Problem
Teams in banks, retail, delivery, taxi and telecom hold location data, but often do not know which tasks it actually helps with. Staffing and opening hours are often set from each outlet's own history alone, which misses foot traffic and competitors nearby. A new district or city has no history at all. Routing and driver assignment must account for traffic, courier capacity and the last mile. For telecom, every task depends on how accurately a subscriber's position is known.
Example from the article: in a taxi project, a driver-customer matching model increased trips by up to 0.4%.
Solution in brief
The talk is a map of geo tasks by industry, with a method for each. Enrich per-outlet time series with neighbours, foot traffic and POI. Estimate new locations with look-alikes and gravity models. Use road graphs for routing and zones, and base-station handover embeddings for subscribers. Check that the task needs geo at all.
Questions this case answers
- How to estimate demand at a new location with no history from its spatial neighbours, foot traffic and points of interest
- How to plan routes and assignments under road-network and capacity constraints using a road graph reconstructed from sparse location signals
- How to staff bank branches and shops and set their opening hours using location data
- How to choose where to open a retail outlet when there is no sales history
Tools and methods
- Look-alike outlets and gravity models — placement of outlets and ad stands
- VRP / PDP solvers (VROOM, GraphHopper) — routing and pickup-and-delivery
- Road graph (OSM, OSMnx, NetworkX) — service zones and route reconstruction
- Handover embeddings — MultiSIM detection, outlier removal
- Switchback tests — validation across zones
Google or a one-time email code. No password.
