How does the driving model work?

Overviewanchor icon

  • We have built our own proprietary model for predicting and calculating drive times.
  • This is built around peak and off-peak hours, rather than real-time traffic data, in order to provide reliable, consistent results.
    • Peak hours - 07:00-10:00 and 16:00-19:00
  • We take into account the time needed to walk to and from the road and the actual address, providing a true human routing system.

Full door-to-door routing

  • The model itself takes a range of data sets such as OpenStreetMap and country-specific data, and then constructs speed profiles based around road types, the surrounding area, and features of the road itself. 
  • These speed profiles are then benchmarked on test maps, fine-tuned, and then set live.
  • Once live, the models are regularly re-tested and updated.


Process in detailanchor icon

  1. The map is broken down into urban and rural areas at a detailed local level, using data such as municipality limit shapes and speed limits
  2. The final areas are: Rural, Village, Town, Urban, City, Extra Urban, Dense Urban
  3. All roads are tagged with a road type, e.g residential, tertiary, trunk, motorway, etc.
  4. A library of representative trip samples is built
  5. Statistical modelling techniques are applied to build speed profiles for all combinations of road types and area, including the density of different connecting road types
  6. A test map is built 
  7. The speed profiles are applied to real-world routes, taking into account map features such as traffic lights and zebra crossings, and peak and off-peak times 
  8. Where appropriate, driving routes are book-ended with walking to represent how people actually travel, rather than simply ‘snapping’ to or from the nearest road 
  9. Drive time predictions are tested extensively against a number of real-world routes and actual journeys
  10. The results of these tests are used to fine-tune the algorithm and speed profiles
  11. The model goes live, and is then regularly tested and benchmarked, and updated where appropriate