secret sauce edited

51Degrees' Not So Secret Sauce


2/23/2018 2:00 PM

DeviceAtlas WURFL ScientiaMobile Opinion

Comparing 51Degrees to ScientiaMobile's WURFL and DeviceAtlas

When 51Degrees arrived in 2011 we changed device detection with our innovative open source business model. The software, source code and intellectual property is freely licenced. A flexible subscription model is used to provide data updates.

Designed in this decade for the next, 51Degrees are the:

  • First to use vendor only data sources with no reliance on historic WURFL data or intellectual property.
  • First to add support for native applications alongside web.
  • Only solution which continues to innovate, filing new patents in August 2017 and running at more than 1,000,000 detections per second on a $30 Raspberry Pi computer.
  • First to track over 900,000 permutations (referred to as Device Combinations) and forecasts over 1,000,000 by early 2018.
device combinations
Figure 1 Volume of the “cube” relates to device combinations a key measure of a solution’s breadth. As one dimension increases the number of device combinations and therefore associated signatures increases non linearly.

We do Device Detection better than anyone else. We’ve driven all the innovation in the Device Detection space. See our Industry firsts here

Competitors were designed when Nokia and Blackberry ruled the mobile industry and simple computing techniques could be used to solve the device detection problem. The permutations of operating system, web browser and device were comparatively small. Such techniques are no longer optimal and have significant deficiencies.

The computing technique Patricia Trie (or tree) first described in 1968 is at the core of the DeviceAtlas solution. Afilias, DeviceAtlas’ parent company, were granted a contested patent for the technique in Europe during 2017 after 8 years of examination. Patricia Trie notoriously overfits and isn’t well suited to device detection where machine learning rather than basic retrieval is required.

patricia trie overfit model
Figure 2 The green line represents an overfitted model and the black line represents a regularised model. While the green line best follows the training data, it is too dependent on it and it is likely to have a higher error rate on new unseen data, compared to the black line. (

That’s why 51Degrees don’t use Patricia trie in this manner. DeviceAtlas will appear to be quick and accurate when evaluated using web traffic known when DeviceAtlas generated the data set, but will perform poorly when evaluated with data that was unknown at the time. When combined with DeviceAtlas’ one size fits all mantra it’s only worth considering for a very narrow set of use cases. More information on the technique and the intellectual property background is available here.

ScientiaMobile’s WURFL uses comparatively slow regular expressions and embedded data to evaluate a hierarchy of limited device combinations. The technique works well with unseen data but is too slow and lacks data breadth and flexibility for many use cases. A key difference is that with 51Degrees you do not need to reinstall for any API upgrades. Also they refer to their 'extensibility' which simply indicates clients can maintain the solution themselves. Who would want to do that?


We’re naturally biased and proud of our intellectual property. But don’t just take our word for it. Evaluate us alongside DeviceAtlas and ScientiaMobile to see for yourselves. Find out how to tune 51Degrees for use cases as diverse as Adtech, eCommerce, Analysis, User Interface optimisation and more. We offer open source tools to help, and data migration guides for those coming from competitors. Contact us to find out more.

Supporting Information: