As PPC advertisers, you need to constantly monitor and measure your marketing initiatives. The end goal being: determining the best way to spend your PPC budget while still understanding where your customers come from. With so many interactions leading to a purchase, attribution gives a bird’s-eye view of how different channels are performing and what gives the final conversion. While attribution surely helps you take important decisions regarding your business goals, often choosing the right model might be trickier than you’d imagine.
So in this episode on PPC Town Hall, we invited over some of the top data and attribution experts in the industry to share their tips and tactics on leveraging attribution with the help of Google Analytics.
Ken Williams, Senior Data Engineer, Search Discovery
Brooke Osmundson, Director of Paid Media, NordicClick Interactive
Christopher Gutknecht, Teamlead Acquisition & Optimization, Bergzeit
As always, you can view this week’s episode as well as previous editions of PPC Town Hall right here.
Here are 5 insights to attribution and Google Analytics.
1. The biggest measurement challenges in PPC
Chris: We’ve been working on profit bidding and attribution for a year now, and they’re both very challenging. The more you dig down into the rabbit hole, the more challenges you find. You’ve to make so many assumptions to move forward. So far, we’ve tested two models, one of them, the Markov chain model, is already in production. We use the Markov chain model, which we calculate on a daily basis and add the fresh results to attribute channel waste.
Brooke: If measuring different attribution models is new to you, start researching models to understand how your marketing efforts would benefit from each of the available models. If you have them built in Google Ads and Google Analytics, start comparing them against each other to understand the different touchpoints on what will work best for your business. These are the conversations worth having with our clients.
I feel like we’re a little bit behind ourselves in terms of adopting newer attribution models, especially with user behavior shifting. We’re seeing users take longer and there are other touchpoints to make a decision. If clients are still using last-click attribution, marketing efforts won’t show the full picture and can be detrimental in decision-making. Trying to find a model that doesn’t kill your upper funnel is extremely important.
2. Go for the ‘best choice’
Ken: We are working with pretty imperfect data. Being comfortable with acknowledging that, and thinking through the implications of different models to find the best fit is a business challenge. We have to start with the questions that the business needs to answer. And there’s no perfect fit. There’s going to be a ‘best choice’ for your customer experience and you just have to think through what your options are. Experimentation is really important!
3. What’s new with Google Analytics 4
Ken: There’s a lot that’s new in GA4. One of the most fundamental is the concept of an ‘event-driven’ data model, a structure that Firebase Analytics has used for years. It works well with mobile-apps, and now web and mobile will be sharing the same structure. Moreover, the way that we measured engagement has changed quite a bit. All of the key engagement metrics that we relied on with legacy versions of GA have been replaced by a new feature called ‘engagement time’ was rolled out with the ‘event-driven’ data model, which solves the problems with session-based engagement metrics. Finally, GA4 is built on the global site tag, which can help you make changes to the user interface that actually modify the code on your site without requiring a change in the tag manager.
4. Challenges with the tracking cookies fading out
Chris: The fading-out idea is hard to predict and focuses more on cross-site tracking than the first-party context. I think we should be working on small steps from the things that are already hurting us. So for instance, Safari ITP 2.1 kills first-party cookie data after 7 days, but they respect server-side cookies. One thing you could improve right now is to transfer to server-side cookies to preserve GA cookies on Safari devices.
Ken: There are two privacy-related things that we need to respond to: one of them is regulatory (we’ve been dealing with GDPR for a while and CCPA is new in the US), and the second is the momentum from popular browsers to restrict cookies. The vision that both Apple and Mozilla have put out is a desire to restrict the ability to monitor your cross-domain activities, by companies like Facebook.
One thing that you can do right now is to write your cookies from the server rather than with javascript. This is not a permanent solution, and it’s not easy for many companies because it requires the assistance of skilled developers.
5. Filling in data-gaps for attribution optimization
Brooke: I think that user-stitching or device ID attribution will be a more long-term solution. But right now, it’s extremely complicated. Whether you have the infrastructure in your team or at a large agency, but you do have essential people in roles that drive developmental work. Since you know what kind of data is available, the best thing will be to have these conversations with your clients. You should also do a bit of testing to figure out what works best for your organizational goals. We know that there will always be gaps in data, so until clients are able to invest in holistic solutions such as device ID attribution, you may have to make assumptions based on data trends available to you. It’s getting more difficult to understand everything there is about every customer, so focus on identifying what can move the needle for your company.
Ken: 10 years ago I believed very strongly in this 360° view of a customer. The idea was that since we’ve got all this digital data, we are gonna get better and better at understanding why customers behave the way they do. And we’ll eventually get to a point where we’ve got amazing data where we know everything. I feel like that is something we need to give up on as an industry because it was never a realistic aspiration. We’re never gonna fill all these data-gaps and have to get used to that.
Conclusion
Let’s be honest. There’s no ‘ideal’ answer to many of these questions regarding attribution. Even to pick the perfect model for your business, you have to constantly evaluate your marketing initiatives. And as Google continues to take away our data, we just have to become a little bit broader in our thinking and go back to the initial question - what’s the business trying to achieve?
This is where experimenting and doing multiple field tests with different models can pay off. The other thing that we can do is look for sophisticated attribution modeling to get that ‘best choice’ for the customer experience. Discussing this with your in-house/agency development teams can only help you finetune your buyer funnel better.
Machine Learning (ML), Artificial Intelligence (AI) and Automation are three trending topics in the industry today. It’s an accepted fact that automation is here to stay so it’s our job to learn how to make the most of it for our PPC accounts. In my book “Digital Marketing in an AI World”, I explain that one of the roles humans will have to play when their old job has been automated is that of the “PPC Doctor”: someone who knows the right medicine for their patient and who also understands potentially dangerous interactions. This post covers one such interaction that can lead to disastrous results in PPC.
We’re talking here about Google Ads’ smart bidding strategies. Even though they’re designed to help advertisers reach a determined goal, they lack the human intuition for understanding how to deal with gray areas, and are prone to bad decisions when they’re fed bad data. Specifically, they can do major damage to accounts that are using last-click attribution (LCA) models.
Understanding Last-Click Attribution Model
Last-Click is one of the 6 different attribution models offered by Google Ads. It gives all the credit to the ad and keyword which was last clicked before a conversion.
For example, let’s say you are advertising athletic shoes. There’s a sequence of queries done by a user that goes something like this: “Sneakers” > “Running Shoes” > “Adidas Running Shoes” and finally they search for “Ultraboost 19”. This is just a simple example to illustrate that users tend to start with broad queries and get more specific as they get to understand what it is they might want to buy.
If your campaign is using the Last-Click attribution (LCA) model, then all the credit for the conversion will be given to the ad shown for the final query: “Ultraboost 19”, and no credit will be given to any of the queries that preceded it.
Conversion Funnels and LCA
So why is this so bad? When you give all the credit to the last-clicked ad/keyword, it’s like saying you don’t think there was any value to all the queries along the way that helped the user become aware and familiar with your offering. You’re assuming the user would have discovered to search for “Ultraboost 19” without having been exposed to any of your other ads. This is generally a false assumption, especially for consumers who are not very familiar with your brand and its latest offerings.
Consumers today have more interactions than ever before with brands while researching what to buy. Brands that are not present at the earlier stages of a user’s discovery process may not be in contention to win their business later down the line.
So using last-click attribution would mean that “Sneakers”, “Running Shoes” or “Adidas Running Shoes” are assigned no value.
Attribution Models Inform Optimizations
Why is it so important to assign the correct value? Doesn’t the attribution model just change the numbers in reports? The answer is ‘no,’ the attribution model populates the conversion and conversion value metrics and most account managers rely on these to decide where to allocate their budgets, where to change bids, what queries to add as keywords, and what negative keywords to add.
This could all be okay if a human was managing all this manually. For example, while the lack of conversions for a keyword like ‘sneakers’ might normally be grounds for a bid reduction, an account manager would likely realize that they’d still want to bid for this keyword. Human judgment would win out over purely following some logical rules and the account might do fine.
But like I said before, automation is increasingly doing more of the day-to-day account management and it lacks the human judgment that averted disaster in this scenario of an advertiser using last-click attribution.
Smart Bidding + Last-Click Attribution
When last-click attribution is being used, the keywords “Sneakers”, “Running Shoes” or “Adidas Running Shoes” from the example above, will be reported as non-converting, although they are still valuable keywords because they help consumers unfamiliar with your brand discover your brand’s offerings as they do their research.
Now here’s where results can get really bad… by combining bid automation with last-click attribution. The job of automated bidding, like target CPA (tCPA) or target ROAS (tROAS) bidding from Google, is to calculate the appropriate CPC that is needed for the ad to enter the auction.
The ‘right’ CPC is determined one of two ways:
For tCPA, Google uses the predicted conversion rate to calculate CPC
For tROAS, Google uses the predicted conversion value for a click to set the CPC
But if the attribution model hasn’t been assigning conversions to upper-funnel searches, it will predict that conversion rate will be low and that the value per click will be low. So now the automated bidding system will start to reduce bids for these upper funnel keywords. And eventually bids will get so low that the ads may stop showing altogether.
This is bad because it means you’re reducing the volume of prospects who will be exposed to your brand at earlier stages. Eventually your funnel just dries up and the only sales you’re left with are those from people who already knew your brand and products very well — the people who knew to search for “Ultraboost 19”.
Final Thoughts
Considering the significant risk of making bad decisions for the reasons explained above, we advise all our customers to switch away from using Last-Click attribution. If anything, simply switch to a time-decay model which is most similar to last-click while still giving some value to all stages of the funnel.
When it comes to automations like smart bidding strategies, or automated bids using another platform, knowing how they interact with your measurement systems is an absolute must if you want to avert an account blowup.
As PPC advertisers, you need to constantly monitor and measure your marketing initiatives. The end goal being: determining the best way to spend your PPC budget while still understanding where your customers come from. With so many interactions leading to a purchase, attribution gives a bird’s-eye view of how different channels are performing and what gives the final conversion. While attribution surely helps you take important decisions regarding your business goals, often choosing the right model might be trickier than you’d imagine.
So in this episode on PPC Town Hall, we invited over some of the top data and attribution experts in the industry to share their tips and tactics on leveraging attribution with the help of Google Analytics.
Ken Williams, Senior Data Engineer, Search Discovery
Brooke Osmundson, Director of Paid Media, NordicClick Interactive
Christopher Gutknecht, Teamlead Acquisition & Optimization, Bergzeit
As always, you can view this week’s episode as well as previous editions of PPC Town Hall right here.
Here are 5 insights to attribution and Google Analytics.
1. The biggest measurement challenges in PPC
Chris: We’ve been working on profit bidding and attribution for a year now, and they’re both very challenging. The more you dig down into the rabbit hole, the more challenges you find. You’ve to make so many assumptions to move forward. So far, we’ve tested two models, one of them, the Markov chain model, is already in production. We use the Markov chain model, which we calculate on a daily basis and add the fresh results to attribute channel waste.
Brooke: If measuring different attribution models is new to you, start researching models to understand how your marketing efforts would benefit from each of the available models. If you have them built in Google Ads and Google Analytics, start comparing them against each other to understand the different touchpoints on what will work best for your business. These are the conversations worth having with our clients.
I feel like we’re a little bit behind ourselves in terms of adopting newer attribution models, especially with user behavior shifting. We’re seeing users take longer and there are other touchpoints to make a decision. If clients are still using last-click attribution, marketing efforts won’t show the full picture and can be detrimental in decision-making. Trying to find a model that doesn’t kill your upper funnel is extremely important.
2. Go for the ‘best choice’
Ken: We are working with pretty imperfect data. Being comfortable with acknowledging that, and thinking through the implications of different models to find the best fit is a business challenge. We have to start with the questions that the business needs to answer. And there’s no perfect fit. There’s going to be a ‘best choice’ for your customer experience and you just have to think through what your options are. Experimentation is really important!
3. What’s new with Google Analytics 4
Ken: There’s a lot that’s new in GA4. One of the most fundamental is the concept of an ‘event-driven’ data model, a structure that Firebase Analytics has used for years. It works well with mobile-apps, and now web and mobile will be sharing the same structure. Moreover, the way that we measured engagement has changed quite a bit. All of the key engagement metrics that we relied on with legacy versions of GA have been replaced by a new feature called ‘engagement time’ was rolled out with the ‘event-driven’ data model, which solves the problems with session-based engagement metrics. Finally, GA4 is built on the global site tag, which can help you make changes to the user interface that actually modify the code on your site without requiring a change in the tag manager.
4. Challenges with the tracking cookies fading out
Chris: The fading-out idea is hard to predict and focuses more on cross-site tracking than the first-party context. I think we should be working on small steps from the things that are already hurting us. So for instance, Safari ITP 2.1 kills first-party cookie data after 7 days, but they respect server-side cookies. One thing you could improve right now is to transfer to server-side cookies to preserve GA cookies on Safari devices.
Ken: There are two privacy-related things that we need to respond to: one of them is regulatory (we’ve been dealing with GDPR for a while and CCPA is new in the US), and the second is the momentum from popular browsers to restrict cookies. The vision that both Apple and Mozilla have put out is a desire to restrict the ability to monitor your cross-domain activities, by companies like Facebook.
One thing that you can do right now is to write your cookies from the server rather than with javascript. This is not a permanent solution, and it’s not easy for many companies because it requires the assistance of skilled developers.
5. Filling in data-gaps for attribution optimization
Brooke: I think that user-stitching or device ID attribution will be a more long-term solution. But right now, it’s extremely complicated. Whether you have the infrastructure in your team or at a large agency, but you do have essential people in roles that drive developmental work. Since you know what kind of data is available, the best thing will be to have these conversations with your clients. You should also do a bit of testing to figure out what works best for your organizational goals. We know that there will always be gaps in data, so until clients are able to invest in holistic solutions such as device ID attribution, you may have to make assumptions based on data trends available to you. It’s getting more difficult to understand everything there is about every customer, so focus on identifying what can move the needle for your company.
Ken: 10 years ago I believed very strongly in this 360° view of a customer. The idea was that since we’ve got all this digital data, we are gonna get better and better at understanding why customers behave the way they do. And we’ll eventually get to a point where we’ve got amazing data where we know everything. I feel like that is something we need to give up on as an industry because it was never a realistic aspiration. We’re never gonna fill all these data-gaps and have to get used to that.
Conclusion
Let’s be honest. There’s no ‘ideal’ answer to many of these questions regarding attribution. Even to pick the perfect model for your business, you have to constantly evaluate your marketing initiatives. And as Google continues to take away our data, we just have to become a little bit broader in our thinking and go back to the initial question - what’s the business trying to achieve?
This is where experimenting and doing multiple field tests with different models can pay off. The other thing that we can do is look for sophisticated attribution modeling to get that ‘best choice’ for the customer experience. Discussing this with your in-house/agency development teams can only help you finetune your buyer funnel better.
Machine Learning (ML), Artificial Intelligence (AI) and Automation are three trending topics in the industry today. It’s an accepted fact that automation is here to stay so it’s our job to learn how to make the most of it for our PPC accounts. In my book “Digital Marketing in an AI World”, I explain that one of the roles humans will have to play when their old job has been automated is that of the “PPC Doctor”: someone who knows the right medicine for their patient and who also understands potentially dangerous interactions. This post covers one such interaction that can lead to disastrous results in PPC.
We’re talking here about Google Ads’ smart bidding strategies. Even though they’re designed to help advertisers reach a determined goal, they lack the human intuition for understanding how to deal with gray areas, and are prone to bad decisions when they’re fed bad data. Specifically, they can do major damage to accounts that are using last-click attribution (LCA) models.
Understanding Last-Click Attribution Model
Last-Click is one of the 6 different attribution models offered by Google Ads. It gives all the credit to the ad and keyword which was last clicked before a conversion.
For example, let’s say you are advertising athletic shoes. There’s a sequence of queries done by a user that goes something like this: “Sneakers” > “Running Shoes” > “Adidas Running Shoes” and finally they search for “Ultraboost 19”. This is just a simple example to illustrate that users tend to start with broad queries and get more specific as they get to understand what it is they might want to buy.
If your campaign is using the Last-Click attribution (LCA) model, then all the credit for the conversion will be given to the ad shown for the final query: “Ultraboost 19”, and no credit will be given to any of the queries that preceded it.
Conversion Funnels and LCA
So why is this so bad? When you give all the credit to the last-clicked ad/keyword, it’s like saying you don’t think there was any value to all the queries along the way that helped the user become aware and familiar with your offering. You’re assuming the user would have discovered to search for “Ultraboost 19” without having been exposed to any of your other ads. This is generally a false assumption, especially for consumers who are not very familiar with your brand and its latest offerings.
Consumers today have more interactions than ever before with brands while researching what to buy. Brands that are not present at the earlier stages of a user’s discovery process may not be in contention to win their business later down the line.
So using last-click attribution would mean that “Sneakers”, “Running Shoes” or “Adidas Running Shoes” are assigned no value.
Attribution Models Inform Optimizations
Why is it so important to assign the correct value? Doesn’t the attribution model just change the numbers in reports? The answer is ‘no,’ the attribution model populates the conversion and conversion value metrics and most account managers rely on these to decide where to allocate their budgets, where to change bids, what queries to add as keywords, and what negative keywords to add.
This could all be okay if a human was managing all this manually. For example, while the lack of conversions for a keyword like ‘sneakers’ might normally be grounds for a bid reduction, an account manager would likely realize that they’d still want to bid for this keyword. Human judgment would win out over purely following some logical rules and the account might do fine.
But like I said before, automation is increasingly doing more of the day-to-day account management and it lacks the human judgment that averted disaster in this scenario of an advertiser using last-click attribution.
Smart Bidding + Last-Click Attribution
When last-click attribution is being used, the keywords “Sneakers”, “Running Shoes” or “Adidas Running Shoes” from the example above, will be reported as non-converting, although they are still valuable keywords because they help consumers unfamiliar with your brand discover your brand’s offerings as they do their research.
Now here’s where results can get really bad… by combining bid automation with last-click attribution. The job of automated bidding, like target CPA (tCPA) or target ROAS (tROAS) bidding from Google, is to calculate the appropriate CPC that is needed for the ad to enter the auction.
The ‘right’ CPC is determined one of two ways:
For tCPA, Google uses the predicted conversion rate to calculate CPC
For tROAS, Google uses the predicted conversion value for a click to set the CPC
But if the attribution model hasn’t been assigning conversions to upper-funnel searches, it will predict that conversion rate will be low and that the value per click will be low. So now the automated bidding system will start to reduce bids for these upper funnel keywords. And eventually bids will get so low that the ads may stop showing altogether.
This is bad because it means you’re reducing the volume of prospects who will be exposed to your brand at earlier stages. Eventually your funnel just dries up and the only sales you’re left with are those from people who already knew your brand and products very well — the people who knew to search for “Ultraboost 19”.
Final Thoughts
Considering the significant risk of making bad decisions for the reasons explained above, we advise all our customers to switch away from using Last-Click attribution. If anything, simply switch to a time-decay model which is most similar to last-click while still giving some value to all stages of the funnel.
When it comes to automations like smart bidding strategies, or automated bids using another platform, knowing how they interact with your measurement systems is an absolute must if you want to avert an account blowup.
