The concept of Attribution is quite simple: identifying which events or touch-points contributed most to a conversion throughout a customer’s journey and then assign each event and touch point a certain percentage of credit for that conversion. However, attribution starts to get challenging when consumers start interacting with a brand across different channels and on different devices. This is called Cross-Device customer journey.
Consumers use a range of Internet-enabled devices to do a wide range of things and this new generation of multi-device users has given rise to the cross-device customer journey. A majority of online consumers who use multiple devices start their purchase on a smartphone and then continue on a PC or tablet. This increase in cross-device usage has brought the need for cross-device attribution into focus.
Essentially, cross-device attribution is an extension of cross-channel attribution. Cross-device attribution provides data about how a user interacts with a brand, online and offline, regardless of device. With unified data across devices, marketers are empowered with the data necessary to make marketing decisions that help to achieve marketing effectiveness.
Deterministic and Probabilistic Models
There are currently two different models for performing cross-device tracking and attribution: Probabilistic and Deterministic.
Deterministic models rely on a user being logged in to a website or app for the process to work. In order to identify users across multiple devices, deterministic matching searches through data sets and links all user profiles that belong to the same physical person together with a common identifier. Common identifiers may include:
- First and last name (if uncommon)
- Email address
- Date of birth
- Phone numbers
Applications like Facebook, Google Apps, and Twitter are able to deterministically match users quite easily, as they require users to sign in with an email address to access their services across different devices.
The data gathered using this method is more reliable, but without the guarantee that users are always logged in, the data can be limited.
Probabilistic models make assumptions about cross-device user identity based on algorithms and can offer accuracy rates of up to 90%. The key to achieving accurate probabilistic matching lies in linking together user profiles that contain the same highly specific pieces of information.
For example, if a married couple living together each had a smartphone, tablet, and a desktop, then each device would access the same IP address, have the same Wifi ID, and be at the same location. The way to probabilistically match the devices to the same users would be to look at other pieces of personal data, such as age, gender, and interests that are consistent across all devices.
Probabilistic matching uses anonymous signals like:
- Location (IP addresses)
- Conversion Type
- Device IDs
- Landing Page
- Interests and web history
Probabilistic matching may not be as accurate as deterministic matching, but it does use deterministic data sets sometime to train the algorithms to improve accuracy. This works by taking a small group of deterministic and probabilistic data sets and teaching the algorithms to make the necessary connections. Then, the newly trained algorithms are applied to data sets not containing the deterministic pieces of information, which can possibly be in the millions.
Which one is better?
Although both methods of cross-device attribution have its pros and cons, Probabilistic model is usually preferred as it can be applied to a wider audience than deterministic matching can. Another in-built advantage of probabilistic model is security. With increasing data privacy concerns, it is better and easier to collect non-personally identifiable information for mapping users. However, due to the same reasons, probabilistic matching could be less accurate.
Deterministic tracking is better suited for websites that can guarantee that a large proportion of its users will be logged in when they are using it. This is because without the login, deterministic tracking does not have the capability to accrue data. Probabilistic tracking is the wiser choice for sites with fewer login requirements and smaller consumer bases. As probabilistic tracking models are based on scaling probability rather than a record of actual consumer data, a smaller consumer base will not lead to misrepresentative reporting.