Day 3 – How to Develop Multichannel Attribution Models That Move the Needle
Speakers:
1. Janet Driscoll Miller (@janetdmiller)
2. Simon Poulton (@spoulton)
3. Adam Proehl (@adamproehl)
Moderator:
1. Ginny Marvin (@ginnymarvin)
One of the most common challenges in marketing today remains the ability, or lack of it, to measure the success of programs. This is especially true in B2B.
To go about the task of measuring success, it’s important to figure out what to track, specifically:
- Source/medium – Which channels are being tracked
- Content type – Specific offers such as webinars, guides or checklists
- Content focus – Such as paid versus organic/earned channels
Other than Google’s data-driven attribution (DDA), there are a few standard classes of attribution commonly used in B2B:
Single-source (which attributes all credit for conversions to a single click)
1. First touch/First click – Attributes all credit for a conversion to the first click in the funnel
2. Last touch/Last click – Attributes all credit to the last click immediately preceding conversion
Factorial (which measures multiple factors)
1. Linear – Attributes conversion credit equally to all clicks in a specific journey
2. Time decay – Attributes partial credit to clicks in a specific journey, emphasizing the latest clicks the most
Choosing which attribution model will work best for you will often come down to testing.
For B2B, Miller recommends using a combination of Google Analytics to track top-funnel interactions; marketing automation solutions like Marketo to track mid-funnel interactions for marketing-qualified leads (MQLs) and sales-accepted leads (SALs); and customer relationship management solutions like Salesforce to manage deep-funnel interactions such as sales-qualified leads (SQLs) and closed-won deals.
A common challenge in marketing attribution is figuring out whether to invest in programs that are proven winners versus programs that have not yet driven significant revenue. There is no clear-cut answer to this question – instead, attribution should be a study of where to deploy future resources for potential future performance.
Google has already unveiled a number of standardized attribution models including last click, first click, linear, position-based and time-decay. However, Poulton argues that Google’s machine learning-based DDA is the future.
DDA volume requirements – It should be noted that in order to use DDA, advertisers must have recorded a minimum of 15,000 clicks and 600 conversions within the past 30 days.
Poulton compares the concept of DDA to Lloyd Shapely’s cooperative game theory, which attempts to assign “credit” for different players on a team for contributing to a game outcome. Specifically, Shapely’s theory attempted to determine attribution should a specific player be removed from the team – and how much of the burden would respectively be shouldered by the remaining players.
Therefore, DDA assembles sales and performance metrics from various channels such as Google Shopping, brand and non-brand terms, computing normalizing factors for outsize contributions of one channel or another. These aggregate calculations also take into consideration counterfactual gains (the most obvious example being individual channels whose isolated revenue doesn’t add up to the total for the campaign).
Limitations of DDA – While DDA seems like an exciting new field to explore, it’s still a work in progress and its algorithm remains a black box that advertisers ultimately have to trust on faith. More importantly, DDA is available only for Google and doesn’t talk to other important channels such as Facebook.
