Attribution modeling is one of my favorite topics in web analytics because it is so complicated and challenging. It forces you to use both your left and right brain simultaneously. It is something which requires deep understanding of your client’s business model and how different marketing channels work together to create sales and conversions. So before use any attribution model, please make sure that you have really very deep understanding of your client’s business, his industry and the target market. Because if you don’t then you may end up applying/creating a wrong attribution model and lose money.
Attribution modeling goes beyond Google Analytics and is a very broad topic. This paper talks only about understanding and using attribution modeling in relation to Google analytics. To cater to the needs of all type of audience (esp. the one who are brand new to attribution modeling and multi channel funnel reports) I have explained the very basic concepts too. Please feel free to skip the sections which you think you are already aware of. If you are a senior web analyst, take a jump straight to the meat section.
Before I dive deep there are few topics which I would like to explain upfront to facilitate the understanding of advanced topics later in the paper.
Acquisition channels
Also known as ‘marketing channels’, ‘digital channels’ or ‘channels’ are the sources of traffic to your website. For example: Paid search, Organic Search, Direct, Social Media, Email, Affiliate, Referral etc are all examples of acquisition channels. In multi channel funnel reports, the acquisition channels are commonly referred to as marketing channels or channels.
Conversions
The definition of conversion is different in Multi channel funnel reports. It can be a Goal conversion or an e-commerce transaction.
The total conversions in multi-channel funnel reports is the sum of total number of Goal conversions and total number of e-commerce transactions.
Note: For the remainder of this post, whenever I talk about conversions, I am talking about the conversions in relation to multi channel funnels. If I talk about Goal conversions I will explicitly mention it. So please remember and don’t get confused later on.
Purge your Analytics data
Please be careful what you track as Goal conversion. I have seen analytics accounts where marketers were tracking ‘visit to the home page’ as a conversion or ‘visit to the product category page’ as conversion. Track only those goals which are beneficial to your customers and company. Irrelevant goals can drastically skew your conversion rate and the data in the multi channel funnel reports. Conclusions drawn from erroneous data will result in wrong marketing decisions and your client won’t appreciate that.
Multi Channel Funnel Reports
Through multi channel funnel reports you can determine:
- How marketing channels work together to create conversions.
- How much time elapsed between visitors’ initial interest and his purchase
- What role did prior website referrals, searches and ads played in a conversion.
- How to attribute conversions to a marketing channel.
There are 5 types of multi channel funnel reports available in Google Analytics:
- Overview report– This report contains a ‘muti-channel funnel visualizer’ through which you can visualize how different marketing channels are working together to create conversions.
- Assisted Conversions Report – This report shows how many interactions each marketing channel initiated, assisted and completed. It also shows the value of assisted and last interaction conversions.
- Top Conversion Path Report – This report shows all of the unique conversion paths that lead to conversions. It also shows the number of conversions from each path and value of those conversions.
- Time Lag Report – This report shows how long it took (in days) for visitors to convert. Through this report you can get an insight into the length of your online sales cycle.
- Path Length Report – This report shows how many interactions it took for visitors to convert.
Data discrepancy between multi channel funnel reports and other reports in Google Analytics
- In multi channel funnel reports, a conversion can be a Goal Conversion or e-commerce transaction. Whereas in non-multi channel funnel reports a conversion means Goal Conversion. The e-commerce transactions are mentioned separately. So total number of conversions in multi-channel funnel reports can be different than the total number of conversions in non-multi channel funnel reports.
- Multi channel funnels data collection lags by up to 2 days. So their results are temporary out of sync.
Conversion Paths
It is the sequence of interactions (clicks, visits, impressions) with digital marketing channels during the 30 days period that lead to conversions. For example consider the following hypothetical conversion path:
Fig.1
Here a visitor is exposed to 6 marketing channels before he made a purchase. Google Analytics will show this conversion path in the ‘top conversion path report’ as:
Fig.2
The conversion path is created for each conversion recorded by Google Analytics. The conversion paths are recorded via visitor cookies (_utma). To know more about Google Analytics cookies, check out the post: How Google Analytics uses cookies
Note: There is no limit to the number of conversion paths, Google Analytics can record.
Multi channel Funnel Data
This data is a combination of conversion data and conversion paths and is compiled from un-sampled data. Since the multi channel funnel data collection lags by up to two days, you can’t see this data for today or yesterday in your multi channel funnel reports. You also won’t see this data if not a single conversion has occurred on your website in the last 30 days.
Interaction
It is an exposure to a marketing channel. Interaction is also known as ‘touch’. For example in the chart above, the visitor is exposed to 6 different marketing channels before he made a purchase. Each exposure is known by the name of ‘interaction’ in multi-channel funnel reports.
Note: Google Analytics can record up to 5000 interactions per conversion path.
Types of Interactions
There are several types of interactions. For example:
- Interactions based on position – First interaction, middle interaction, assist interaction and last interaction. These interactions are also known by the name first touch, middle touch and last touch which are more commonly used in the web analytics world.
- Interactions based on type – click, impression, direct visit
- Interactions based on campaign or traffic source type – keyword, campaign etc.
Any interaction other than the last interaction can be called the assist interaction. Any interaction other than the first and last interaction can be called the middle interaction.
Types of Interaction Analysis
In Google Analytics you can do two types of interaction analysis:
- Assist Interaction Analysis
- First Interaction Analysis
As the name suggest, the ‘Assist Interaction analysis’ is the analysis of assist interactions (any interaction other than the last one) and ‘First Interaction Analysis’ is the interaction of the first interactions. You can do such type of analysis in the ‘Assisted Conversions’ report.
Channel Labels
It is the label applied to a digital marketing channel. For example ‘paid search’, ‘organic search’, ‘social’, ‘display’ etc are all examples of channel labels.
Types of Channel labels
There are two types of channel labels in Google Analytics:
- Default Channel Labels
- Custom Channel Labels
The default channel labels are the predefined channel labels. For example: ‘paid search’, ‘organic search’, ‘referral’, ‘display’, ‘email’, ‘social’, ‘direct’ and ‘other advertising’ are default channel labels. The custom channel labels are the labels defined by a user. For example, ‘branded keywords’, ‘non-branded keywords’ etc are custom channel labels.
Fig.3
Channel Grouping
It is a set of channel labels. There are two types of channel grouping in Google Analytics:
- Basic Channel Grouping
- Custom Channel Grouping
The ‘Basic Channel grouping’ is the set of predefined channel labels. The ‘custom channel grouping’ is the channel grouping created by a user. You can see the example of basic and custom channel grouping in Fig.3. Defining channel labels is part of creating your own channel grouping. You can define a channel label by creating specific rules. Each rule is based on one or more conditions:
Fig.4
Fig.5
The rules (labels) are interpreted in the order in which they are defined. To know more about defining your own channel labels and channel grouping, check out this help article from Google Analytics
Note: You can create as many channel groupings as you want.
Conversion Path Analysis
You can analyze a conversion path by changing its primary dimension in the ‘Top Conversion Paths’ Report:
You can switch to following 17 primary dimensions to analyze your conversion paths:
- Basic Channel Grouping Path
- Source/Medium Path
- Source Path
- Medium Path
- Campaign Path
- Campaign (or Source/Medium) Path
- Keyword Path
- Keyword (or Source/Medium) path
- Adwords Campaign path
- Ad group path
- Adwords Keyword path
- Ad content path
- Matched search query path
- Placement domain path
- Placement URL path
- Display URL path
- Destination URL path
All these dimensions are pretty self explanatory and if you play with them in your reports, you can get a pretty good idea of how they analyze a conversion path. You can access the last 13 dimensions by clicking on the ‘other’ drop down list in your ‘top conversion paths’ report:
Segmenting Conversion Paths
You can segment conversion paths through ‘conversion segments’. They are just like the Google Analytics advanced segments but are meant especially for multi channel funnel data.
Fig.6
Through conversion segments you can isolate and analyze specific subsets of conversion paths. There are two types of conversions segments: Default conversion segments and User defined conversions segments (as you can see in Fig.6 above).
So if you want to see all those conversion paths where the first interaction was paid search then select ‘First Interaction is Paid Advertising’ from the ‘Default Segments’ and then click on the ‘Apply’ button. If you want to see all those conversion paths where the first interaction was Facebook then create a user defined segment named ‘First Interaction is Facebook’ by clicking on the link ‘create new conversion segment’ (as you can see in the Fig.6 above) and then click on the ‘Apply’ button.
Check out this help article to learn about creating your own conversion segment.
Note: You should look at your multi-channel funnel reports in an un-filtered profile. A filtered profile can corrupt your conversion path data. Use ‘Conversion segments’ instead of filtered profiles.
What is Attribution Modeling?
It is the process of understanding and assigning credit to marketing channels that eventually leads to conversions. An attribution model is a set of rules that determine how credit for conversions should be attributed to various touch points in conversion paths
Why I should use Attribution modeling?
You should use attribution modeling to understand the buying behavior of your website visitors. Why people buy from my website? What happens before they buy? What prompted them to make a purchase or complete a predefined goal? The biggest insight that you can get from attribution model is that you can determine the most effective marketing channels for investment.
Many marketers / analyst still evaluate the performance of a marketing campaign according to the number of conversions it completed. This is sub-optimal way of evaluating the performance of a marketing channel. If a marketing channel is not directly completing a conversion, may be it is assisting in conversion. May be it is initiating the conversion process. So before you discard or label a marketing channel as ineffective or over invest in any particular channel determine following things:
- The number of conversions initiated by the marketing channel.
- The number of conversions assisted by the marketing channel.
Both ‘Display’ and ‘email’ are poor cousins of Search Marketing campaigns. This is because they generally don’t get the credit of completing a conversion. But they do/can help in initiating or assisting a conversion. So before you label them as ineffective or under invest in them, look at the number of assisted conversions in your ‘Assisted Conversions’ report.
When you are into inbound or multi channel marketing, no one marketing channel is solely responsible for conversions. Inbound marketing is like a football game. The success of the game depends upon the whole team. You can’t single out a single person and give him the whole credit for winning the game just because he happens to be the person who directly stroked maximum number of goals. The people, who passed the ball, defend the ball, the goal keeper; all play an important role in winning the game.
Types of attribution models
Attribution models can be broadly classified into two categories:
- Baseline Attribution Models
- Custom Attribution Models
Baseline Attribution Models
The baseline attribution model defines how credit should be distributed to interactions (or touch points) in a conversion path before the custom credit rules are applied. There are several types of baseline attribution models:
- Last interaction attribution model (popularly known as last touch attribution model) – This model assign 100% credit to the last interactions. Google Analytics uses this model by default.
- First interaction attribution model (popularly known as First touch attribution model) – This model assign 100% credit to the first interactions.
- Linear attribution model – This model assign equal credit to each interaction in a conversion path.
- Time Decay attribution model – This model assign more credit to the interactions which are closest in time to the conversion.
- Position based attribution model – This model assign 40% credit to the first interaction, 20% credit to the middle interaction and 40% credit to the last interaction.
- Proportional Multi Touch attribution model – I developed this attribution model back in the March of this year. This model assigns credit to interactions in proportion to their contribution in a conversion. Please note this model is not supported by Google Analytics and is known to only handful of web analyst in the whole world (off course except you now)
Note: In my ‘selecting the right attribution model’ blog post I stated various attribution models (like first touch, last touch or multi touch) as being flawed. I now disagree with these statements. My analytics knowledge has grown exponentially since I last wrote this post and now I believe that there is no one size fit all attribution model and the selection of attribution model depends mainly on the client’s business model and advertising objectives.
Custom Attribution Models
As the name suggest these models are developed by people like me and you. When you build your own attribution model, you create your own rules to assign credit to different interactions in a conversion path. These rules are known by the name ‘custom credit rules’ in Google Analytics. Unfortunately you can create these rules only in Google Analytics Premium.
How to create your own Attribution Model in Google Analytics
If you have access to Google Analytics Premium, follow the steps below to create your own attribution model (or feel free to skip this section and move on to the next nuggets):
Step-1: Go to the ‘Attribution Modeling tool’ (conversions > Multi Channel Funnels).
Step-2: From the ‘model’ drop down menu select ‘create new custom model’.
Step-3: Name your model and select a baseline model (linear, time decay or position based)
Note: You can’t select ‘Proportional Multi Touch attribution model’ as baseline model because Google Analytics Premium doesn’t support that.
Step-4: Create and apply new custom credit rules. These rules define how credit should be distributed to various interactions in a conversion path. Each rule is based on one or more conditions.
Note: Attribution Modeling tool lets you create and apply custom attribution model. It also lets you compare up to three attribution models simultaneously.
How to select an Attribution model
You select an attribution model on the basis of your client’s business model and advertising objectives. The attribution model you select has a great impact on conversion volume and conversion value. Thus they greatly impact the valuation of your marketing channels. Both conversion volume and conversion value can vary from one attribution model to the other.
For example if you select ‘first touch attribution model’ then all the marketing channels that initiate conversions will be credited with high conversion value. If you select the ‘last touch attribution model’ then all the marketing channels that completed conversions will be credited with high conversion value.
Following are some general rules you can apply while deciding an attribution model:
1. If you are a company like ‘Tesco’ and you sell products like FMCG (fast moving consumer goods) which involves least amount of consideration by a buyer, then last touch attribution model may be appropriate for you as you don’t need to assign more credit to the first and middle interactions in your conversion path.
2. If you are new player in your niche then you need more brand awareness than your established competitors. Consequently your advertising goals will be more ‘brand building’ centric. So you need to assign more credit to interactions which initiate the conversion process. For this reason, First touch attribution model can be more appropriate for you.
3. If you have a business model where each interaction (or exposure to a marketing channel) is equally important for your conversions then Liner Interaction Model can be more appropriate for you.
4. If you want to understand the buying behavior of your clients during a promotional campaign then you would like to assign more credits to the interactions which occur closest in time to the conversions as they are more relevant than the interactions which occurred few days ago. Consequently, Time Decay Model can be more appropriate for you.
5. If you have a business model or advertising objectives which value first and last interactions more than the middle one then Position based attribution model can be appropriate for you.
Conversion Values
In multi channel funnel reports there are 3 types of conversion values:
- Assisted Conversion Value
- Last Interaction Conversion Value
- First Click conversion value
The assisted conversion value is the total value associated with assisted conversions. Higher the assisted conversion value the more important a marketing channel is in assisting conversions.
The Last Interaction conversion value is the total value associated with last interaction conversions. Higher the last interaction conversion value the more important a marketing channel is in completing conversions.
The First Click conversion value is the total value associated with first click conversions. Higher the First Click conversion value the more important a marketing channel is in initiating conversions.
How Assisted/Last Interaction conversion of a marketing channel is calculated and what insight I can get from it?
This ratio is calculated as:
= Number of assisted conversions/Number of last interaction conversions
If the value of this ratio is close to zero then it indicates that the marketing channel functions primarily in completing conversions.
If the value of this ratio is close to 1 then it indicates that the marketing channel functions equally in both assisting conversions and completing conversions.
If the value of this ratio is more than 1 then it indicates that the marketing channel functions primarily in assisting conversions.
How First/Last Interaction conversion of a marketing channel is calculated and what insight I can get from it?
This ratio is calculated as:
= Number of First Click conversions/Number of last interaction conversions
If the value of this ratio is close to zero then it indicates that the marketing channel functions primarily in completing conversions.
If the value of this ratio is close to 1 then it indicates that the marketing channel functions equally in both initiating conversions and completing conversions.
If the value of this ratio is more than 1 then it indicates that the marketing channel functions primarily in initiating conversions.
Case Study
How to measure the performance of branded Vs Non Branded Keywords?
Many marketers/analyst will create and apply branded and non branded advanced segments on keywords report and then toggle between the Goals and e-commerce tabs. While this approach is better than nothing, it won’t produce optimal results. Certainly not fit for an analytics ninja.
Remember Google Analytics use last touch attribution model by default. So whatever conversions and ecommerce transactions you see in your regular analytics reports are attributed to the last interactions which directly resulted in goal conversion or ecommerce transaction. You need much more insight than this.
So head towards your ‘top conversion paths’ report and create a new ‘custom channel grouping’. Name it ‘Branded Vs. Non Branded Keywords’. Now check Fig. 4 and Fig.5 above. You define the ‘branded’ and ‘non branded’ channel labels using the same conditions you use while creating advanced segments. The ‘Branded Keywords label include those keywords which match a certain regular expression like: s?eo ?takeaways)|seo|takeaways.
Once you have defined your labels, apply the new custom channel grouping. Note down the number of assisted conversions and conversion value of branded and non-branded keywords.
Fig.7
From the chart above, we can see that the non-branded keywords have ‘assisted/last interaction’ value (1.15) higher than that of ‘branded keywords’ (0.35). Also the ‘assisted/last interaction’ value of ‘non-branded keywords’ is more than 1. This means the ‘non-branded’ keywords are more important in assisting conversions than completing conversions.
Since Google Analytics by default shows the last interaction conversion values, we may assume that only branded keywords are driving sales and conversions if we just rely on the regular analytics reports. If we stop bidding/targeting on non-branded keywords, then the assisted conversion value of non-branded keywords will go down and since non-branded keywords are assisting conversions a lot more than branded keywords, the overall sales will eventually go down. So the rule of thumb is, determines the assisted conversion value and assisted/last interaction conversion value of a marketing channel before you discard/label it as ineffective. Similarly through ‘first interaction analysis’ of the branded and non-branded keywords we can determine which one is initiating more conversions than the other.
