Google Data

Evelyn recently provided an introduction to the AdWords Search Funnels reports and I would like to expand on her post by looking at these new reports in more depth. I want to provide some insight into how the Search Funnels reports can help you to re-evaluate your keywords

What can I learn about my advertising from the Search Funnels reports?
A lot There’s a lot of information in the reports and you might wonder where to start. You can see information as diverse as how long it took from users first seeing an ad to converting to which keywords had the best ratio of providing assists to last click conversions. There’s a lot of insights to be gleaned. In this post I’ll focus on re-thinking how you attribute credit to your campaigns, ads and keywords.

Some thoughts on attribution
One of the great things about online advertising is it’s accountability. Using tools like Conversion Tracking and Google Analytics we can track whether visitors to our site go on to convert after clicking one of our ads. Traditionally if someone clicked our ad and then converted we recorded a conversion against that ad, ignoring any interaction that user had with our site or ads previously. This has been the de-facto way to evaluate our online advertising performance for quite a while now, but this sort of measurement doesn’t always tell the full story. If I was to buy a new DSLR camera for example I would be unlikely to do one search and buy immediately. The same is true for most of our purchases online and offline, typically there will be a decision process that will involve some research. I might start out searching on general reviews for DSLRs, then reviews for specific models and finally move on to searches for one specific model that I have decided to buy (I’m almost at this stage ).

Studies have shown that users do anything between 5 to 6 searches and 9 to 12 site visits prior to converting for products like mobile phones, computers and clothing.*

The truth is, looking solely at conversion rates based on last click can lead to some flawed decision making. I previously wrote a post where I recommended looking at additional metrics like bounce rate in Google Analytics when evaluating keywords. A smart reader commented that showing assists for keywords further back in the purchase cycle and granting them a share of revenue would be the ideal way to deal with this scenario.

The obvious question then is; how much value or share of revenue do we attribute to our assisting keywords? There are different approaches you could take to answer this question. You might just decide to divide the credit equally between all keywords clicked on prior to conversion, you might value clicks that happened closer to conversion higher reasoning that there’s a ‘time-decay’ effect, you might give special credit to the first ‘introducer’ keyword and last ‘closer’ keyword or you may use some other model of your own making. I think it’s fair to say that it’s difficult to single out a clear one size fits all model. Also a ‘complete’ attribution solution would ideally focus on a variety of mediums/channels including offline so as I said I don’t think that perfect solution exists yet. However attempting to do multi-touch attribution is a vast improvement on evaluating keywords solely on how they perform in terms of last click.

The value of Search Funnels

The data that Search Funnels provides enables us to do multi-click attribution for our AdWords campaigns. It shows us which keywords were viewed and clicked on on Google search domains prior to conversion. It can also show us the sequence of those impressions/clicks.

One metric I really like in the search funnel reports is the ratio of assist clicks to last clicks as found in the ‘Assist Clicks & Impressions’ report. The reason I like this metric is because it removes the bias of volumes. If you have a thousand keywords and you only look at the first 10 rows in your report you sometimes forget about the other 990. In the remaining 990 lie keywords that you might have de-prioritised because of their low count of last click conversions but they might be creating awareness and help in setting up that final sale. The assist/last ratio highlights these keywords.

An Example: Twiddy.com

Let’s take a look at an example.

Twiddy.com rents holiday homes on the east coast of the US. If you take a look at the site I think you’ll agree that the places they rent out are quite nice! They’re generally big and lavish and not likely to be the sort of place you’ll rent without some consideration.

Diving in to their Search Funnel reports we order the Assist Clicks & Impressions report by the assist/last ratio. Here is the list of keywords we get:

The first few keywords on the list are very low volume but row 4 grabs our attention. It is outperforming the average ratio by quite a bit, 1.25 vs .84, and has enough volume to warrant a bit more exploration.

We now check out the Top Paths report:


We see the [outer banks vacation rentals] keyword appear in quite a lot of paths. The conversion rate had never looked particularly good for this keyword so we didn’t invest too heavily in it. But this keyword seems to be a lot more important than what we thought when we looked at last click. Its real value seems to be as a keyword that drives initial awareness of the site.

If we don’t have much time we could try and get this keyword more traffic by upping our bid. We could check in in a week or so to see the impact on the paths above. If it continues to provide the same level of assists and last clicks relative to ad clicks then we could consider retaining the increased bid.

If have some more time we should consider trying to do some more precise attribution. We could combine Search Funnels data with AdWords cost data to achieve this. Our goal here will be to redistribute conversion credit amongst our keyword portfolio to account for the benefits of assists. We will see keywords that are better at last click conversions get less credit and hence higher CPAs (cost per acquisition) and keywords that contribute more in terms of assists get more credit and lower CPAs. In the above example it’s likely we would end up with a lower CPA for [outer banks vacation rentals] at the expense of a higher CPA for the Twiddy brand keywords.

If we decided that distributing credit equally was the form of attribution that made most sense for our company we could simply divide the credit from conversions for paths in rows 3 & 4 in the top paths screenshot by two (as there are two keywords in these paths). To do this in bulk it will be easier to export all the data to a spreadsheet and apply some logic there to calculate the number of keywords in the path, divide the conversion value by that figure and then credit the keywords. We could use the same method for the other models I mentioned earlier, building in weights (percentage increase or decrease) if we deem some positions in the path to be of higher or lower value.

Nothing to tell and the value of testing

I talked about the bias of volumes earlier in the post. Lower volume, high value keywords can be discovered using the ratio of assists to last but this pre-supposes that we are bidding on keywords nearer the start of the buying cycle in the first place. If we have taken our last click obsession to the extreme we may have stopped bidding on more generic terms altogether.

When this happens Search Funnels can’t really tell us much and you should consider experimenting with some more general keywords. If adding generic keywords in to your account I would recommend doing so on a test basis. See this post for ideas on expanding your keyword list. Once they have enough volume (at least well into the double figures, ideally into treble) you can evaluate the number of last clicks, assists and where in the conversion path they fit. If you then find they provide real value, you can decide to retain these keywords.

Posted by Brian O’ Sullivan, Google Analytics Team.

*Source: comScore custom analysis- UK Population, 15+