywords and 90%/95% BMM keywords.

Think about building a good structure and keyword base for account optimization through keyword bidding.

Happy bidding.

lukas-blazek-mcSDtbWXUZU-unsplash (1)

Marginal CPA vs Average CPA: What most people get wrong.

I still get quite surprised about how people within the industry still confuse the two terms: average vs marginal cost per acquisition (CPA).

I remember once I was renewing my drivers license at the local DMV office in Buenos Aires. While attending the mandatory class, the person teaching the class asked us this question: What is the speed limit in a city avenue? Everyone answered simultaneously “60 Km/h”, we were quite confident on our response only to be reprimanded by the DMV teacher that our answer was incomplete. He corrected us saying that the correct response was “30 Km/h as a lower limit and 60Km/h as the upper limit”. That response stuck with me regarding how sometimes one metric by itself might give an incomplete picture of a particular situation.

I believe the same happens relating to CPA, a more complete understanding of how your paid channels are performing involves knowing both average CPA AND marginal CPA.

Calculating Marginal CPA

The average CPA is pretty straightforward, it is the total cost divided by the total amount of paid conversions. An easy way to muddy this metric is to not only use paid conversions but use total conversions (including free/organic).

Ok, so how about marginal CPA? The first key insight of marginal CPA is that it is not a constant. Marginal CPA depends on the amount of conversion one takes as a starting point. The marginal cost for the 100th conversion is not the same as the marginal cost of the 500th conversion. If the sorting of the conversions has been done right (less expensive first), what you will see is that marginal costs is always incremental.

The graph below tries to show both concepts(1). If one were to grab and adwords account and detail the information by grouping the data by keyword adding total cost and total conversions for each keyword. Later ordering the keywords from least expensive to most expensive.

The X-axis shows the accumulated amount of conversions while the Y-Axis tracks the accumulated investment.

In the above example activating only a subset of the keywords, the least expensive keywords to achieve 4000 conversions. One would have to activate 887 keywords for a total of 4012 conversions at a total cost of $31,374. Making the average CPA $7.82.

However the marginal CPA at 4000 conversions is calculated differently. Taking two very close data points we get:

Point A: 3991 conversions, total investment $31,090

Point A’: 4012 conversions, total investment $31,374

In this example and at this specific point marginal CPA is close to double of what the average CPA is. Another benefit of this graph setup is that the slope of the curve corresponds to the marginal CPA.

I usually consider marginal CPA to be a more important metric than average CPA. It is a good indicator that shows how scalable are your current SEM efforts. In the case of selling ad inventory do consider that the real BATNA (best alternative to a negotiated agreement) of the buyer is the marginal CPA and not the average.

Building the graph

Extract and adwords report by keyword export it to an excel. There divide cost by conversions for each keyword (CPA), then sort this column in ascending order. In the same file add two new columns: 1) accumulated conversions: the first row should have the number of conversions of first keyword (KW) , the second row should have conversions of second KW plus the accumulated conversions of the previous row. 2) accumulated cost: same concept of the previous column but instead of accumulating conversion, add total cost invested in each KW. After processing these columns, the last row should show total amount of conversions for the account and total cost.

Once you have this data calculated, select the cells of these two columns and generate a scatter plot where the Y-axis is cost and X-axis amount of conversions. Your plot should like something like the graph shown above.

1.The Google Ads data was taken from an account in the auto industry in Argentina, the absolute values were modified for confidential purposes

Google-position

What Google doesn’t want you to know about average position

An oldie but good one. Originally published in 2015.

The average position myth

A common phrase that I have heard about average position is the closer to the top of the page the better. I have even heard that the 1st ad position should be coveted by digital marketers specially on the branding benefits. The proliferation of this myth leaves me baffled.

I wanted to shed some light and some humble evidence to contradict this point. That first position isn’t the best position SERP for a performance oriented account. Insteat a target position between 2 and 3 stands what I would call a performance “sweet-spot”. 

Average CTR vs. Average Position

Above shows a graph of accumulated information on a high volume (clicks & conversions) ad group. We will later see it is representative of other ad groups that show similar conclusions. The graph was generated with information of 12 month window. Each of the ad groups analyzed were composed of highly homogeneous keywords and high percentage of ad group traffic was consolidated in a few exact terms.

The graph was generated aggregating daily reports by average position cluster. Within the analyzed time period competitive structure was stable and no major changes were made to the ad group just bidding CPC price updates. 

The X-axis represents the average position cluster, the Y-axis on the left shows average CTR.

Nothing mind-blowing here, CTR generally decrease with an increase of average position. Now (below) we add an overlay of data points with their average CPC (right Y-axis)

Average CPC vs. Average Position

Average CPC declines as well, the closer the position to 1.0 the more expensive it is. Again, no major news.

From this point on, it gets more interesting, let’s dive deeper on the CTR and CPC curves. 

CTR Linear Fit

In this example a linear regression for the CTR curve offers a good fit (R^2=71%)

CPC Exponential Fit

However, the CPC curve has a better fit with an exponential curve (R^2 = 86%). This type of curve shows that CPC has an accelerated decay or accelerated increase, depending if you move left or right within the graph.. 

Finding the sweet-spot

Considering that the conversion rate is stable within the same ad group and independent of average position, then CPC and Cost Per Acquisition (CPA) go hand in hand. 

We can see an area where CTR is stable and CPC has local minimum, behold, our sweet-spot. Let’s call this area the Efficient Position Area (EPA, environmental pun intended). 

As we can see from the graph, as we move to the left of the area, CTR does not increase much however average CPC (also conversion cost) increases dramatically without any significant gain in volume. On the other side, moving to the right, CTR has a significant drop as we get close and cross the 3.0 average position barrier. This decline in click volume is seen across many ad groups. I can be related to going from top ads to sidebar ads in the SERPs.

Conclusion

Optimizing bids taking into account ad position within efficient area can generate savings that can be used to invest in other keywords/adgroups lowering overall CPA of global account. 

So try making this analytic exercise with your own adwords account, let me know through comments if you find similar findings.