Waze: how the navigation app works
How Waze decides your route
I have asked myself many times how Waze works so well. That curiosity led me to study the subject and write this post.
This article is not based on internal information or official Waze documents, because the company does not disclose the full details of its algorithm.
The content is a synthesis built from public sources, observations of how the app behaves, and technical references tied to its ecosystem, including graph search, the A* algorithm, and machine learning models used to estimate arrival time.
If you have ever thought, “how does Waze know this path is bad?”, you are not alone.
Behind the map and the voice saying “turn right,” there is a system that combines map data, live traffic, and real driving behavior.
Waze is more cloud than app
The app on your phone acts as a messenger: it sends your GPS position, your preferences, and your route request to cloud servers, then receives the path, the ETA, and the navigation instructions.
And that does not happen only once. During the trip, Waze keeps updating your position, while servers can recalculate the route and ETA as traffic changes.
Where its information comes from
The base map is maintained by a volunteer editor community and enriched by official and partner sources. That is what makes the map detailed, with road directions, restrictions, limits, and points of interest.
Drivers also feed the system constantly with automatic data, such as GPS and speed, and with manual reports about accidents, road work, traffic jams, closures, cameras, and more.
Practical translation: Waze learns with the city in motion and with the city standing still.
How Waze “thinks” about routes
Behind the scenes, Waze sees streets as a network of road segments and intersections, and each segment has its own rules, such as road type, direction, tolls, and restrictions.
When calculating a route, it uses a graph search algorithm, usually explained through A*, to find the path with the lowest cost, where the main cost is travel time, not simply distance.
After finding possible paths, it does not return only one. It can generate several candidate routes and then rank the best options.
This ranking usually respects restrictions and your preferences first, such as “avoid tolls,” and only then compares which option gets you there faster.
Why ETA changes so often
Waze estimates the speed of each road segment using recent data, such as moving averages from the last few minutes, and historical models that know how traffic usually behaves by hour and day type.
To predict what lies ahead, it blends real-time information and historical patterns, giving more weight to the signal that best matches the current situation.
ETA can also be predicted with machine learning models trained on large sets of real trajectories, considering trip context and route characteristics.
During the trip, since the app continues sending updates, the server can recalculate route and ETA and suggest detours when incidents or congestion appear ahead.
What you control and why two people get different routes
Settings such as avoiding tolls, ferries, highways, long dirt roads, and difficult intersections work as filters and penalties during route calculation.
Waze also considers preferred and natural routes, which are the options you usually accept, so recommendations can be personalized beyond pure speed.
As a result, two people leaving from the same place at the same time can get different routes, and both can make sense within each person preferences.
When things go wrong
If you lose connection, Waze may fall back to a local mode using cached maps and without dynamic traffic or closure data, which reduces the quality of the estimate significantly.
Even with internet, ETA can still be wrong because of sudden traffic changes, outdated maps, unexpected stops and detours, weather, and local driving patterns.
The part nobody outside the company knows
The general principles are understandable, but the exact weights, limits, and fine rules of Waze are proprietary, so any explanation should be treated as a solid mental model, not as the full recipe.
The next time Waze suggests a strange detour, think of it this way: it is not improvising for no reason. It is comparing possibilities based on the map, live signals, and historical behavior for that time of day, while constantly updating everything as you drive.
👇 Enjoyed this content?
Share it and keep following my publications.
Take the opportunity to read more articles!
