Find the other parts in this series here:
- Part 1: The state of attribution in 2023
- Part 2: How we do attribution (this post)
- Part 3: Attribution outputs & actions
What is Multi-touch attribution?
Multi-touch attribution is a method of marketing measurement that accounts for all the touchpoints on the customer journey and designates a certain amount of credit to each channel so that marketers can see the value that each touchpoint has on driving a conversion.
The core functionality of an attribution engine is its ability to match touches to conversions based on a series of rules, known as ‘attribution models’.
When tackling an attribution project, it’s important to first consider the key areas that will feed the engine and its definitions. We’ll go through the main ones: the models. the touch points, identity definitions, and relevant marketing attributes.
Area 1: Multi-Touch Models
The first step in the Tasman Attribution Engine is that we select one or multiple MTAs to report or analyse our data with.
There are two types here.
Firstly, we have relatively standard marketing-theory models such as:
- Last touch - 100% conversion credit is applied to the touch point immediately before the conversion event.
- First touch - 100% conversion credit is applied to the earliest occurring touchpoint.
- U-shaped - 40% conversion credit is given to both first and last touches, with the remaining 20% split across all others.
We also can use more data-driven models (slightly more scientific) such as
- Shapley - the Shapley model considers the marginal impact of each channel on the overall conversion. It is attractive because it does not rely on user-level data. It does not however implicitly include the concept of position in the conversion journey.
- Markov chain - a Markov model considers a sequence of touchpoints and evaluates how they contribute to a Lead that is changing states as it moves through the conversion process. They are mathematically attractive but can be very complicated to explain.
- MVR - multivariate regression is able to consider a bespoke model of the conversion journey and so is the most adaptable to a specific setup. It can be complex to setup and solve though and so requires more resources to run and interpret.
- and Survival analysis - survival analysis considers the experiences of users that don’t convert more explicitly alongside those that do convert. It can be useful in environments with very low conversion rates.
Area 2: Touchpoints
What its defined as a touchpoint? These being marketing interactions that provide a meaningful change in the consumer attitude or behaviour. Examples of touchpoints well-defined could be:
- Web Sessions - sessions are defined in logic that is applied to segment pageview events. A session is defined for a specific user. A session starts when there has been no activity on the website for longer than 30 mins OR the user has left the website and returned via a new channel. A session stop is captured as the last event within a session.
- Mobile Sessions - sessions are defined on mobile web as they are for desktop website. For the mobile app, it is defined using app_background and foreground events. A session starts when the last app_background event happened more than 30 mins ago.
Area 3: Identity
All attribution requires the concept of an individual user to be tightly defined. If we cannot identify a user then what should be analysed as related touchpoints will be lost. Tracking tools like Segment’s identity (anonymous_id) and Snowplow’s identify (domain_user_id) IDs provide us with a secure user identifier for the website and (separately) for the mobile web.
When having log-in capabilities and the user logs-in, it will be used to identify it with the client-id as a customer.
- When tracking is well implemented, all historical sessions can be retrospectively tied to the single user at the point of login.
Where there is no login, we may not be able to tie desktop and mobile devices together. This can be a problem in two main areas:
- Companies that have a website and app strategy and haven’t thought enough about user identification.
- For activities such as emails because they are typically opened on mobile devices but purchases can be made on the desktop or vice versa, therefore need to build a system to tie the mobile devices and desktop devices together using an identification that the CRM tool may have available for the different channels.
Area 4: Marketing Attributes
For each touchpoint, we want to define a set of marketing attributes that allow us to analyse the touchpoint. Typically this information will be captured either through UTM parameters or through the HTTP referrers and can be tied to a session. For Email opens, these can be captured from the source data.
Each session will be connected to one marketing attribute-set. This will provide the primary information associated with that touchpoint, but because different channels have different levels of granularity, there are three tiers (some times more!) of marketing attribute-sets for each touchpoint.
Some examples of attributes are:
- Channel Grouping (Organic/Paid)
- Specific marketing channels
- Campaigns
- Audiences
- Ad / Creatives.
Area 5: Conversion events, flows, and time-windows
One of the most important steps when working with attribution is to define the conversion flows as well as which conversions are going to be the ones to use.
The engine allows for multi-touch as we mentioned, but also multi-stage. If you are a B2B company and the goal of your marketing team is to drive qualified leads, but most of the control they have is volume of leads, it has a way for them to understand attribution not only based on volume but in all the different stages that a lead would have between Marketing and Sales (Marketing Lead, Qualified Lead, Sales Qualified lead).
The model is so versatile that it can address challenges involving distinct flows and phases spinning in multiple months.
Consider an event where the primary objective is to generate awareness and gather leads for an early ticket release. Months later, when tickets become available, the goal shifts to selling tickets. However, the effectiveness of the marketing strategy hinges on evaluating the effectiveness of the initial lead generation phase.
The work to do here is to know exactly what needs to be tracked, and map it out to the sources and flows.
So, what is next?
Once all these steps have been agreed and mapped out with the sources, channels have been reviewed to capture how teams use them and ensured there is a mapping between them and the data being collected, and tracking has been checked to ensure the model has good data quality, the development kicks off.
What you can expect out of these models, and what impact it has on the business, is something you can expect in our next blogpost.