Optimization
Judge every road and rail project by what it does to the whole network.
Appraising transport projects one at a time is a mistake, because each one can change how people and freight move across the whole network. extendfuture would test every road, rail line and bus change on a model of the region's full network before public money is committed. Those choices last for decades. The WHO estimates that road crashes kill about 1.16 million people a year and cost most countries 3% of their GDP. The IPCC found that transport produced 23% of global energy-related carbon dioxide in 2019.
How can AI help plan roads and railways?
AI helps only when it runs on a calibrated model of the whole network, meaning one adjusted until it matches measured traffic. We would combine OpenStreetMap, public timetables, traffic counts and census travel data in a simulation of how people and goods move. Optimization software then searches for the plan that meets the targets at the lowest cost. AI agents, software that sets up runs and drafts summaries, compare the scenarios. Planners compare travel time, crashes, emissions and cost, and an official signs off every decision.
Judge a project by its effect on the whole network
Transport projects are often appraised one at a time, such as a single bypass, rail upgrade or bus corridor, yet their effects spread across the network. A model of the whole network can test whether a new road draws traffic from other routes, whether a faster train changes where people live, and whether a bus change cuts a neighborhood off from jobs.
UN-Habitat data in the UN's 2026 SDG report show that convenient access to public transport rose from 51.5% to 60.4% of urban residents in 350 cities between 2020 and 2025. The report says data and digital mobility systems will be critical to improve service planning.
Start with open data, not a data-sharing agreement
A public body should see results before it shares any records of its own. The model would start from open data and cover every mode, including walking and cycling.
- Road, rail and path networks from OpenStreetMap
- Bus and rail timetables in GTFS, the open standard for public transport schedules
- Census travel-to-work tables and population data from the national statistics office
- Traffic counts, ticket sales and crash records held by the road authority or operator
Trust the model only after a back-test
A model should reproduce last year's traffic before anyone uses it to plan next year's. We would build the network, estimate who travels where, when and by which mode, and calibrate it against traffic counts and ticket data. Then we would back-test it on a period it did not see, and the report would state where it misses and by how much. Every assumption, such as population growth or fuel prices, sits in one file that reviewers can read and change.
Give planners options with evidence, not one answer
A single recommended answer hides the trade-offs that officials have to weigh. Planners would get a short list of options, with the evidence for each.
- Travel time, crowding and access to jobs for each option, mapped by neighborhood
- The expected change in crashes and emissions, with the method and its limits stated
- Capital and running costs side by side
- A plain-language summary drafted by an agent and checked by an analyst
- The model, data and code, so staff can rerun any scenario
Show who gains and who loses
An average saving in travel time can hide a district that loses its bus service. So each option would be checked for who gains and who loses, by area and income group. Data stays in the public body's own cloud account where possible. Every run is logged with its inputs and model version. Agents propose and compare options, and named officials decide.
Build on open tools that anyone can check
Public money should buy models that the public can inspect and rerun. Each tool below is open source or open data.
- OpenStreetMap, the map of roads, railways and paths maintained by a worldwide community of mappers
- SUMO, a traffic simulator that models each road vehicle, bus and pedestrian, with tools for emission calculation
- MATSim, a framework for large-scale agent-based transport simulation, in which each traveler is simulated as an individual
- ActivitySim, travel demand software led by a consortium of metropolitan planning organizations and other transport agencies
- OpenTripPlanner, a multimodal trip planner, for measuring access to jobs and services by public transport
- OR-Tools, Google's open-source optimization suite, for timetables and fleet scheduling
Does this replace our existing transport model?
No. The open-source model can run beside an existing one, check its results or answer questions it was not built for, such as hourly bus operations. Where the two disagree, the report shows where and why.
Can it test road pricing or congestion charges?
Yes, as a scenario. The model would show how a charge may change routes, traffic and travel times by area, and who pays most. Whether to charge, and how much, stays a decision for elected officials.
How do you check that the model is right?
We would compare it with traffic counts and ticket data, first to calibrate it and then on a period it has not seen. The report states the error on each main road and line.
Do you need data from people's phones?
No. A useful first model runs on open data, census tables and traffic counts. If a public body already holds aggregated mobile or ticketing data, the model can use anonymous totals by area and hour. It never needs the trips of a named person.
Does this cover freight and rail?
Yes. Rail lines come from OpenStreetMap and operator timetables, and freight vehicles share the simulated roads. Freight routing and fuel use have their own page.
Talk to the people who build.
One call. An honest read on what AI can do for this, and the number it has to beat.