AI can optimise a route. It cannot fix the fragmented data landscape that makes UK transport unnavigable by machine.

The UK's ambition for AI-enabled transport is serious and well-funded. The Department for Transport published its Transport Artificial Intelligence Action Plan in June 2025, calling for responsible AI embedded in a resilient transport system delivering cheaper, cleaner, and safer journeys for all. The Connected Places Catapult has funded two successive programmes bringing innovators into transport AI. Autonomous vehicle trials are being accelerated ahead of the Automated Vehicles Act coming into full effect in 2027.

The technology is maturing. The policy intent is clear. But the thing that will determine whether any of it actually works, the question of whether AI systems in transport can communicate, share data, and function together, is still largely unresolved. Interoperability is not a technical footnote. It is the central unsolved problem in AI-enabled transport, and most of the sector has not yet reckoned with it seriously.

The Problem Starts With the Data

AI systems in transport run on data. Journey planning tools consume real-time service data, timetables, accessibility information, and live disruption feeds. Traffic optimisation models ingest sensor data, speed measurements, and incident reports. Predictive maintenance systems rely on asset data from infrastructure operators. Freight routing algorithms need port status, road capacity, and weather models.

In theory, all of this data exists. In practice, it exists in separate silos, built over decades by different operators, local authorities, and national bodies, each using different standards, different formats, and different update cadences. The DfT's own Transport Data Action Plan, published in 2026, acknowledged directly that data remains locked away, siloed or difficult to use, slowing innovation, undermining passenger confidence, and preventing operators from deploying more advanced analytics, AI tools and automation.

This is not a niche infrastructure problem. It is the reason why a passenger trying to plan a journey across three modes in two local authority areas frequently finds that the information available to a digital tool is incomplete, inconsistent, or simply absent.

A 2025 Open Data Institute and Nortal study of ten UK local authority cases found that most datasets held by councils remain unfit for algorithmic use, despite widespread AI pilot activity. The problem is not ambition. It is foundation.

What Fragmentation Actually Looks Like?

Fragmentation in transport data manifests in several distinct ways, each with different consequences for AI deployment.

Format inconsistency

Different operators and local authorities use different data formats, naming conventions, and classification systems. The DfT has identified this explicitly in its commitment to develop a Transport Data Ontology, a common language for how datasets, assets, and terminology relate across modes. Without this, an AI system trained on one operator's data may simply fail to interpret another operator's output.

Update frequency mismatch

‍Real-time AI applications require real-time data. But many transport datasets, particularly those held by smaller local authorities, are updated manually, infrequently, or not at all. An AI journey planning tool that relies on accessibility data about bus stops, for example, cannot function reliably if that data is months out of date. The DfT is now piloting AI agents to introduce accessibility data at public transport stops precisely because manual updating has failed to keep pace.

API gaps

‍The DfT's Transport Data Action Plan promotes suitable APIs as the expected method of sharing transport data. The ambition is to move beyond static file exchanges and legacy system integrations. But as of 2026, many local transport operators still do not expose data through APIs at all. This means AI systems cannot access live feeds and must instead rely on periodic batch transfers, breaking the real-time capability that makes AI valuable in transport.

Governance inconsistency

‍Even where data exists and is technically shareable, the question of who is permitted to share it, under what conditions, and with what liability, remains unresolved across much of the sector. Different operators have different contractual positions, different interpretations of data protection obligations, and different risk tolerances.

The Equity and Accessibility Dimension

Fragmentation does not create equal problems for all passengers. It creates systematically worse outcomes for passengers who depend most on transport AI to navigate the system.

Passengers with disabilities rely on accurate, detailed accessibility data to plan journeys that are genuinely feasible for them. When that data is incomplete, inconsistent, or siloed within a single operator's system, AI journey planning tools simply cannot surface the information needed to make those journeys possible. The consequence is not just an inconvenient experience. It is exclusion from the transport network.

Regional inequality follows a similar logic. AI tools tend to work best in the areas where data infrastructure is strongest, typically well-funded metropolitan areas with integrated transport authorities and modern data systems. In rural areas, smaller towns, and regions with fragmented transport provision, the data quality is lower, the API coverage is thinner, and the AI tools deliver less. The benefits of AI-enabled transport, better journey planning, more predictive disruption information, smarter ticketing, accrue disproportionately to areas that already have better transport.

Parliamentary research published in early 2025 confirmed that transport remains a significant barrier to opportunity for young people, women, and communities in coastal and rural areas. AI tools that amplify existing data inequality will worsen, not improve, this picture.

The Vendor Lock-In Risk

There is a second-order problem that receives less attention than the data quality challenge, but which has significant long-term implications.

When local authorities and transport operators procure AI tools from commercial vendors, they are often effectively committing to that vendor's data architecture. The AI system ingests local data, learns from it, and produces outputs in a proprietary format. Over time, the local authority's operational processes become dependent on that vendor's specific implementation. Switching becomes expensive and disruptive, even if better tools become available.

This is the pattern of vendor lock-in that the DfT's interoperability agenda is partly designed to prevent. If transport operators procure AI tools without mandating open standards, common data interfaces, and portability requirements, they will find themselves in the same position in 2030 that many found themselves in with legacy IT systems: dependent on suppliers they cannot easily replace, paying for upgrades they did not choose, and unable to adopt better solutions because the migration cost is prohibitive.

Interoperability is not just a data quality issue. It is a procurement and governance decision that will determine whether public sector AI investment retains value over time, or becomes the next generation of legacy lock-in.

What DfT's Action Plans Signal

The direction of travel from DfT is unambiguous, but the implementation timeline is long and the challenges are structural.

The Transport Data Action Plan committed to developing a Transport Data Ontology to create a common language across modes. It promoted APIs as the default for data sharing. It called for data to be held in a condition that is findable, complete, accessible, interoperable, and reusable, in language borrowed directly from FAIR data principles. It set out a commitment to move toward data being open and shared by default.

These are meaningful commitments. But they require action at multiple levels simultaneously. National standards bodies, regulators, transport operators, local authorities, and technology vendors all need to align. The CIHT has noted that a platform or community is needed to share knowledge and best practice, and that local authorities need appropriate funding, guidance, and procurement frameworks to implement AI in transport successfully. Without that coordination, the DfT's action plans will produce pilots without scale.

What Needs to Happen Next

Three things need to change to make AI-enabled transport interoperability real rather than aspirational.

Procurement standards need to embed interoperability requirements. Every AI tool procured by a transport body, whether central government, an arm's-length body, or a local authority, should include requirements for open data interfaces, portability, and adherence to the Transport Data Ontology as it develops. This is where the structural lock-in risk is created or prevented.

Data governance needs to be treated as infrastructure. Interoperability is not just a technical challenge. It requires clear rules about who owns what data, under what conditions it can be shared, and what liability attaches to sharing. Without this governance layer, even technically compatible systems will not actually share data because the legal and contractual framework does not support it.

Local authorities need capability support, not just guidance. The 2025 GoLLM AI Readiness research found that AI activity in local authorities remains largely pilot-dominated, with scaling constrained by legacy IT, fragmented data estates, and unclear ownership. Publishing action plans is not sufficient. The bodies that hold much of the transport data that AI needs are under-resourced, technically constrained, and unable to implement interoperability standards without direct investment in capability.

The AI Opportunity and the Interoperability Condition

AI has genuine transformative potential in UK transport. Better demand forecasting, smarter network optimisation, more accessible journey planning, safer infrastructure monitoring, and more resilient freight logistics are all achievable. The technology to do these things exists or is within reach.

But the returns on AI investment in transport are directly conditioned on the quality and interoperability of the data environment in which AI operates. An AI journey planning tool that cannot access reliable, real-time, interoperable data from across the modes a passenger needs to use is not a tool that will transform the passenger experience. It is a more sophisticated version of the same incomplete picture passengers already have.

The organisations that invest in the data architecture now, before the AI tools are fully deployed, will find that the AI capabilities follow naturally. The organisations that deploy AI tools first and hope the data infrastructure catches up are building on foundations that may not hold.

Interoperability is not the most exciting conversation in transport AI. But it is the most important one.