Attribution Modeling is a mouthful to say and an earful to listen to. Digital and database marketers use attribution modeling to find the most effective mathematical method of attributing value to a point on the conversion path so you can optimize and repeat thereby acquiring customers at the lowest possible cost while delivering the highest conversion rate and highest AOV. And this is where you can find a rabbit hole and wonder, will the LTV make up for the low 1st purchase AOV. And this is why this subject is so complicated and why you need a detailed analyst documenting all hypothesis and outcomes with a complete testing matrix. You will find that each customer segment may have a different model based on the segment persona.
Here is an example of why you need to evaluate your acquisition with attribution modeling:
A customer is reading a blog about a product she did not know she wanted or needed and poof, the interest is sparked. She now starts searching to figure out which model she needs. She clicks on your Adwords ad and goes to your website. She browses 4 pages and leaves. Three days later she comes back to your site from social media and browses 4 more pages. A day later she clicks on your email marketing and returns to your site to browse. Four hours later she makes a purchase.
…Now, which touch point is responsible for the sale? You will have many analysts in a heated debate over this. May be the only time analysts get heated. Well let’s take a look at Google’s 7 different models to see which approach you may take to solve this quandry.
Also before we start, let’s get some ground-rules or definitions out here so we’re on the same page going forward. Goals can be less than a transaction, meaning a goal in your GA model may be a product page visit, then an add to cart, then check out, then review order, then confirm order and then the final goal is the thank you page which is your conversion. For attribution modeling, the total conversions is the sum of goals and eCommerce conversions. Naming conventions for acquisition media also varies: paid search, organic search, social media paid, social media organic, eMail and Affiliate are all examples of channels. In attribution modeling, these acquisition verticals will be referenced as channels.
A few of the multitude of marketing challenges that attribution models can aide in answering are the following:
- Which combinations of media channels contributed to this specific sale for this specific persona?
- What is the time-lag from first interaction to eCommerce conversion? Can I impact this?
- How to attribute marketing channels to a conversion.
Today we will explain how Google approaches attribution modeling and the tools they make available in Google Analytics to test your attribution theories and find the models that work best for each of your customer segments. Google also provides an Attribution Comparison Tool to show the different models using your data so you can determine best fit. You can also create your own model and test it here.
The Last Interaction model attributes all credit and value associated with the conversion to give 100% credit to for the conversion to the last channel the customer came through.
Tip: This model may be best used for direct response style selling or for products that lack a consideration phase. This may include low-priced products and services or impulse and emotional services. This can be tricky as the AMA says 75% of all transactions are emotional. One to keep in your pocket and try on for size with each new product and/or channel introduced.
The Last Non-Direct Click model ignores any direct interaction with your website and/ or online brand besides the last non-brand owned interaction. This means the last channel you came from prior to visiting direct would be given 100% credit for your conversion. This is how GA attributes conversion in tradition analytics not employing multi-attribution.
Tip: This model is the default in GA and considered a great benchmark for how you won customers prior to them getting to your website. You may consider filtering out direct traffic and this will get you to the referral level of understanding your acquisition.
The Last AdWords Click model is a little self-serving, if ya know what I mean here. Ahem, this model gives 100% of the conversion credit to the last AdWords ad clicked on. I am sure you can see how self-serving this is to Google. If this model somehow fell on def or dumb, we all would pour all of our ad dollars into AdWords. Luckily free will and intelligent analysts really dwell on these models and this has yet to happen.
Tip: This model is effective in identifying your most effective AdWords Ads that drove the most conversions. This can be determined n AdWords console as well.
The First Interaction model credits the first channel engaged with the entire 100% of the conversion credit. This theory asserts that it was that first channel that changed the prospect’s behavior and everything after that was as a result of the first channel and creative.
Tip: This is a great model for launching a start-up brand where brand awareness is at ZERO and you need to determine the channel and creative that elicited enough emotion to stop the scroll and change behavior in an instant. Based on media studies, we know it was probably not this ad, but this is all we have that we can track.
The Linear model gives equal credit to each channel along the conversion path. This model seems the most fair until genuine multi-variate predictive analysis using regression can be employed.
Tip: This follows a more traditional marketing funnel where you expect to have multiple touchpoints and engagement with the prospect or returning customer many times before conversion. Think about your product or service and price point.
The Time Decay model is based on the theory of exponential decay and focuses on the channel touch points closest to conversion. The half-life on this model is 7 days, meaning that a channel touch point from 7 days ago will get half the credit of a channel that was used the day of the conversion.
Tip: This model is good for products and services that have a low consideration phase. This is typically a low-priced product or consumable that the customer needs for existing products they own. This is also a good model to use when you run one or two day promotions. You want to credit the impetus not the previous brand interactions as you know the price is what drove the conversion here.
The Position Based model enables you to create a hybrid model between the last interaction and first interaction models. This puts you in the driver’s seat knowing your customers and your channels. You can split the value how you want, though a commonly agreed use of this equation is to assign 40% to both first and last and divide the remaining 20% across the channels in between.
These are Google models, you can derive your own models as well as use even more advanced methods using multi-variate statistical analysis with software such as SAS or SPSS. Share with us experiences where models could have made or broken the acquisition strategy.