AI Traffic Management: Cutting Travel Time and Emissions
Adaptive signals cut travel time by up to 25% and vehicle emissions by up to 15%. How the mechanism works, what deployment needs, and how to measure it.
Adaptive traffic signals are one of the few urban AI applications with a track record long enough to judge honestly. Cities running adaptive signal control have reported travel-time reductions of up to 25 percent on treated corridors, and McKinsey estimates that AI-driven traffic optimisation can cut urban vehicle emissions by as much as 15 percent in dense metropolitan areas. Those two numbers are linked: a vehicle that spends less time idling at a red light burns less fuel.
This article looks at how that mechanism actually works, what it takes to deploy it, and how an organisation should measure whether it worked — using Karaj, in Iran's Alborz province, as a working example of a dense commuter city.
What adaptive signal control actually does
A conventional traffic signal runs a fixed timing plan. It gives the same green duration at 7 a.m. and 11 a.m., whether forty vehicles are waiting or four. The plan is set once, often years ago, and drifts further from reality as travel patterns change.
An adaptive system replaces the fixed plan with a feedback loop:
- Sensing. Cameras or inductive loops count vehicles per approach and estimate queue length in real time.
- Prediction. A model projects how the queue will develop over the next few minutes based on current arrivals and historical patterns for that time and day.
- Adjustment. Green time is reallocated between approaches, and adjacent intersections are coordinated so a platoon of vehicles can clear several signals without stopping.
The gain comes almost entirely from that third step. Isolated smart intersections help little; coordinated corridors are where the measurable improvement lives.
Why the emissions effect is larger than it looks
Vehicle emissions are not evenly distributed across a trip. A large share is produced during acceleration from a standstill. Every avoided stop therefore removes a disproportionate amount of pollutant relative to the seconds of travel time it saves.
This is why traffic optimisation appears in air-quality strategies rather than only in mobility strategies. For a commuter city — and Karaj, sitting on the main corridor into Tehran, is a clear example — a large share of daily vehicle movement is concentrated into a few hours on a few arterial routes. That concentration is what makes corridor-level intervention worthwhile: a small number of treated intersections can influence a large share of total delay.
What deployment actually requires
Existing infrastructure counts for more than new hardware
Most cities already have traffic cameras installed for monitoring or enforcement. Computer vision can often derive vehicle counts and queue lengths from those existing feeds, which removes the largest line item from the budget. The first question in any feasibility study should be what is already installed, not what needs buying.
Pick one corridor, not the whole network
The most common failure pattern is scope. A project that tries to cover an entire city at once takes years to show anything and rarely survives a budget cycle. A corridor of six to ten consecutive intersections is enough to demonstrate the coordination effect and small enough to deliver in months.
Establish the baseline before touching anything
Without a measured before-state, results are unprovable. Two to four weeks of travel-time data across the corridor at consistent times of day is the minimum. This is unglamorous and routinely skipped, and skipping it is why many pilots end in disagreement about whether they worked.
Define the success metric in advance
Average corridor travel time during peak hours is the clearest single measure. Secondary metrics — number of stops per trip, queue length at the worst approach — help explain the result. Effects on travel time typically become measurable between the second and fourth week after activation.
Build the internal capability
The step most often omitted, and the one that most determines whether a system is still running in three years. A platform nobody inside the organisation can read or tune degrades quietly once the implementation contract ends. A short AI training course built around the specific deployment, delivered to the transport and IT staff who will operate it, is what keeps the system in use. An AI consultant working alongside the internal team through the first months of operation serves the same purpose.
Frequently asked questions
How much does an adaptive signal pilot cost?
It depends almost entirely on whether usable camera infrastructure already exists. Where it does, the cost is dominated by software, integration and calibration rather than hardware. A corridor pilot is a fundamentally different order of expenditure from a citywide programme, which is the main argument for starting with one.
Does this replace road construction?
No, and presenting it that way sets up disappointment. Signal optimisation extracts more capacity from existing infrastructure. It complements construction programmes; it does not substitute for them where physical capacity is genuinely exhausted.
How long before results appear?
Data collection and baselining take a few weeks, model calibration a few more. Measurable change in corridor travel time generally appears between the second and fourth week after the system goes live.
Where this leads
The infrastructure built for an adaptive signal project outlives the project. Continuous, structured data on how a city actually moves becomes the input for every subsequent transport decision — where to place an overpass, how to redesign bus routes, which junction to prioritise next. That accumulated data can eventually feed a broader city model, which we cover in Urban digital twins for city decision-making.
The same narrow-scope logic applies across infrastructure; for a water-sector example see AI and non-revenue water reduction. And the sharp fall in AI operating costs over the past year has improved the economics of all of these projects — the details are in Recent AI developments and business automation.
Related services from Qatreh
Qatreh is based at the Alborz Science and Technology Park in Karaj and works with organisations across:
- Computer vision and robotics — vehicle counting, traffic analysis, image processing and automated monitoring systems
- AI consulting for business — reviewing processes and identifying which ones return the most from AI
- AI training courses and AI consulting for education — building capability inside organisations and educational institutions
- AI automation and custom enterprise chatbots
- Data science and predictive modelling
If you work in urban management and want to know which corridor would return the most from a project like this, talk to us.