What artificial intelligence and big data change in transit planning, what they leave to planners, and how Flow puts both to work.
Most planning tools assume a person draws, models, and checks every variant. So networks get simplified, studies examine a handful of alternatives, and the passenger impact of each arrives days or weeks later from a separate model.
Agencies now produce an enormous amount of detailed operational data as a byproduct of running service: passenger counts, vehicle locations, fare records, and schedules. Optimization models can now search network designs at a scale no team could draw. Together, they turn a separate modeling task into a question a planner can answer the same morning.
Today's toolsModels leave variables out to stay tractable.
With FlowTransitFlow models far more of the real network, closer to true simulation.
Today's toolsEach variant is hand-drawn and modeled over days or weeks, limiting how many get explored.
With FlowTransitFlow proposes new networks, computing millions of options in minutes and returning the strongest for review.
Today's toolsRidership impact needs a separate travel-model run.
With FlowTransitEvery change re-scores the whole network as you make it.
Today's toolsReports are assembled by hand after the fact.
With FlowTransitAuto-generated reports explain each change with maps, scorecards, and accessibility analysis.
AI is a loose label for software that does work that used to need human judgement. In transit planning, it helps to separate three kinds.
Dashboards and reports on ridership, running times, and on-time performance. The foundation for everything else, though it recommends nothing on its own.
Machine learning trained on past operations forecasts running times, loads, and demand, including how riders respond to a change.
Optimization searches possible designs and recommends the strongest. Flow uses deep reinforcement learning trained on real and synthetic cities, run against your objectives and constraints.
Flow works in the second and third, and every score traces back to its inputs and assumptions. Timetables, vehicle blocks, and crew rosters stay in your scheduling system.
Most planning tools cope with a network's complexity by leaving variables out. Flow brings them back in: the decisions planners control, what riders experience, what operators carry, and the physical limits of the street.
Everything is scored together on real street geometry with observed speeds, with passenger choices modeled across walking, cycling, and transfers. The model gets closer to a true simulation of the network, so there are fewer surprises once service is on the road.
Even a mid-sized city has an astronomical number of possible route configurations, and a hand-drawn study examines only the few its schedule allows. The strongest design may never be among them.
Flow's optimizer searches routes and frequencies together, within the objectives and constraints you set, and returns complete candidate networks, each scored the same way and labeled by what it optimizes.
Flow re-scores the whole network as you edit: estimated travel time, ridership, coverage, and cost, including the riders who shift between routes. Effects on lines the edit never touched show up while the change is still on screen.
Costs are planning-level estimates your schedulers refine when they build vehicle and crew schedules.
Every scenario reports the assumptions it was scored under, attributes changes route by route, and generates the maps, scorecards, and equity analyses a board or committee packet needs, including the tables for a US Title VI service equity analysis. When a board member asks why Route 6 improved, the report shows the modeled rider shift behind it.
A city-wide network redesign, or a new tram or BRT line the bus network has to reorganize around.
A new neighborhood, hospital, or campus that needs service, and the routes that change to reach it.
Scheduled roadworks, a changed speed limit, a crash, or a demonstration that closes streets, with service adjusted and rolled back after.
Flow runs on data agencies already have: the GTFS feed, passenger counts, AVL history, fare records, and census and land-use data, plus the origin-to-destination patterns inferred from them. Models trained on your past year of operations forecast how each day will run, and keep updating through the day.
Where counts are thin, Flow estimates demand from land use and journey-to-work data, and each scenario notes which inputs were measured and which were estimated. Results export back out as GTFS.
Whether a fixed budget should chase ridership or guarantee coverage is a choice between legitimate goals, and it belongs to planners, elected officials, and the public. A model also cannot know a city the way its planners do: which neighborhoods depend on a route no data set would flag, what residents asked for at the last public meeting, and why past decisions were made.
That knowledge enters Flow as objectives and constraints, such as a protected corridor or a minimum service guarantee, and planners refine the candidates until the network fits their community. Flow works with the planner as a co-pilot, taking on the search and the scoring so planners can spend more of their time on the creative and strategic work of shaping a network.
More on Flow: the product overview and Flow for consultants and operators.
Start with one corridor or the service change you're already working on.