Why AI-driven fraud protection matters now

In transport, payments are part of the passenger journey, not a back-office detail. A mobile top-up on the way to the platform, a ticket bought mid-commute: these moments need to be quick, dependable and almost invisible. When they don’t work, the impact is immediate. Customers abandon checkout, complaints rise and operational teams feel the strain.

At the same time, the fraud landscape has shifted. Automated attacks are more common, scams are more convincing and tactics change quickly. In that environment, leaning on defences built for a simpler era can create gaps that only show up when it’s too late. Put simply, we must fight fire with fire.

Account takeover and passenger experience

Transport ticketing is a particularly tough environment for fraud controls because the stakes are high and the decision window is tiny. Many transport payment journeys happen at points of entry and transfer, where a slow or uncertain outcome creates friction straight away. 

Open-loop models are designed to improve convenience and security, and card networks explicitly position open-loop transit as enabling enhanced payment security and lower fraud risk at transit points of entry. But the shift to open-loop and digital-first journeys also increases the number of places fraud can show up: account creation, top-ups, stored credentials, customer service flows and post-journey adjustments.

Old-school fraud detection

Rules-based fraud detection is the traditional way of protecting payments: if a transaction matches defined conditions, block it. Rules-based fraud detection has been the default for years. Rules still have a place, especially for known patterns, but using them alone creates real drawbacks in high-volume, time-sensitive environments like transport ticketing.

• False positives:
  Rules are binary by nature and struggle with context. That often leads to false positives, where genuine customers are declined because their behaviour happens to resemble a known fraud pattern. In transport, false declines don’t just lose a sale: they can interrupt a journey and chip away at trust.
  
  

• Inefficient and hard to scale: 
  A rules-only approach means the library must keep expanding as fraud evolves. Over time, the system becomes slower and harder to maintain. It also puts a heavy burden on fraud analysts who spend their time adding, tuning and testing rules rather than focusing on higher-value investigation and strategy.

Rules can support a wider approach, but they’re rarely enough as the primary defence when fraud tactics evolve quickly.

Why machine learning suits fraud detection

• Speed 
  Fraud decisions need to happen fast. The longer a buyer’s journey takes, the less likely they are to complete checkout. In transport, that friction can be especially costly because customers are often purchasing on the move. Machine learning can evaluate large numbers of signals in real time, returning a decision in milliseconds and keeping the journey flowing.
  
  
• Scalable 
  Machine learning systems improve with larger datasets because they have more examples of genuine and fraudulent behaviour. That helps models spot differences and similarities more quickly and apply that learning to future transactions. As transport networks grow, channels expand and more ticketing journeys shift to digital-first experiences, this ability to scale without constant manual rule-writing becomes a clear advantage.
  
  
• Efficiency and cost
  Machine learning does the heavy lifting of data analysis at a speed and scale that even large teams can’t match. It can run continuously, handle repetitive tasks and only escalate edge cases where human judgement is genuinely needed. That reduces manual workload, helps teams focus on meaningful exceptions and supports a more sustainable operating model.
  
  
• Accuracy 
  A strong machine-learning model can identify risk signals before a chargeback has even happened. It can also look beyond transaction history and assess behavioural patterns that are hard to capture with rules alone. For example, it might flag suspicious activity based on how a user navigates a purchase flow, how they enter information or signals that suggest automation. Catching risk earlier helps reduce losses while avoiding unnecessary friction for genuine passengers

   

The takeaway for transport operators

The goal isn’t to add hurdles. It’s to protect revenue and trust while keeping journeys seamless. As fraud becomes faster and more sophisticated, the most resilient approach is layered: combine human expertise with systems that can learn, adapt and respond at the same pace as the threat. That’s why AI-driven fraud protection matters now.

At Transport Ticketing Global 2026, running from 17-18 March at Olympia London, John Dobson will share practical ways to apply AI and machine learning to cut fraud, protect conversion and build scalable defences for evolving digital threats. His presentation is titled ‘The Future of Payments: Why AI Fraud Protection Matters Now’.