Toll Revenue Leakage: 4 Leaks AI Is Closing

What is the most expensive vehicle on your road?

The one that was never recorded. It crossed your infrastructure, used its capacity, and left no transaction. Sometimes it goes unnoticed; other times it is identified, but the revenue still remains unrecovered. That is toll revenue leakage: cumulative, persistent, and easy to underestimate.

Think of every journey as a chain of steps: detect the vehicle, identify it, count it, process it, and finally charge it. A weak link at any point allows revenue to leak away. For operators and public authorities facing rising operating costs and higher service expectations from users, this matters more than ever. The good news is that many causes of leakage are identifiable, and AI in tolling offers the industry new ways to close them.

In this third article of the AI For Tolling series, we examine four leaks, some hidden and others in plain sight, and show how AI helps close each one, with insights from Adria Ferret, New Solutions Manager at Emovis.

Leak #1: The Vehicle Your Sensors Never Saw

For years, detecting a vehicle involved combining multiple sensors and, even then, some cases went unnoticed because of certain vehicle characteristics or road conditions. Today, computer vision lets cameras resolve more cases and reduces dependence on certain roadside equipment. The result is simpler infrastructure and a lower chance of missed transactions due to undetected vehicles.

In practice, the key change is that AI is now inside the camera itself. Before, the camera only captured the image and relied on other sensors to know when a vehicle was passing. With integrated computer vision, the camera detects the vehicle, chooses the best moment to capture the image, and extracts information such as its classification or visual characteristics. This adds a very powerful detection layer that complements existing sensors and recovers passages that previously slipped through without leaving a trace.

That camera intelligence also makes it possible to design much lighter solutions. Today, we can do without a full gantry and use side-fire cameras as the primary detection sensor, with equivalent performance and a much lower implementation cost. For the operator, this means less equipment to maintain, fewer points of failure, and more passages captured with better evidence from the outset.

Leak #2: The License Plate No One Could Read

A damaged or difficult-to-read license plate can prevent a transaction from being correctly associated with a vehicle. AI-assisted image interpretation can resolve cases that conventional systems do not identify with sufficient confidence, improving identification and reducing potential revenue losses.

But identification is not enough: each vehicle must also be recorded only once and reliably. Vehicle fingerprinting and deduplication help recognize the vehicle and avoid duplicate records, improving data quality and protecting revenue.

The value of these capabilities becomes clear when each passage generates evidence from different equipment, angles, and moments. This creates more opportunities to identify the vehicle, but only if they can be linked; without intelligence, more data generates more exceptions, manual review, and cost. Emovis’ OBO incorporates AI to do this: vehicle fingerprinting recognizes that multiple detections belong to the same vehicle based on its visual characteristics, even if none can read the license plate on its own, while deduplication ensures that the passage is recorded only once.

Together, these capabilities turn partial evidence into an identified transaction and more reliable data, with each passage associated with a single record and supporting evidence. For the operator, this helps recover previously lost passages, reduce exceptions, duplicates, claims, and subsequent corrections, and protect more revenue without increasing manual review.

Leak #3: When Conditions Affect Performance

A system may perform correctly under normal conditions and yet generate more exceptions when visibility, weather, or traffic density deteriorates. AI can strengthen image interpretation and pattern recognition in these scenarios, helping to maintain more consistent performance and reduce the impact of adverse conditions on transactions.

The problem is that traditional systems rely on rules designed for specific conditions. When the light changes, it rains, traffic increases, or the capture environment changes, the number of cases that don’t fit those rules grows, and each one is revenue at risk.

Faced with this limitation, AI interprets more complex patterns and works with confidence levels. Clear cases are processed automatically, and ambiguous ones are referred for validation, rather than forcing a potentially incorrect decision. In addition, corrections made during operations help the models adapt progressively to the real conditions of each road.

AI is not infallible, but it recognizes uncertainty better, learns from exceptions, and maintains more stable performance when conditions deteriorate.

Leak #4: Manual Identification Errors

Not all vehicles are resolved automatically. When the system can’t identify one, a person reviews the image through Manual Image Review (MIR). It is essential work, but it is also human, and a misidentified vehicle is lost revenue or an incorrectly processed transaction.

Here, AI doesn’t remove people from the process; it helps them work more consistently. Emovis’ MIR application can analyze operational patterns to identify situations associated with a higher risk of error, allowing additional controls to be introduced before an incorrect identification affects the transaction.

Specifically, MIR can anticipate many errors before they affect the transaction by combining rules and AI that analyze two sources: the data provided by the roadside for each vehicle and the reviewers’ own actions. With this information, it decides in each case whether the identification can be accepted or whether it should be escalated for a second review.

In this way, effort is concentrated where there is uncertainty and additional controls are not introduced where they are not needed. Manual review works in both directions: people validate cases where AI has lower confidence, and AI helps detect potential human errors. Every error detected in time is revenue that is neither lost nor disputed.

Every Vehicle Counts

At Emovis, we see AI as a layer of intelligence that enhances the tolling value chain without replacing operational expertise.

With the right knowledge and controls, it can interpret images and data at scale, identify gaps, reduce exceptions, and protect more revenue from each journey.

Next Article