Bart: Can you explain this in more detail?
Marcel: Sure. I should start by explaining how advertisers struggle currently to evaluate the performance of walled gardens platforms like Facebook, TikTok and others. We can take Facebook as an example.
Brands basically have three options - to use last click data from GA that would typically give Facebook the least credit and the performance channels later on in the consumer journey are as a result often overestimated.
A bit more sophisticated advertisers and analysts would use click-based data-driven attribution models which typically adjust the results a bit in favor of Facebook, but this is still limited as it is only using click data.
The third option is then to rely on Facebook Attribution or what Facebook business manager reports about the performance of FB campaigns. That, on the other hand, gives unrealistically high and inflated numbers about FB contributions.
So marketers, at the end of the day, know that the truth is somewhere between DDA click based and FB self-reported results, but the range is still VERY wide.
The problem is that platforms are all claiming the same conversion and the marketer has no idea who is right? It is like a classroom where a teacher asks who in the class was responsible for cleaning the blackboard so nicely and multiple kids put up their hands. Who should the teacher praise?
Bart: Wait a minute, Marcel. I understand what you say but how you state it, it seems all to boil down to how everything is being tracked by each of the platforms. But it seems there are advertisers that found a way to overcome many of these issues. Can you explain how this is done and where your solution comes in?
Marcel: I’ll come to that in a minute, but let’s first start with a real-life example which will explain a lot about how this can be solved. Let’s have a look at how I could have bought my jeans online.
First, I saw some ads of an e-shop on my phone on Facebook and TikTok during the day. In the evening, I went directly to the relevant e-shop and after some searching on their website, I buy a pair of jeans and make a 125 EURO purchase at 8pm.
At the moment of purchase, information about my transaction is sent to all the ad platforms, the web analytics tool of the e-shop and the CRM of the e-shop. Both FB and TikTok will say that they are responsible for 100% of the conversion and report this out. But Google Analytics will say my 125 EURO purchase at 8pm comes from Direct. They all have measured a part of my journey and attribute my purchase in isolation.
Our solution, called Roivenue, works differently and works in two major steps:
First, it uses enhanced connectors, which pull conversion and impression data from all platforms. Pairing algorithms use this information to enrich incomplete customer journeys. As a result, synthetic impressions are inserted into the customer journey. Only when all journeys are enriched, it looks at the attribution of each touchpoint in the journey. No additional pixels or 3rd party cookies are needed, and it works cross-device.
The moment you are looking at complete journeys, the attribution models will perform much better and give a more accurate read on the fair contribution of each of the digital channels which avoid under- and overestimating channels.
Bart: Got it and this makes sense. So, your solution starts with clever pairing of information and only in the next step does the attribution. Can you elaborate a bit more on both steps?
Marcel: Yes, sure. Let’s start with the last step, meaning attribution and work backwards.
Attribution models can be classified in two groups. Rule-based and data driven models.
Rule-based are often simple models that assign contribution in a fixed manner. Data driven attribution, however, uses all the available data of all consumer journeys to quantify the contribution of each channel. It is making use of regression techniques, Markov chains or more sophisticated machine learning models.
Roivenue is using a very clever attribution model developed by our AI colleagues in Prague. They are utilizing recurrent neural networks to predict the likelihood of conversion and assign credit based on the contribution of each touchpoint in increasing the chances of conversion. A lot of variables are taken into account including the order, the time stamp, the length of the journey, etc.
Another strength is that the model also uses qualitative data about each website visit. GA4 (Google Analytics 4) considers every visit in the same way as it bounces within a few seconds. Roivenue recognizes the difference and assigns credit based on that. As a result, it is also able to recognize decided customers and exposes parasitic channels.
Bart: That sounds clever, but how does this help the solution to go beyond click attribution and also bring in the impact of ad impressions?
Marcel: This brings us to the smart pairing. We want these clever attribution models to work on the best available paths instead of inconsistent ones. Our solution addresses the challenge of post-view attribution, by leveraging impression-level data from walled gardens and generating synthetic impressions and adding these to the relevant journeys.
GA4, in comparison, faces limitations in capturing the true impact of such activities and therefore will always overvalue (click) channels.
Additionally, Roivenue integrates data from multiple sources including Google Analytics and impression-level data from DSPs and walled gardens.
There are various connecting variables that are used to connect the data from various sources to enrich the consumer journeys. Think about the UTM tags but also the order id and more importantly the order value, time stamp and from which campaign or ad group the conversion is coming from. All this information is partially available in the various platforms and the moment this data comes together, we stitch it where possible in a deterministic way.
Think about the jeans example: Facebook knows about a conversion and tells you they were responsible for this conversion via the ads that were served on their platform. However, we can use this info not to give them the full credit, but we can use it to insert FB impressions in the journey of this conversion.
Yes, this is possible for many advertisers as long as the set-up is done right. If the set-up is right, you can for a large part insert these impressions deterministically. The rest can be done probabilistically, especially in cases when the delivered impression data from a walled garden could link to multiple conversions.
Bart: But are you saying that you are just using the available information and don’t need additional cookies to implement this properly?
Marcel: True. We believe this is not only the right way to properly measure performance from walled gardens, but also the most future proof one as it doesn't rely on any additional third-party cookies.
It's also easy to implement and to a certain extent it puts us also further on a path of solving cross device tracking problems as the platforms like Facebook use and require persistent logins on all devices and know exactly on which devices the user see the ads.
To summarize, the main requirements are a good set up or configuration on the client side. That is why we always start with an audit of a client’s web analytics measurement to ensure the server-side tagging is done correctly, all pixels are set on the right places and the taxonomy is consistent across the various platforms.
Bart: We have so far spoken a lot about the how, but I am also interested to learn about how the results of the solution can be applied by clients. Does the solution deliver the actionable insights clients need to optimize their digital campaigns and is this really complementary to MMM and Experiments?
Marcel: Yes, it does. The moment the pairing and the attribution happens, you have the whole digital marketing funnel in view. From investments, to impressions, clicks, visits, conversions and if linked to your CRM or ERP system, also revenue and profit.
This is all fully accessible in the software that is available for clients. The possibilities to drill in the data are endless as the front-end works on the most granular consumer journey data. As such, you can filter on specific periods, campaigns, formats or platforms. The software also contains a budget optimizer which helps you to shift money to the more effective channels in your plan. The models and data are refreshed overnight, daily, so a user is always looking at the most recent data.
Bart: The software solution looks great, but is everything available in the software or is there more?
Marcel: There is more, as the real strength goes beyond the software. We believe that transparency is crucial for any type of marketing effectiveness solution. Why would anyone use outcomes where it is not completely clear how they are derived? That is why we provide clients with the opportunity to export the most granular data towards their own data lake or bi solutions. We call this “raw data export”. This opens an endless way of using the data in various other applications.
Bart: This is great. Based on what we have seen today, can we state that MTA still very clearly deserves is place next to MMM and Experiments?
Marcel: Yes, I fully agree. It has many strengths and if done right, MTA is clearly complimentary to these other applications. Leveraging multiple measurement methods through triangulation provide marketeers with both strategic and tactical insights to optimize their marketing effectiveness.