Getting ready for AI: improving bus services

AI is the always-on dispatcher in the control room. It ensures that nothing gets missed during a tea break, that manual tasks are automated and alerts sent. This frees up capacity so that humans can focus on the real challenges – those that need insights beyond the capabilities of AI

AI is creeping into a wide range of transport-related tools, services and procedures. But is the sector ready? The government has been busy with preparation for AI, with much happening in the space for public sector, transport operator and supply chain AI-readiness. As the Department for Transport (DfT) works on its forthcoming Transport AI Strategy and AI Strategy Action Areas, much work has gone into identifying challenges to AI adoption. 

The transport, logistics and warehousing sector in the UK has struggled to adopt AI. "There are divergent opinions, fragmented ecosystems, policy, politics and regulation at play. But we also have the right fundamentals for transformative change with the use of AI and machine learning technologies,” says the 2023-2024 Innovate UK Bridge AI programme’s annual report. 

It adds: “Parts of the sector are still learning how to leverage value out of more basic digital ways of working – let alone using AI. But there are forms of AI that hold the most near-term promise for the sector. Firstly, the use of machine learning models to support quantitative analytics, unlocking insights in complex system design and operation, from dynamic logistics re-routing optimisation to stock and crew balancing. Secondly, the (measured) use of LLMs to support human-understandable engagement with, and presentation of, mobility information to communicate complex scenarios to diverse audiences, whether a journey planner or a summary report.

Read more about AI readiness in the transport sector

By and large, suggests Bridge AI, the type of AI being deployed is not the core of the adoption challenge: instead, a multitude of barriers exist ranging from dynamics around industry culture, skills in the workforce, ethics and security, perception of AI, procurement and funding models, data governance and standards. None of these issues are new, and documents ranging from the Transport Data Strategy all the way to more recent technical work done by Digital Catapult’s AI adoption toolkit have laid out sensible ideas to address and minimise these challenges.

To drive the biggest impact, the key challenges blocking AI adoption in transport are procurement and funding models, AI perception (from users to those who hold the budgets), and skills of those purchasing, designing and using AI-driven tools safely and effectively. 

“Early signs of success will become evident with the communication of more case studies and an uptick in cross-transport-sector communication. AI adoption in transport is coming, but it will take a big effort in public and private alignment to unlock the real value,” says the report.

Buses leading the way

One sector, however, is showing the way forward and trying out practical applications of AI. In early 2024, says Tim Rivett, RTIG held a one-day conference on the topic of Using AI to Improve Bus Services. As is typical of an RTIG event, some attendees were real experts with long and in-depth experience developed over many years, and others were there to find out what all the fuss was about and come away with an understanding of the subject.

---

Join Tim and other experts at Transport AI 2025 for a deep dive into AI innovation and procurement in the transport sector

---

Across the day it was clear that data is everything when it comes to AI. Garbage in means garbage out, and to be successful you need as much volume as possible from replicable sources of understood quality. 

Before you get to the point of deciding what AI to use, it’s necessary to develop and refine the business case and determine what the objective is going to be – and what success looks like. Is the aim to improve productivity, efficiency, or decision-making?

Adoption and use

Success with AI is often not as much about the AI tools and models used, but how an organisation adopts it and uses it. The teams it is being applied to need to buy into the project, understand how it works, and be able to accept the outcomes it is producing. 

Problems in a project start when people think AI is smarter than it is, and those implementing systems need to be clear with their clients what the limitations are and where problems can exist. 

Managing risk

Organisations need to start with small projects which can be used to build understanding and skills and show success without significant risk. 

Only once there is some confidence should riskier projects be developed and planned. Because of the risks of bias and uncertainty in outcomes from AI projects, it is important to develop within your organisation a governance process for the development and use of AI. 

Whilst each organisation is different, there is a developing base of helpful governance advice and support, including from the UK government. The Government framework sets out five principles to guide and inform AI development in all sectors, as well as within Government itself.

Selection and management of data being used for training should form a significant part of the governance processes. 

Understanding the context

It is not appropriate simply to point a model at some data without understanding what is there, and also understanding where there may be bias or inconsistencies in the data or its quality. 

This is of particular importance where decisions may be made which impact particular groups of people, and especially so when models are being trained using personal data as, in this case, significant privacy controls will need to be put into place.

Published data on its own cannot determine whether data is right or wrong – operational insight is needed. For bus services this means knowing the operators’ intentions for any particular service – for example the schedule, and applying the real world delivery to it – as well as any live running information. 

Know local conditions

Without understanding all aspects of the data, an algorithm applied to the live running data would detect problems which weren’t really there – for example a route variation which was planned and not due to an unexpected diversion. The converse is true – the need to use operational insights to determine and verify data accuracy.

It is not always possible to apply AI systems developed in one place to another, for example using algorithms to improve journey times by improving traffic light timings. What would work in North America where bus stops are typically located before traffic signals wouldn’t necessarily work in Europe where bus stops are placed after the signals. Understanding the local conditions and contexts is important to maximising benefit.

During our discussions, it was highlighted that a key challenge for transport planners is understanding the demand and usage of services, particularly where interchange is required. 

Transfer between vehicles

This challenge is a classic case for using AI with ticketing data and image processing. In some far eastern countries, they are already using image processing using street CCTV to identify vehicle transfers. This is currently being done post-event, but some research papers are suggesting it could be done in real-time in the near future. 

Image analysis has been done in the UK in trials for transfer analysis between vehicles using on- bus CCTV only, and models are clever enough to know the same person even if something changes, such as a passenger removing a hat, or no longer carrying a bag.

There are some practical applications of English Bus Open Data Service data being used with generative AI and planning models. These integrate with Urban Traffic Management Control systems to manage bus journey times, during periods of disruption where evidence shows significant improvement compared to previous approaches.

When it comes to public transport, local knowledge has always been key to the success of bus services – and always will be. AI will not change this.

Using the data from a model alongside local knowledge is critical. There may well be reasons, other than practicality, as to why an intervention should take greater priority than the data would suggest – political visions or manifesto commitments, for example.  

What an AI platform should do is to provide the quantitative data to support the decisions proposed, which is little different to how decisions are made and evidenced now. 

Speed of analysis

What AI really helps with is the speed of analysis and helping to ensure that more possibilities are analysed – AI isn’t dependent on having available resource to manually review and guarantees 100% analysis. Therefore with careful data management and selection, AI could help reduce human bias towards a particular solution.

Freeing up human resources

Where it is used for assisting with the operational management of bus services, AI is the always-on dispatcher in the control room. It ensures that nothing gets missed during a tea break, that manual tasks are automated and alerts provided to changes that need investigating. This frees up capacity so that humans can focus on the real challenges – those that need insights beyond the capabilities of AI.

---

Tim Rivett is General Manager, RTIG-Inform