The mobility sector has enthusiastically adopted AI as a narrative. It has been slower to adopt it as a discipline. Here is what that distinction means for the industry – and for the passengers it is supposed to serve.
AI has become the dominant narrative in mobility. Whether you are at a transport conference, reading an industry publication, or evaluating a new platform, it is almost impossible to avoid. And yet, if you ask the average commuter whether artificial intelligence has meaningfully improved their journey, the answer is likely no.
This is not because AI is without substance in mobility. There are genuine applications already delivering measurable value. But the gap between what is being claimed and what is actually working has grown wide enough to warrant some honest scrutiny. A December 2025 report by the MIT Mobility Initiative and Kearney Advanced Mobility Institute – drawing on input from 55 organisations including Google, Uber Freight, Deutsche Bahn and Lyft – put it plainly: “most AI deployments remain isolated pilots that haven’t achieved scale, and the gap between AI’s promise and its execution is widening.” The sector does not serve itself, or its passengers, by confusing aspiration with capability.
What AI Genuinely Does Well Today
The most productive place to start is not with the hype but with the evidence. There are several areas where AI is already delivering meaningful, measurable improvements – and they tend to share a common characteristic: they are grounded in real data, deployed in real networks, and evaluated against real outcomes.
Traffic signal optimisation is one of the more mature applications. Pittsburgh’s Surtrac AI system, which dynamically adjusts signal timings based on real-time traffic rather than fixed schedules, reduced travel times by 25%, cut braking by 30% and reduced idling by over 40% at the intersections where it was installed. TRL’s SCOOT adaptive signal system – deployed in more than 350 cities globally – recently launched an AI-predictive version capable of forecasting congestion up to 30 minutes in advance, with journey-time reductions of up to 15% over current-generation systems. These are infrastructure-level improvements that rarely make headlines. They are also delivering measurable results.
Demand-responsive transit is another area where AI-based dispatch and routing can make a genuine difference. Services with variable, complex demand patterns – particularly community transport for older adults, rural areas and people with disabilities – are poorly served by fixed timetables. AI optimisation of routing and dispatch in these contexts can improve both vehicle utilisation and coverage area.
Predictive maintenance, while less visible to passengers, has real operational impact. Machine learning systems analysing sensor data to anticipate component failures before they occur can reduce unplanned service cancellations – which have a direct effect on passenger trust and reliability metrics. The specifics vary considerably by operator and network, and the field is still maturing, but the underlying principle is well-established and the early evidence is credible.
Multimodal journey planning and disruption intelligence is perhaps the most passenger-facing area where AI is adding genuine value – and the one closest to SkedGo’s own work. Traditional journey planners are static: they calculate the best route at the moment of query and leave passengers to improvise when conditions change.
SkedGo, said:AI-grounded journey planning can do something meaningfully different – understanding not just the network state but the individual passenger's context, accessibility requirements, and live situation mid-journey.
The result is a shift from informing passengers that a service is disrupted to actually helping them navigate around it in real time, accounting for who they are and where they are in their journey. We have written about this in more depth in our piece on the next frontier in journey disruption, but the principle is worth naming here: this is an area where AI is already useful, provided it is grounded in verified, live transport data rather than general-purpose language models.
What these applications share is that they are specific and measurable. They do not require the industry to wait for breakthrough technologies. They are working now, in real networks, with existing infrastructure.
The LLM Problem the Industry Needs to Confront
The launch of large language models into the mainstream in 2023 triggered a wave of AI feature announcements across the mobility sector. Many of these amounted to adding a conversational interface on top of an existing platform and presenting it as intelligent mobility assistance. This warrants direct scrutiny.
General-purpose LLMs are trained on static data and generate probabilistically plausible responses. They do not have access to live timetables, real-time service disruptions, or the specific context of an individual journey in progress. A passenger who receives a confidently-stated but factually incorrect journey suggestion has been failed by the system. In transport, where accuracy has direct consequences for people’s ability to reach work, healthcare appointments or essential services, this is not a minor product flaw.
The architectural distinction that actually matters is between AI that operates on top of verified, live, structured transport data – and AI that generates text based on patterns in training data. The former can be genuinely useful. The latter introduces risk into a safety-relevant context.
This is not an argument against AI in passenger-facing applications. It is an argument for being precise about what the AI is actually doing, and what data it is grounded in. The Deloitte 2024 Gen AI in Transportation survey found that just one in five transport companies surveyed had matured beyond pilots to broad AI implementations. That caution is not necessarily a failure – in many cases, it reflects appropriate rigour about what is actually ready to deploy.
The Data Infrastructure Problem
Beneath a significant proportion of AI shortcomings in mobility lies a data problem that the industry has been slow to address publicly.
The Intelligent Transportation Society of America’s report on AI in transportation noted that “the current gap between AI research and implementation is still vast,” in part because many AI algorithms are developed “without considering how a human is going to be able to understand and use them” – but also because they are built in conditions that do not reflect the complexity of live transport data. Real-world transport data is inconsistent across providers, delayed in its feeds, and incomplete in its coverage of multimodal options, accessibility information and real-time disruptions.
AI systems can perform well in controlled demonstrations with clean data. Deployments in live networks encounter a different reality. This does not make the challenges insurmountable, but it does mean that the foundational work of building reliable data infrastructure – connecting providers, managing quality, maintaining real-time feeds – is a precondition for effective AI, not an afterthought. The MIT/Kearney report was explicit on this point: without shared data infrastructure and interoperable standards, regions risk “fragmenting into competing and potentially incompatible AI futures.”
Execution Over Invention
The framing the MIT/Kearney report put forward at CoMotion GLOBAL is worth borrowing directly: the challenge in AI mobility is no longer primarily one of invention. It is one of execution.
The models exist. The processing power is available. The theoretical applications are well-understood. What is harder – and what the sector has underinvested in relative to the AI narrative – is the domain expertise required to deploy these tools responsibly in complex, multimodal transport environments. Oliver Wyman’s 2024 performance transformation survey found that while 95% of executives at AI-capable companies expect the technology to have a significant impact on their businesses, only 45% report fully implemented AI solutions. The gap between expectation and execution is not primarily a technology problem. It is an implementation and expertise problem.
Understanding how transport networks actually operate – the edge cases, the accessibility requirements, the multimodal dependencies, the behaviour of live data under disruption – is not something that can be shortcut by a capable general-purpose AI vendor. It is accumulated domain knowledge. And it is the thing that separates AI deployments that improve journeys from AI deployments that simply add a layer of technical complexity to existing processes.
What This Means in Practice
The mobility organisations making the most credible progress with AI are not necessarily the ones with the highest-profile announcements. They are the ones investing in data foundations, provider integrations, and the contextual intelligence that makes AI useful in real conditions.
For transport agencies, councils and mobility platform operators, the practical question is not whether to use AI, but how to evaluate it honestly. That means asking what data the AI is actually grounded in, how it handles edge cases and accessibility requirements, and whether it has been tested against the complexity of live networks rather than clean demonstration environments.
The next phase of AI in mobility will be shaped less by which models are available – these are advancing rapidly and becoming increasingly commoditised – and more by who has the domain knowledge to deploy them in ways that are accurate, reliable, and genuinely useful to passengers. That is a less dramatic story than the broader AI narrative in mobility would suggest. It is also, on the evidence, the more consequential one.
A Note on Where SkedGo Stands
We are not neutral observers in this conversation. SkedGo builds journey intelligence grounded in real-time data from over 4,000 transport providers globally, and our TripGo Mobility AI reflects the architectural choice described above: AI capabilities built on verified, live transport data rather than on general-purpose language models generating plausible responses.
That principle – grounded, tested, continuously improved – may sound less exciting than the broader industry narrative around AI. But we think it is the right one.
SkedGo provides the journey planning technology behind some of the world’s most innovative mobility platforms. If you are thinking about how to deploy AI capabilities in a transport or mobility context, we would be glad to talk about what data-first AI looks like in practice.
This article was originally published by SkedGo.
