Data-Driven Attribution Models Explained

Data-driven attribution uses machine learning to decide how much credit each marketing touchpoint deserves for a conversion. Rather than giving everything to the last click, it compares the journeys of people who converted against the journeys of people who did not, and assigns credit to the touchpoints that actually changed the outcome.

It is the default model in Google Ads, and one of three models available in Google Analytics 4. Google builds it on the Shapley value method. Because it reads whole journeys instead of single clicks, you can judge campaigns on one number, the revenue or conversions they contributed, instead of reading clicks and click-through rates side by side.

The two different data-driven attribution models

There are two widely used algorithmic attribution models: Shapley value (this is the one Google uses) and Markov model.

The markov model can work on less data and is faster computationally so it’s easier to go down to keyword and content level in the analysis when there are less conversions.

Data-driven attribution takes into account all marketing touchpoints in a customer journey. So it is multi-touch attribution.

Compared to last-touch or last-click attribution it makes it possible to move into early funnel marketing channels to acquire more customers.

How does data-driven attribution work?

Here is the example Google provides on how the data-driven attribution works:

You own a tour company in New York City, and you use conversion tracking to track when customers purchase tickets on your website. In particular, you have one conversion action to track purchases of a bike tour in Brooklyn. Customers often click a few of your ads before deciding to purchase a ticket.

Your data-driven attribution model finds that customers who click your “Bike tour New York” ad first, and then later click “Bike tour Brooklyn waterfront,” are more likely to purchase a ticket than users who only click on “Bike tour Brooklyn waterfront.” So the model redistributes credit in favor of the “Bike tour New York” ad and its associated keywords, ad groups, and campaigns.

Now, when you look at your reports, you have more complete information about which ads are most valuable to your business.

More on this here: https://support.google.com/google-ads/answer/6394265?hl=en

In this example the earlier touch-points also get conversion credits with multi-touch attribution, for example the first touch-points with Adexchange and Light Reaction.

Data requirements for Google data-driven attribution

Google has removed the old minimum data requirements. Every conversion action now qualifies for data-driven attribution, whatever its volume.

What remains is a recommendation rather than a gate: at least 200 conversions and 2,000 ad interactions in supported networks within a 30 day period. The model still runs below those numbers, but more history helps it tell real patterns from noise. If your volumes sit well under that line, the answer is to grow traffic and conversions, not to pick a different model.

The old fallback is gone too. Accounts that dropped below the threshold used to revert to the linear model. Linear, first click, time decay and position based were all retired in November 2023. Google Ads and Google Analytics 4 now offer data-driven and last click only.

Windsor.ai delivers the granular data your model needs, so it can work with less data and go to keyword level much easier.

We recommend to switch to multitouch attribution or data-driven attribution because it makes the analysis so much easier. It becomes possible to look at one KPI and make decisions based on that.

Especially with complex customer journeys and programmatic platforms it can sometimes be directly misleading to look at clicks, click-through rates etc. So it is much better to really look the contribution of the conversions and put that into perspective to the costs to get a data-driven ROAS or a data-driven CPA.

Data-Driven Attribution in Looker Studio

Marketing teams like StoryBox, which strive to understand the true performance of their marketing channels, need to look beyond last click attribution.

People make passionate arguments about first or last click attribution. When it’s my budget and company on the line, I go with the data and science every time.

Bill Macaitis, former Slack CMO

Visualising non last click attribution models in Looker Studio is possible. To jump straight into it here are two ways you can get started with non last click attribution.

Your two options

1.Connect your raw Google Analytics data and setup the Looker Studio Multichannel Attribution Dashboard Template. In case you have additional dimensions from other data sources in Looker Studio you can just blend them together with the Windsor.ai data source. The steps necessary are:

1. 1. Copy Dashboard
   2. For the New Data Source, select CREATE NEW DATA SOURCE
   3. Under Community Connectors, go to EXPLORE CONNECTORS, find Ad Data + Attribution By Windsor.ai, add the connector

2.Use your existing Google Analytics connection in Looker Studio and follow the 4 steps below:

1. 1. Copy the Looker Studio Facebook Ads E-commerce Dashboard Template
   2. To get the data for your attribution model set up, select CREATE NEW DATA SOURCE in Looker Studio
   3. Under Community Connectors, go to EXPLORE CONNECTORS, find Ad Data + Attribution By Windsor.ai, add the connector
   4. Install it and blend the data (see video below)

Data-driven attribution in digital marketing with AI & ML tools

Data-driven attribution gives the most credit to your best channels and eliminates underperforming ones. However, applying this model properly usually requires an expert data scientist and dedicated analytics resources.

Data collection of the marketing actions is a pain for marketers, as in some cases, like with offline marketing, it’s hard to evaluate a customer’s behaviour. Tying those offline touchpoints back into the same attribution framework as your online data is one of the harder parts of getting data-driven attribution right. While using different touchpoints in marketing, the use of data-driven attribution tools that help to understand when and how various marketing channels contribute to those conversions are very important for defining a marketing strategy.

Thanks to data science, and its tools like AI or machine learning, marketers can predict ROI and ROAS in their planning more accurately. Marketing managers apply AI and machine learning to unified cross-channel campaign data.

Data-driven attribution per keyword or ad content

Some of our clients like to not only use the high-level budget optimiser but go very granular with their analyses. They sometimes analyse the attributed revenue for every keyword and the customer journeys containing those keywords. This way they can model data-driven attribution per keyword and journey on top of that data.

Here is an example of what it can look like to have a data-driven attributed ROAS per keyword:

When the ROAS is calculated with multi-touch attribution it simplifies the analyses because then the keyword gets credit wherever in the customer journey the keyword has appeared so the analyses become simpler.

Those who want to drill down into the keywords on a customer journey level can of course do so also. Then it can look like this:

Then it becomes possible to search for the keywords so one can find the keywords wherever they appear in the path.

Doing data-driven attribution per keyword and journey usually provides lots of insight and data so most of our customers don’t go down to this level but some like to quite granular.

Data-driven attribution for affiliate and price comparison channels

Affiliate and price comparison channels are one of the clearest cases for data-driven attribution, because they often sit mid-journey where last-click reporting hides them. Once the attributed revenue is matched against the cost data from each programme, you can compare them on the same basis as paid search or paid social. Windsor.ai imports cost and revenue data from affiliate platforms such as AWIN and Connexity, so the model has both sides of the calculation. The Benz24 case study shows what that looked like in practice.

Windsor.ai helps clients free themselves of unprofitable campaigns and focus their marketing with data-driven customer journey insights, unifying the data behind the attribution modelling that tells you what works and what doesn’t.

Connect your data in 1 minute with a free forever plan, and read the stories of success from some of our clients.