Attribution Modeling SimplifiedBy Uri Bishansky on May 26, 2016
Reading Time: 17 minutes
While going through their buying journey, users will visit a lot of sites. Depending on the visitor’s circumstances, they may be using different devices. It is not uncommon for a user to visit your site many times before making a purchase decision.
Most marketers measure the effectiveness of their advertising channels by marking the channel a user came from when they converted. But what about all the other supporting channels that the buyer passed before that final encounter? How much did they help or hinder?
In this post, I will dig deeper into the buyer’s journey and teach you how to measure this more precisely using attribution modeling.
So what is Attribution Modeling?
Definition: “An attribution model is the rule, or set of rules, that determines how credit for sales and conversions is assigned to touchpoints in conversion paths.”
You probably already take actions based on the progress of your business goals (sales, leads, etc) through one or more marketing channels. This is measured through analytics programs. All of these programs have some sort of attribution modeling, like the aforementioned link between the site your visitor came from just before the conversion happened.
So without even knowing it, you are already using attribution modeling. The problem is that you are missing out on a lot of information from data you already have and making important optimizing decisions based on partial information.
This is where more advanced Attribution Modeling techniques come in handy.
Let’s take an example of a college that advertises to attract and enroll new students to next year’s MBA studies. Let’s call this college BIA (Business Is Awesome.)
Joe is an accounting grad and wants to expand his knowledge. He searches the web for information on post-grad studies in his field.
Here are the steps that took place before he converted on BIA’s landing page for MBA studies (only the steps that took place on BIA’s site):
|1||Google Search||Search query: post-grad studies for CPA’s|
|2||Facebook Remarketing Ad||Landed on a BIA landing page|
|3||Google Search Ad||Search Query:
“MBA Studies for CPA’s”
|4.||Google Display Ad||Converted|
In this example, the user had 4 touchpoints with BIA’s site before actually converting.
Google analytics will show the source for the lead as a Google Display ad since this was the traffic source when the user converted.
Don’t Google Organic, Facebook Ads, and the Google search ad deserve to get any credit for this conversion? Of course, they do.
This is attribution modeling: attributing to each touchpoint the value it deserves for contribution to the conversion.
Common Attribution Models
Here are some attribution models you may find useful. Later in the post, you’ll learn how to make your own.
All the conversions/conversion value is attributed to the source of the first click. In our example, the Google Search. This attribution is mainly used to measure the impact of a brand-awareness campaign. Beyond that, it doesn’t have much use.
All the conversions/conversion value is attributed to the source of the last click – this is Google Analytics’ default attribution model and the most common among other analytics companies. It is very popular in the mobile app industry.
Each touchpoint is attributed an even amount of conversion/conversion value. While this seems the most logical attribution model to measure everything, it’s not the most accurate. Nevertheless, it is quite simple to implement and measure and will provide you with much more relevant insights than last/first click.
This model attributes a higher attribution percentage to the touchpoint the closer it gets (by time) to the conversion itself. It is most relevant to businesses who run time-bound campaigns and want to attribute the most of the value to the clicks that were closer (by time) to the campaign. This model can be hard to implement because you have to judge how much to attribute to a click in a particular time span.
This model combines first click, last click, and linear models. The first and last touchpoints are attributed X% of the conversion value and the rest of the touchpoints are attributed linearly the rest (100-X)%.
These are just some of the available models, and the most useful. Here’s a list of the built-in model types in Google Analytics:
Let’s apply these models to our BIA example and try to figure out the ROI of our channels. Let’s say Joe eventually enrolled and signed up for the MBA. He will pay about $100,000 (Average customer lifetime value).
|Google Search – $100,000|
|Facebook Ad – $0|
|Google Search Ad – $0|
|Google Display Ad – $0|
|Google Search – $0|
|Facebook Ad – $0|
|Google Search Ad – $0|
|Google Display Ad – $100,000|
|Google Search – 25,000$|
|Facebook Ad – 25,000$|
|Google Search Ad – 25,000$|
|Google Display Ad – 25,000$|
|(We don’t have enough information for this example.)|
Position Based (70/30 example)
|Google Search(35%) – 35,000$|
|Facebook Ad(15%) – 15,000$|
|Google Search Ad (15%)– 15,000$|
|Google Display Ad(35%) – 35,000$|
In this table, you can see how much of the profit from Joe is attributed to each touchpoint for each model. The reason we don’t have information for the time-based model is that we’d need a time measurement to compare.
Attribution Modeling using Google Analytics
Let’s see how to apply and compare the different standard attribution models using Google Analytics. On the sidebar on the Conversions tab, choose Attribution -> Model Comparison
In the attribution modeling home screen, you can see how much is currently attributed to each channel. The default Google Analytics model is Last Click.
Here’s what the sections mean:
- Conversion: Choose which conversions to include in the report
- Type: What type of conversions (Adwords/All)
- Lookback window: Choose the timeframe when the initial clicks occurred and add it to your report (1-90 Days)
- Model A and Model B: These let you select different attribution models for comparisons. The default is for Model A to use Last Click.
- Primary Dimension: Choose the dimensions breakdown (Source/Medium/Default/Other)
If you select two models, a new report will be generated comparing one model to the other.
As you can see in the above example, paid search has 16.45% more conversions attributed to the position based attribution model compared to last interaction model.
Creating your own attribution model using Google Analytics
Each business has its own unique niche, USP, and buyer personas. Therefore, one attribution model may fit one business but not another. Some businesses need to define their own unique attribution modeling. This is where Google Analytics’ ability to create custom models comes in handy.
Click the model comparison view and select a model to compare to the current one.
Click “Create new custom model”. This screen will appear:
Here’s what the different sections mean:
- Model name: This is the name of your model that will appear in the reports
- Baseline Model: Set the basis for your new model
- Set half-life of decay: This element exists when you choose time decay as a base model and determines the time frame to decay by.
- Lookback window: This sets how many days you want to look back prior to the conversion for measurement.
- Adjust credit based on user engagement: If you’re measuring user engagement rather than conversions, this lets you decide how you want to measure it, whether by time on the site or by page depth.
- Apply custom rules: This enables you to set various rules regarding different dimensions and segments. Complex And/Or statements can be applied. This is the meat-and-potatoes part of your custom model.
Revenue Attribution – ROI Measurement
To actually calculate and measure ROI through your attribution model, we must drill down deeper and understand the true impact that each channel has on our revenue. We have to make sure that our attribution model covers all the bases.
I’ll start by dividing all commercial websites in the world into 3 different segments:
- Lead generators
Each one of the websites has its own unique characteristics and must be dealt with differently. E-commerce websites are the simplest to understand for tracking attribution. Lead generator websites and SaaS websites are much more complexed and require a different approach.
In the breakdown below, I will address NEW user Acquisition and not user retention or lead nurturing, which is a completely different story.
E-commerce sites are self-contained. All the activity occurs online, which makes it easier to measure. Google Analytics is great for measuring attribution for e-commerce sites. The trickiest issue is cross-device tracking. Google won’t automatically link the same user when they use two different devices. This can be solved by implementing user registrations for your site. That way, when they log in from other devices you can place a cookie to track them.
Lead generation websites struggle with a different issue. The actual sale may not occur on the same site or even online. Therefore, it’s hard to get an accurate measure of attribution ROI, though you can make a guesstimate from overall averages. Without tracking offline conversions, you won’t get a clear picture. In order to address this issue, I have compiled another post regarding Offline Conversion Tracking in Google Adwords. If you combine the techniques between this article and that one, you’ll get a complete picture of activity and make accurate comparisons.
These are the hardest to crack. User acquisition funnels are much longer and it takes more time to progress through the funnel up until they actually pay. During their journey, the buyers might, and probably will, use different devices. There are also often multiple conversion types with SaaS sites. For instance, a SaaS site may have a trial or free version. When do you start and stop measuring different points on the journey? Should the visits that happened prior to signing up for the trial count in the model after the trial starts? These are tricky questions.
In order to understand the buyers journey on SaaS sites, I will demonstrate the complexity using a fictitious example. A cyber security company has a SaaS product that assesses your website daily for cyber security issues, threats, and attacks. The site provides solutions and active defenses to secure your website. This product is cloud based and is distributed by a SaaS Product called ACS. ACS has a 60-day free trial and provides some of the features available. After 60 days it stops working unless you upgrade to a paid account.
Here is an example of one of the users journey:
|1||Google Search||No action|
|2||Google Search||No Action|
|3||Direct Ad||No Action|
|4||Facebook Ad||Registered for a trial|
|5||Email Marketing||No Action|
|6||Facebook Post||No Action – Trial ended|
|7||LinkedIn Post||No Action|
|8||Email Marketing||Upgraded (120 days after first click)|
How much credit should each channel earn? How do we track a 120-day buying cycle if Google Analytics has a 90-day lookback window?
Let’s roughly divide the buyer journey to 2 segments:
- Non-registered user
- Registered user
Up until the user registers or submits his credentials, the tracking stays the same. But, when the user first registers – that’s when the magic begins! This is where the connection between the marketing and sales (CRM) takes place.
What we need to do is to copy all of his touchpoints up until the point he registered and save the path on the CRM. Now, once you know who the user is, every touchpoint is added to his record on the CRM. This helps us create a longer, more accurate time-unbound tracking. Once the user upgraded his account to a paid account, the attribution takes place.
We can then measure ROI for each of the touchpoints in the 120 days of data. But, how much should we attribute? The user just upgraded his account. Worse, let’s say that the customer can choose submodules for the program that add additional cost, and they can quit the service at any time.
There are two approaches to this issue:
1. Using average customer lifetime value (LTV) – we can take the number of customers we can attribute to each channel and measure that against the average LFT of the customer. This problem with this is that the data will be normalized. Great for a broad picture, but terrible for specifics.
2. Look at a small subset – This involves calculating the revenue from a specific set of users to make predictions. For example, a company might look the users that registered between 6 and 12 months ago and what they spent..
The best practice is the combine the two. Customer LTV will do just fine for the ongoing optimization, but once in a while you should take longer, deeper lookbacks to see if the data stays the same. This is very difficult to do manually. I recommend using attribution software to assist:
I personally recommend using HubSpot as a complete marketing cloud, especially for SaaS companies that are practicing Inbound Marketing. Don’t confuse Inbound marketing with PPC. PPC is a part of Inbound marketing but that’s a whole different discussion. You can learn more about it in this post about Inbound Marketing vs. PPC.
Pitfalls to avoid
Like any other model, there are ways to make it inaccurate. Here are some of the common errors.
Not testing or testing without a coherent thesis
Like every digital activity – attribution models should be tested constantly. Since these tests take much longer than a/b testing, usually a few months, you have to know what you’re testing.
Testing different theses simultaneously or changing between tests rapidly will confuse your results. Try to make each test as clean as possible so the results will be obvious and you won’t make inaccurate assumptions.
Google Analytics has a known issue of breaking sessions if not handled correctly. This results in incomplete data, and could completely distort the insights. Session breaking happens when Google Analytics creates two or more separate user profiles for the same person. Make sure that all the funnels and all the pages the users can reach on your website do not cause a session break.
What can cause sessions to break?
- Cross domain tracking – sometimes businesses use different sub-domains for different purposes (blog, an app for SaaS, etc). Crossing from one sub-domain to the other if unhandled can cause a session break
- Incorrect implementation – not implementing the Google Analytics tag correctly throughout the website can cause sessions to break. Try consolidating all your tags using the Google Tag Manager.
- Sessions are taking too long – especially on SaaS products – sessions could take longer without activity than normal websites. If a session times out it will break. Try maximizing Session Time-out in Google Analytics or create “keepalives” to show Google that the sessions are still active.
Trusting the system too much
Always test the information you have on your systems to make sure it’s in line with the actual data and that there are no significant discrepancies.
Not drilling in deep enough
It’s very compelling to use raw data and round up averages from different segments and then optimize with respect to the average. You might assume that it’s much easier, faster and probably good enough.
Well, it isn’t. For some businesses, a 5% difference can shift a campaign from profiting to losing. And, of course, the top 5% of your revenue is the most profitable. Don’t settle for average data reporting. Dig in.
Missing out on the big picture
You must remember the attribution model is a model and not a fact. It considers a set of rules you set for it. The rules are different for each business and every model should be tested to see if it generates more business or not. Always question the model. Trust, but verify.
No matter what type of business you have or you are advertising for, harnessing a correct attribution model can greatly increase your campaigns and business profitability. Not understanding the actual contribution each marketing channel provides to your business could lead to future business failure. There’s a sentence I like, I’m not sure who said it – “If you don’t move forward you are moving backward” and exactly that way with business growth.
Always test the effectiveness of your campaigns. Focus on increasing the ROI while examining where you can find new audiences and channels to add to your media mix.