Last month, we introduced Full Reach, our new targeting solution enabling publishers to deliver an audience on every impression by relying both on ID-based and ID-less predictions. In this blog post, you can learn more about what an ‘ID-based prediction’ is and how we can predict audiences based on first-party data.
Full Reach is our new targeting solution delivering a high-quality segment on every impression.
Instead of relying on only one predictive model, Full Reach is based on a hybrid model which uses several different methodologies (both ID-based and ID-less) and chooses between these in real-time.
While Full Reach uses the ID-less model when just a single impression is available, the ID-based model operates where historic data in connection to a device identifier is available, hence it predicts based on multiple impressions.
Leaving third-party cookies behind
The inputs the ID-based model receives from a page view are the same we see for the ID-less model:
- Content (e.g. sports site)
- Device (e.g. mobile)
- Web browser (e.g. Chrome)
- Operating system (e.g. iOS)
- Geolocation (e.g. city)
- Time (e.g. 12.30 pm)
The difference is that the ID-based model is able to connect the inputs to a device identifier and therefore, it can discover more about the user by looking at historical data before making a prediction. Please notice that a device identifier is not necessarily a cookie, but can be many different things such as:
- Login profiles
- Domain-specific UIDs
- Device IDs
- CRM-hashed emails
Just one input from a page view can already tell us a lot about the user. With the ID-based model, Full Reach is able to recognise a user and attribute multiple inputs to her by relying on more data points and thus predict an impression. Full Reach is able to make this prediction by translating the different data points into sociodemographic information thanks to our algorithm based on the AudienceProject panel.
Let’s make an example and say we see a user who logged in and visited a web page at 9 am in the morning. With the ID-based model, our prediction won’t be based on just this one input, but on historical data connected to a device identifier.
In our example, we may observe a user visiting a web page at 9 am and from a very central location of the city, let’s say from the central station. If we then see that the same user is logging in later in the evening from a wealthy suburban area, and this pattern repeats every working day, we may draw the conclusion that she is working in the city during the day and coming back home after work. Therefore, she is probably employed and has a rather high income.
Same technology as used in self-driving cars
The beginning of the 21th century has witnessed breakthroughs in many different fields, ranging from facial recognition and photo restoration to fraud detection and language translation. These developments are largely driven by variations of deep learning algorithms, due to the versatility of the technology. The autopilot system of Tesla, the AlphaZero AI of Deepmind and Google’s language translation services are examples of solutions based on this technology.
AudienceProject brings the benefits of this technology to the field of targeting with the ID-based model of the hybrid solution*. The technology scales incredibly well with large amounts of complex data, which is why it is an ideal choice for leveraging our panel data to increase targeting precision.
Targeting is a hard problem to solve, due to the complexity of human behaviour and the vast amount of content available online. It is possible to develop solutions based on simpler technology and less data, but it compromises the quality of the end product. Obtaining high-quality insights from behavioural data requires both data quantity and technological sophistication, which is why AudienceProject is in a unique position to deliver high-precision targeting data.
How the Full Reach hybrid solution works
As mentioned above, the ID-based and the ID-less models work separately, but how do they coexist?
Full Reach follows a waterfall logic where the first thing it considers when predicting an impression is whether or not the user is associated with a device ID. Depending on this, the system knows whether to use the ID-based or ID-less model.
The explanation is fairly easy and logical and can be exemplified with these three different cases:
- If a user navigates a web page and she has no device ID, and all Full Reach collects is a page view, the solution will use the ID-less model to predict the impression.
- If a user navigates a web page and Full Reach does see an ID, but it is the first time this ID is encountered, the solution will use the ID-less model to predict the impression.
- Finally, if the user navigates a web page and Full Reach is able to see her ID and recognise her, the solution will use the ID-based model to predict the impression.
Get ready for the new digital scenario
Implement our Full Reach solution today to boost your share of addressable users and let media buyers know that you are able to offer accurate audience targeting for all impressions and video views.
And most importantly, make sure that you are future-proofed for what is yet to come, no matter what Google says.
Reach out to firstname.lastname@example.org or your usual contact at AudienceProject and learn how to get started targeting your cookieless audience.
Full Reach is available for publishers in the UK, Germany, France, Denmark, Sweden, Norway and Finland.
*For the more tech-savvy reader, the algorithm is based on a deep convolutional neural network with multiple parallel convolutional layers trained with MXnet.