Adwords Conversion Funnels & Attribution Models: Can They Work Together?

First click. Last click. Linear. Reverse decay. Linear reverse decay…everyone accepts that attribution is a necessary consideration in analytics, but no one seems to be able to crack the code. There are countless ways to attribute revenue, yet the major analytics systems stick with the traditional last entry.

The reality is that we, they, you, I, don’t know the right answer. One thing is for sure: we should pay attention to the funnel of clicks leading to an event because there are generic, upper funnel terms driving clicks on more product specific and branded terms which lead to sales.

But how many clicks are relevant and what key performance indicators should we use to keep the account in check?

To answer the first part of this question, we compiled a data set from 40 etailers across multiple verticals.

To keep things consistent, we used an industry standard cookie window of 30 days.

The result: 93% of purchases occur within 3 clicks when respecting the 30 day window. That means for any given purchase, there’s a 93% chance that that the conversion occurs in 3 clicks or less.

Typical Paid Search Conversion Time Lag

Typical Conversion Time Lag

The average number of clicks before a conversion was 1.8.

While not quite strong enough to write a scientific theory, this does give some convincing directional evidence for how we can distribute conversions across multiple converting terms.

The Trouble With Calculating KPIs

Most online retailers sell more than one product, each of which has a different cost of production and margin. As a result, return on ad spend (ROAS) is a more reliable performance indicator than cost per acquisition (CPA; cpa = cost / orders) for standard paid search bid optimization.

If we try and attribute revenue across multiple terms leading to a conversion, we’ll end up with more keywords associated with revenue, but not meeting goals.

Imagine the case where you have 2 keywords in an account with a ROAS goal of 3. One keyword gets all the last click conversions and one keyword is always in the upper funnel. Both generate the same number of clicks at the same CPC, so spend is identical (quite a perfect world, I know).

Aggregate ROAS is 4 and goal is 3. So the last click term is looking great while the upper funnel term looks terrible.

In a linear attribution model (equal distribution of revenue for all terms in the click stream), both terms would have a ROAS of 2 and bid logic would tell you to bid them both down. Boo.

This is why it’s crucial to look at a click stream holistically – considering all costs of upper and lower funnel terms and total revenue in order to understand the true performance of your account. For this reason, I am strongly against using revenue in attribution and believe that attribution models should only consider conversions and thus the most logical key performance indicator is CPA.

Analytics Without Attribution?

So how should you go about optimizing your account, considering the upper funnel, but not using an attribution model to dilute data and screw everything up?

Assuming that you or your company do not have access to advanced conversion path reporting offered by a quality 3rd party SEM technology provider and are limited to what’s available for free in AdWords conversion funnels, here are a couple of relatively simple ideas which can be used to extrapolate the value of upper funnel clicks leading to a conversion on a lower funnel term, commonly referred to as an ‘assist conversion’.

  • Start with a keyword level report within the search funnels section of AdWords and a Keyword report directly from the AdWords campaign user interface.
  • Identify all terms not meeting your CPA or ROAS goal (whichever is your key performance indicator of choice).
  • Isolate these terms and filter out terms without assist conversions. Use the formula below to determine if these upper funnel terms are providing value to your account:

CPA = Cost / (Conversions + (Assist Conversions / avg clicks per conversion))

This calculation uses the average number of clicks per conversion (1.8 from my analysis…your account data is available in search funnels) to calculate a ‘more true’ cost per acquisition for upper funnel terms.

By using CPA instead of ROAS, you are not directly associating revenue but are taking into account that this term is partially responsible for driving conversions. I divide the number of assist conversions by the average clicks-to-conversions in order to de-duplicate assist conversions within each click funnel.

A more extreme measure is to determine a true multiplier for the impact of upper funnel terms on lower funnel terms.

Follow these steps to create a clean environment for testing the latency effect of upper funnel terms on lower funnel terms:

  • Turn off upper funnel terms for 1 cookie cycle (30 days in my test – this is the default cookie length in AdWords). Upper funnel terms can be defined as any terms without or with minimal last click conversions and observed assist conversions.
  • Measure the negative effect on branded/product terms in the following 30 day cycle. If there is a negative impact on brand and product level terms that is greater than the conversions associated via last click to those upper funnel terms alone [during the previous reporting period]m then there is a latency effect for upper funnel terms on lower funnel terms.
  • Calculate the multiplier and this becomes your attribution multiplier for upper funnel terms.

Note: Seasonality and other marketing campaigns can come in to play and skew the numbers in the testing environment. An ideal scenario would be to test at a time when there are no promotions running and there is no significant seasonal impact.

Before you turn off the upper funnel terms in your account, please consider this caveat: this will have a negative impact on overall revenue/conversions despite improving ROAS or CPA in the short term…which is probably why very few advertisers do it.

This is the cleanest way to determine a real attribution model.

In certain scenarios, pressure from management or a client may put you in a situation where you need to conduct such a test. Only resort to this test if you must prove the value of upper funnel terms in your account.

Opinions expressed in the article are those of the guest author and not necessarily Search Engine Land.

Related Topics: Channel: Analytics | Search & Analytics


About The Author: is the Vice President of Performance Marketing and Analytics at SellPoints and is based in the San Francisco Bay Area.

Connect with the author via: Email | Twitter | LinkedIn


Get all the top search stories emailed daily!  


Other ways to share:

Read before commenting! We welcome constructive comments and allow any that meet our common sense criteria. This means being respectful and polite to others. It means providing helpful information that contributes to a story or discussion. It means leaving links only that substantially add further to a discussion. Comments using foul language, being disrespectful to others or otherwise violating what we believe are common sense standards of discussion will be deleted. Comments may also be removed if they are posted from anonymous accounts. You can read more about our comments policy here.
  • George Michie

    Wow, where to begin. Let’s start with the choice of conversions rather than revenue. If in aggregate the ROAS is 4, then the average ROAS of both keywords is…4. Using your example, from last click analysis Keyword X has $100 in sales on $10 in cost, keyword Y ‘upstream’ has $0 in sales and $15 in cost. In aggregate a 4/1 ROAS. If you spread the credit evenly, you get Keyword X has $50 in rev and $10 in cost; keyword Y has $50 in rev and $15 in cost…averaging out to….an ROAS of 4.

    Average touches to conversion = 1.8. If you look at the data and take out the instances of multiple clicks on THE SAME AD, you’ll see that number drop significantly, as that accounts for the majority of multiple click orders.

    I’ll leave the rest for someone else.

  • Siddharth Shah

    Like George, I dont know where to start. Several statements are made without data backing them up. Upper funnel vs lower funnel words for instance. Is there any proof that upper funnel generics are assisters ? My analysis with real data sets has not revealed such a pattern. Keywords like “shoes” which are considered generic and upper funnel attracts consumers both earlier and later down the funnel. As a result , you will on average not see much of an assisting effect. (The story is different when looking at multi channel data).

    Since Search is lower down the funnel with the result (A) a good fraction (over 60% at least) of conversions are single click and even among the multi click transactions you will see a 50-80% self assist effect.

    Like George,I dont agree that you look at conversions for attribution and revenue for measuring performance. Doesn’t make any sense mathematically.

    I have several other problems with the assertions made but I think the readers get my overall point; Before making an assertion please back it up with data. This is Analyze This after all !


Get Our News, Everywhere!

Daily Email:

Follow Search Engine Land on Twitter @sengineland Like Search Engine Land on Facebook Follow Search Engine Land on Google+ Get the Search Engine Land Feed Connect with Search Engine Land on LinkedIn Check out our Tumblr! See us on Pinterest


Click to watch SMX conference video

Join us at one of our SMX or MarTech events:

United States


Australia & China

Learn more about: SMX | MarTech

Free Daily Search News Recap!

SearchCap is a once-per-day newsletter update - sign up below and get the news delivered to you!



Search Engine Land Periodic Table of SEO Success Factors

Get Your Copy
Read The Full SEO Guide