Philip Wright, Group Data Strategy Director at MediaCom North, explains how embracing modelling can help to build out more accurate and trusted results.
Changes in browser technology, from a privacy perspective, are creating a changing measurement landscape. These shifts make the measurement of activity, something that is already broken, harder to do without a rethink.
To measure activity at the moment, we need to bring together multiple data sources. Each of these has differing levels of fidelity. Managing these various sources is difficult to do and can lead to people looking to a single source for all their reporting.
However, drawing data from a single platform doesn’t always lead to measurement results that describe what you want or expect and can be difficult to validate, particularly if you measure activity in a walled garden.
Donning my rose-tinted glasses for a moment, I’d love a single, detailed, complete picture of what impact our digital activity has on a customer’s journey. In reality, this doesn’t exist.
Accuracy vs precision
Precision describes how closely calibrated a series of results are. With digital media, we can often see when something happened down to the second. This precision leads to confidence because granular data feels more ‘correct’ than aggregated data at a lower level of fidelity.
However, precise data might be describing only part of what you need to see, albeit at an alluringly high level of detail.
Accuracy is our ability to measure what is happening. It is entirely possible to have highly precise but inaccurate measurement – especially if we are dealing with incomplete data or different methods or platforms for reporting.
Building an accurate picture that describes everything is more important than having an overly precise one that covers only parts of the customer journey.
With many tracking technologies online based on cookies and device IDs, the increases in privacy protection in browsers impact the ability to build up an accurate picture that links together distinct, granular data sets.
You’ve likely spent many hours trying to square the circle of “Why don’t the numbers from system x match the numbers from system y?”
More often than not, this is because competing measurement methods do not always align. Web analytics is not ad server data, which is not internal BI data, but many businesses’ measurement frameworks are a complex combination of these different approaches.
Trying to make this work is difficult enough when you have complete control over the dataset(s) you are trying to use. It is a challenge accentuated by privacy-compliant tracking prevention.
An increase in acceptance of modelling
The impending ‘cookiepocalypse’ is finally retiring a mechanism for tracking that has stretched a technology far beyond the original intention. To adjust to this, we need to be clear on where we are now, and where we have to move to.
With large scale blocking of third-party cookies and existing limits on the setting of cookies by scripts, we will see more holes in our data.
It will be harder to join up customer journeys where there is a separation of signals. Without that ability to build a complete picture based on directly captured data, we will have to rely on modelled data to fill in the gaps.
Use of modelling has been the case for a while, with different platforms initially adding modelling to fill in the gaps left by the blocking of trackers through technology such as Apple’s ITP and Firefox’s ETP.
It is possible to have two platforms from the same vendor where the numbers don’t match. They’ll be close, but you’ll lose sleep needlessly trying to line them up. This is often down to different applications and thresholds of modelled data. We see this difference between platforms aimed at measurement compared to those aimed at optimisation.
We’re seeing new approaches, like Google’s recently announced Consent Mode, that give more nuanced alignment of data advertising and analytics tracking technologies with the consent settings of users. Using anonymisation and aggregation to report where a user withholds consent, Google is looking to strike a balance between accurate data capture and preserving user privacy. This balance is likely to become the new normal.
Relying on your best source of truth
Whether using numbers for reporting or optimisation, it is essential to remember that in-platform numbers for digital media channels or web analytics are not reality. Your first-party data, such as transaction data, is your business’s reality.
Your internal data is your best view of the world. You already hold data on the number of sales made, for how much, and who to.
Bridging the gap between your data and the platform data you need to report and optimise is essential. To maintain privacy, we see increasing use of cleanroom technologies, such as InfoSum and Google Ads Data Hub, to build connections and enrich across data sets without sharing personal or sensitive data between you and suppliers or publishers.
Using your first-party data allows you to build out advanced analytics around your customers and what they do.
There are longstanding approaches for this, using methods such as marketing mix modelling (MMM), that work to describe correlations between media activity and effect.
The industry has used digital measurement to weave detail into that picture where signals and channels overlap – whether through in-channel reporting or the use of re-attribution.
Increasingly we’ll use algorithms and machine learning to take the digital data we have and use it to understand things in a way that is closer to the scope of MMM. This approach will be healthiest when we link models to real-world data you already hold.
The privacy-compliant measurement of media activity or engagement with your digital footprint will need to accommodate a reduction in concrete precision. But embracing, developing, and rolling out modelled approaches that are rooted in your data will allow you to build a view that is more accurate, more grounded, and, ultimately, more trusted.
Philip Wright is Group Data Strategy Director at MediaCom North.
