What is an Attribution Model?
Put simply, if you need to figure out how to allocate your marketing spends across digital channels to maximise your ROI, then you need an Attribution Model.
An Attribution Model is a framework used in marketing to determine how credit for conversions or sales is assigned to different marketing channels and touchpoints along the customer journey. These models help marketers understand the effectiveness of their various marketing efforts and make informed decisions about resource allocation and strategy optimization.
Importance of Attribution Modeling
Attribution Modeling is crucial for marketeers for several reasons:
- ROI Measurement: It helps marketers determine the return on investment for different marketing channels and campaigns1.
- Budget Allocation: By identifying the most effective channels, marketers can allocate resources more efficiently3.
- Customer Journey Insights: Attribution Models provide a clearer picture of how customers interact with various marketing touchpoints before converting2.
- Strategy Optimization: Marketers can refine their strategies based on which channels and tactics are most influential in driving conversions4.
- Cross-Channel Analysis: Multi-Touch Attribution Models (MTAs) allow for a more comprehensive understanding of how different channels work together3.
In terms of types of Attribution Models, there are two main categories:
Single Touch Attribution models
- First-click attribution: Credits the first touchpoint a customer interacts with
- Last-click attribution: Assigns credit to the final touchpoint before conversion
Multi Touch Attribution models (MTAs)
- Linear attribution: Gives equal credit to all touchpoints
- Time-decay attribution: Assigns more credit to touchpoints closer to conversion
- U-shaped (position-based) attribution: Emphasizes first and last touchpoints
- W-shaped attribution: Highlights first, middle, and last touchpoints
- Algorithmic (data-driven) attribution: Uses machine learning to determine credit distribution
How can MTA models accurately assign value across touchpoints?
This is a very good question because it relies heavily on understanding how your customers journey to your website via the different advertising touchpoints.
But even if you have a feel for these journeys how can you accurately assign values to these touchpoints? And won’t these values change as you change your marketing mix?
With new data processing developments and AI more are turning to the Algorithmic models. Data-Driven Attribution is a dynamic attribution model that looks at your users’ source/medium/campaign combinations over their customer journeys and attempts to determine which are most crucial to the conversion. It uses a weighting system to distribute credit: combinations that were more important toward driving a conversion are weighted more heavily, and less important combinations are weighted less.
Other potential issues with MTA models?
MTA models rely on the use of cookies, and with growing browser limitations on 3rd party cookies, the accuracy of tracking user journeys is becoming more and more difficult. We have Google phasing out 3rd party cookies on Chrome browser (which accounts for 60% of users), and Safari already did 2 years ago (12% of users). According to Measured.com, soon after the roll out of Apple’s ATT (Apple’s user privacy framework requiring apps on IOS to request user permission for tracking users), Meta found that last click models undervalued marketing on FB by 47% on average.)
MTA models will only look at the online journey – but what about the effects of offline marketing and word of mouth?
What about Marketing Mix Models (MMM)?
MMMs essentially look at the data from a top down perspective, and use multi variable linear regression techniques to correlate marketing activity with the target KPIs. What about Marketing Mix Models (MMM)?
In doing so, MMMs don’t assign weightings to touchpoints, the maths does it for you. So the issue of weightings and their accuracy is removed.
And the maths used does have credibility – climate control experts have long used multi variable regression techniques for years/decades to predict the resultant global waming we are now seeing.
And that word ‘prediction’ is important in the last sentence. MTA models, being measurement based essentially look backwards. MMMs however, being analytics based, look forwards to give predictions. Both have their role (we need to understand what is working right now and by channel, but we also need to know the outcome of the current mix as we move forward). But it is important to understand that MTA models will tend to be be weak on predicting, unless of course very large data processes are set up and teams to manage.
Time lags are an important point. Different digital marketing channels will take different times to have an impact. Brand awareness campaigns are a good example. Typically used to feed the top of the funnel, their effect can last far longer than the more immediate Paid Search campaigns responding to people’s searches on the brand. Some people may repond later to a brand campaign and completely miss the other more immediate digital campaigns and be missed off the measurements of ROI. An MMM approach would be more likely to pick these up.
Ideally you should use both models – an MMM to set overall planning going forward, and an MTA to measure the granular results to see if it needs fine tuning.
What KPIs and metrics align attribution models with business objectives?
That depends on the business objectives.
Aligning KPIs to the business objectives needs to be done by working closely with the different department staff of the business.
Getting them involved at this stage not only ensures KPI alignment but brings more of the business in to the model building, and this is important in terms of getting resource (spend and staff) but also to raise the profile of this initiative in the wider business as a key way to drive profitability going forward.
[EG: For a client we linked GA to a CRM software to measure and optimise the journey from digital acquisition of a lead through to offline conversion of the lead from sales staff. KPIs were Cost per Acquisition and Lead Conversion to Order rate, aligning to the business objective of Revenue.]
When selecting the metrics to use, remember they need to clearly measurable and actionable (i.e. clear paths to how you can improve going forward).
How to leverage real-time analytics to optimise campaigns and budget allocation?
Real time data is important for marketing teams to remain agile, response and optimise performance as quickly as possible.
But how realistic is it to work in real time? Do you have the staff resource for this, and is it not better to let the model work for a few days?
And as mentioned above, different campaigns take different time periods to have impact.
If you do need real time data then there are some key steps:
- Aggregating your data – real-time sources include server logs, app activity, online advertising, and clickstream data. Then connect to a stream processor.
- Implement your stream processor – using a tool like Amazon Kinesis
- Perform real time queries – using Amazon Kinesis Data Analytics or Google BigQuery
- Set up alerts triggered by events
Attribution Platforms?
Here are some key strengths and weaknesses of different marketing attribution platforms:
GA4
Strengths:
- Free to use
- Integrates well with other Google products
- Offers data-driven attribution model
- Provides cross-device and cross-platform tracking
Weaknesses:
- Limited customization options
- Lacks some advanced features of paid platforms
- Data sampling can impact accuracy for high-traffic sites
- Privacy limitations due to reliance on cookies
Adobe Analytics
Strengths:
- Robust customization capabilities
- Advanced segmentation and analysis tools
- Integrates well with other Adobe products
- Offers both rule-based and algorithmic attribution models
Weaknesses:
- Expensive, especially for smaller businesses
- Steep learning curve
- Can be complex to implement and maintain
Mixpanel
Strengths:
- Strong event-based tracking
- User-friendly interface
- Good for product analytics
- Offers cohort analysis
Weaknesses:
- Limited marketing-specific attribution features
- Can get expensive at higher volumes
- Less suitable for content-heavy websites
Amplitude
Strengths:
- Powerful user behavior analysis
- Machine learning-powered insights
- Good for product-led growth companies
- Offers predictive analytics
Weaknesses:
- Focused more on product than marketing analytics
- Can be complex for non-technical users
- Limited integrations compared to some competitors
AppsFlyer (Mobile focussed)
Strengths:
- Specialized in mobile app attribution
- Offers fraud detection
- Supports deep linking
- Provides real-time data
Weaknesses:
- Limited use outside of mobile apps
- Can be expensive for high-volume apps
- Some users report occasional data discrepancies
Segment
Strengths:
- Centralizes data from multiple sources
- Offers data governance features
- Flexible and customizable
- Supports real-time data streaming
Weaknesses:
- Not a full attribution solution on its own
- Can be complex to set up and manage
- Pricing can escalate quickly with high data volumes
Attribution softwares?
Amazon Kinesis or Google BigQuery can be used for real time queries.
Overall softwares: Proof Analytics, Supermetrics, WhatConverts (includes tracking of chats, calls and forms tracking), Demandbase (for B2B), C3Metrics (works across digital, TV, radio and direct mail), Rule Analytics (can move between model types, and looks at offline), Hubspot Marketing Analytics (good for bridge sales and marketing data, and can integrate with large range of other apps) and Cake (for real time tracking and optimisation).
Helpful tips to consider when building an Attribution Model
- You should include looking at GA4 Acquisition reports, particularly First User and Session – to get a feel for top of funnel effect and last click view. This will give you another way of looking at the results of your Attribution model as a way of sense checking
- Be careful using Google’s Performance Max campaigns (also run by AI) as part of your Attribtuion :
- It can leave you promoting pages and assets that drive CTR and CPC but don’t drive conversions, or at least the conversions you want. So turn the auto create assets off.
- It will include Brand campaigns, which will score well on KPIs but might not get the conversions of services you want. Best to turn off Brand and create a separate Brand campaign.
- It may cannibalise services that perform well Organic Search and don’t need promoting
- It can compete with other campaigns so you end up bidding against yourself so check impact on other campaigns when putting Performance Max live
- Also isolate and test each ad campaign to make sure you get the best performances and that one is competing with another.
- Look at MCA (Multi Channel Attribution) (a blend of MTA and MMM). MCA (MTA+MMM) looks at the entire offline and online journey. The challenge is connecting the offline and online behaviour. Some softwares/techniques to consider: Foot Traffic (Mobile Marketing companies using beacon technology to determine location of smartphone users), POS data (data from credit card companies on locations of purchase), customer panels (matching location to customer behaviour).
