Start a Graph-Neural-Network Routing Algorithm Company

1. 1

   Understand routing as an AI product, not a utility

   The naive view is that routing was solved by classic shortest-path algorithms, but modern routing and ETA prediction is a machine-learning problem: a graph neural network learns from real traffic patterns to predict travel times far better. The fact that Google runs a GNN engine (built with DeepMind) under both Maps and Waze, tuned to give different routes, proves routing is a product with room to differentiate. Frame your business around that reality.

2. 2

   Get the graph and traffic data

   A routing engine learns from a road-network graph plus historical and live traffic and travel-time data. Source the network from OpenStreetMap and traffic-and-speed data from probe feeds, partners, or your customers' own fleets. The data is as much of the moat as the model, so plan how you will access it lawfully and continuously.

3. 3

   Build and benchmark the GNN engine

   Train a graph neural network to predict travel times and optimal routes, and measure it rigorously against the baselines your buyers use today (classic routing plus their current ETA method). Publish the improvement in ETA accuracy and route quality. Buyers will not switch for a claim; they switch for a measured, reproducible gain over what they already run.

4. 4

   Package it as an API or embeddable SDK

   Your customers want to call routing and ETA, not run ML research, so expose the engine as a clean API and, where needed, an on-premise or edge SDK. Make integration fast and the interface familiar to teams used to existing routing APIs. The easier you are to drop in beside or in place of an incumbent, the shorter your sales cycle.

5. 5

   Target logistics, delivery, and mobility platforms

   Your buyers are the many delivery, logistics, ride-hailing, and mobility companies whose margins depend on better routes and accurate ETAs but who cannot fund GNN research in-house. These are technical, ROI-driven buyers: a small percentage improvement in on-time delivery or driver efficiency is real money. Quantify that ROI in their terms. This is distinct from geospatial route-expansion consulting, which advises on where to expand rather than supplying the engine.

6. 6

   Price on value and volume

   License by API volume, by vehicle or seat, or as a platform fee tied to the scale of routing you power, so pricing tracks the value delivered. Offer a trial that lets a customer benchmark you on their own data. Because the gain is measurable, value-based pricing is defensible and the trial is your strongest closing tool.

Do not take our word for the money. Put your own numbers in and see what this idea has to earn before it works. Free, instant, no account needed.