The Science Behind Modern Device Maps

By | 2018-05-01T16:45:12+00:00 November 29th, 2017|Product|

The content consumption landscape today is a complicated one with a blurry line between the content and the device on which it is consumed. Viewers control when, where and how to watch their favorite shows across a growing number of connected devices including laptops, tablets, smartphones, smart TVs, Apple TV, Roku and more, making it difficult for marketers to understand ad exposure and impact. With this in mind, while it’s common for advertisers to develop strategies and allocate budgets by device, planning in silos creates blind spots in ad effectiveness, path to purchase, frequency impact and overall reach. For example, when measuring what drove a digital conversion, are TV ads included in the attribution strategy as well?

With a cross-platform device map, advertisers can see the true impact of their TV and digital media investments. When digital devices and televisions from one individual or household are linked together, it becomes easier to target people across screens based on behaviors, provide the right message depending on exposure and measure de-duplicated impact, proving out ROAS. However, the science behind building out an accurate device map can be challenging and complicated and is the key to creating an effective bridge between TV and digital.

There are 3 main ways of identifying devices:

  1. Deterministic: A deterministic device map links devices to individuals using login data, like the email IDs of registered users – signing into The New York Times from a tablet, smartphone and laptop using the same ID, for example. This method is limited in scale since it works only when the services being used are registered to that ID and it can’t be shared across vendors or services, creating a walled garden. It also can’t link devices that belong to the same household but don’t use an ID ― like a shared family tablet or devices without a log in, like a TV.
  2. Probabilistic: A probabilistic device map makes its matches using more freely available information, like location or IP address. This is less accurate than deterministic IDs, but far easier to scale. For this to work well, a probabilistic device map needs to use a persistent, privacy-compliant identifier.
  3. Hybrid: This ideal method combines the best of deterministic and probabilistic mapping. It has the scale of a probabilistic device map, but can fine-tune results with a deterministic truth set by adding more persistent identifiers.

Once devices are identified, the next big question is how to identify outliers. For example, which TVs truly belong to a household and which are in offices, hotels or airports? A device map needs a way to scrub out “rogue” devices while correctly identifying the primary/home set. How is this done? With continuous machine learning, algorithms track usage patterns of the TVs and their mapped digital devices, weeding out false positives and false negatives within the dataset.

Identifying missing devices
Missing devices are devices that should belong to the household but are not captured in the device map.

  • Using multiple data sources to identify digital devices, like TV-to-device pairing, cookies/device IDs and bid requests through exchanges, can ensure the device map captures all
    the digital devices in any given household.
  • Households that have additional TV sets that are not Smart TVs are captured by enhancing the Smart TV dataset with set-top-box data.

Identifying False Positives
False positives are devices that have wrongly been included in the device map

  • If a TV is connected to a higher-than-average number of digital devices, and those devices are constantly changing, the environment could be identified as an airport, coffee shop or other public place.
  • If a higher-than-average number of digital devices are connected to a TV, but it’s almost always the same set of devices, the environment could be identified as an office.
  • If the digital devices mapped to a TV change at a higher-than-normal frequency, the environment could be identified as a hotel where users connect to the IP for a few days at a time.

Identifying False Negatives
False negatives within a device map indicate TVs that are wrongly classified as non-household. For example:

  • An IP change for a digital device would technically indicate that it no longer belongs to that household. But, if the IP change is consistent with that of the TV, it could indicate an IP refresh of the entire household. In such cases the map needs to update and continue tracking all devices as part of the same household.
  • All digital devices using the same IP in a household are plotted as part of that household’s device map. But, if a digital device spends just a few hours in that environment, it
    could be a visiting guest.
  • If a TV has an accepted average number of devices mapped to it, it would fall into the definition of a household. However, there may be smaller environments (such as a
    startup office) that would fit that description but should not be considered a household. In such cases, measurements of average time of day and duration of the devices active on that IP would be able to capture the non-household nature of these devices.

A high-fidelity device map for the modern marketer needs to smash the silo and enable marketers to create an integrated experience for the consumer at home and on the go. Device fragmentation is here to stay and marketers need to not only embrace that fragmentation but turn it into an opportunity to engage with consumers on their gadget of choice. A good device map is the first step in enabling marketers to make a holistic cross-screen impact. Read more about our device map technology and do not miss our Device Map Technical Whitepaper to learn more about the science of bridging the devices

Contact Us
Thank you! We'll be in touch with you shortly to assist you.