Guide · Insight
AI in UK Public Transport: what's actually real in 2026
A practical guide for operators and authorities
A commercial leader at a transport operator recently asked me to recommend an AI expert to help their board build a 12 month AI strategy. I couldn’t name a single person. Executives across the industry are being handed ‘Create an AI strategy’ as an objective, with no independent guide to what delivery actually looks like.
So I put together a practical list. It was for my mate, but I wanted to share with the ecosystem. Hopefully it helps!
Full transparency before reading. Firstly, I have spent my career on the supplier side in public transport and have been around AI for several years now. Secondly, I have a declared interest, as I now build AI software for transport tendering. So on the section around bidding below, I have skin in the game. I have, however, advised on the buyer’s side, which is the side I deal with less.
Why now? The franchising window
Bus franchising is creating something rare. As each region franchises its buses, a limited time window opens where the technology currently in use can be genuinely scrutinised. As operations shift from private to public, in the middle of the AI wave, there is a proper chance to ask; ‘Can we do better?’, ‘Is this fit for purpose?’, ‘Is this good value?’, ‘How can we position our technology for the next 10 years of operations?’.
Those windows can close. Contracts get signed, systems delivered and embedded and then we don’t get another chance for a long time. This is why I am so passionate and know that 2026 matters more than the AI hype cycle suggests.
Below are five handpicked questions that operators and authorities have actually asked me over the last two months, with straight answers, followed by eight AI use cases to back the answers up.
The five questions everyone is asking
1. AI has collapsed the cost of software development. Should we be building our own tech?
There are times where you genuinely can now and times where you definitely shouldn’t. Established markets where great competitive products exist, such as network optimisation tools, buying will be best. However, bus operators and authorities are full of niche legacy tools, pieced together over years, that serve a great purpose but are an administrative and technological burden. No software company is going to disrupt a niche tool, there isn’t a big enough market for them, but, now, building it yourself is genuinely viable. There is a catch, you will need to grow a small, lean software team to build and more importantly, maintain.
2. AI feels like a solution built without a problem. Where does it fit in public transport?
It links to the above, AI fits in the unglamorous, slightly dirty, definitely dusty areas. Mining years of defect cards or turning controllers shorthand into passenger comms. AI fits where transport already has mountains of messy data and manual effort…
3. AI companies come to us with no understanding of the criticality, risk and administration in public transport. How do we deal with that?
Proof. Make them prove that they understand. Every use case below includes a delivery reality, from unions to data sharing, because that’s what AI companies skip. Public Transport is brilliant at administration and compliance, which is usually framed as a weakness. In the purchasing process of AI tools, it’s the leverage. My challenge to suppliers at the bottom of this page gives you the exact questions to ask them.
4. Who actually knows AI in public transport and can give us independent advice?
Almost nobody, which is kind of why I did this. The consultancy market hasn’t caught up and most AI expertise won’t have sat in a control room. My advice to anyone is to build your own judgement, use articles, insights to ground your judgement and push suppliers for transparency. Pressure test everything. Pressure test this article! I have built an independent advisor agent that will challenge any idea on this page AI Advisor Agent. I deliberately haven’t made it agree with me. I am happy to be wrong in public, if it means the transport industry gets to right faster.
5. How do we procure critical software, on a 5 year contract, through a 12 month process, when technology changes every month?
Procure and reward adaptability. Write technology step changes into contracts, break points tied to capability shifts and data access rights, so you aren’t forever locked in. As AI models improve, push the supplier to forward those improvements on to you. The autonomous vehicle section below shows what this looks like in practice. The six month shelf life is real, so contracts need to be able to breathe.
The right AI use cases
in reverse alphabetical order so there is no bias
Video Analytics and Safety
The Idea - You installed cameras to explain incidents after they happen. AI turns the same cameras into sensors that see them coming.
The What - Analytics over existing onboard and depot CCTV. Monitoring yard movements, near misses, boarding friction and dwell time
The Delivery - The hardware is already bought. What is needed is a proper DPIA and a straight conversation with staff and unions about what is measured and why. Public Transport is great at admin, that’s an advantage this time.
The Innovation - Joining of ticketing and payments with CCTV. Not a surveillance tool, but a punctuality tool. Measure where the network actually leaks seconds, stop by stop, passenger by passenger, from footage already being recorded.
Tendering and Procurement (declaring my interest in this space)
The Idea - The tender process is silently becoming machine vs machine. AI is already writing bids, AI assisted marking will follow. We need to stop kidding ourselves and say it out loud, so the process stays fair
The What - For an authority, consistency checking specifications, summarising market engagement, compliance checks on submissions and clarification question triage. It cannot be scoring, evaluation has to stay human
The Delivery - Under the Procurement Act, AI can prepare an evaluators desk, but cannot own a score. Once we say that openly, we all relax and push on with the reality.
The Innovation - Publish your own AI use statement in the ITT. Declaring how you use AI and invite bidders to declare theirs. PPN 017 already asks this of the central government, an authority adopting it voluntarily sets the market standard. Also, run AI over your last ten procurements and find the clauses that are a decade old, we have all seen the Windows 7 requirements!
Revenue Protection and Fraud
The Idea - The public told us already where to start. The ONS found in August 2025 that 51% of adults are comfortable with AI being used to identify fraud and transport is the most trusted public service for AI use, at 35% ONS Study
The What - Multi operator, multi modal, pattern analysis over payment and ticketing data. Focusing on evasion hotspots and inspector deployment
The Delivery - The famous question… data sharing agreements between operators, authorities and suppliers. That’s the work.
The Innovation - Sequence your whole AI adoption around public permission. Start with fraud, build credibility and trust, then expand. Most are doing it the other way around.
Predictive Maintenance
The Idea - Failures are already being predicted by sensors. But I bet there’s a lot of proof in the defect cards nobody bothers reading again.
The What - The next generation of failure prediction on vehicles and assets.
The Delivery - The market will sell you sensors and telematics platforms. The real constraint isn’t models, it’s data.
The Innovation - Text, not telematics. LLMs are exceptional at mining years of scrappy notes, defect cards and driver reports. It’s unstructured and bus operators have mountains of it. This is the cheapest and biggest win on this entire list.
Passenger Information
The Idea - Passengers need answers to questions journey planners can’t take. Such as ‘will I make it?’
The What - Disruption comms and rich journey context delivered with LLMs.
The Delivery - Hallucination risk is critical here. Wrong information simply cannot be sent. Any deployment needs to be grounded in live operational data, not a chatbot bolted on to a website.
The Innovation - Point the LLM at the controller, not the passenger. Ground it solely in the control room log and let it turn the controllers shorthand into multi channel messages in seconds. Disruption is such an under loved workflow.
Network and Schedule Optimisation
The Idea - Most networks are optimised for a market that is about to be abolished. Franchising is redrawing the commercial logic underneath every timetable.
The What - AI timetabling, duty and vehicle scheduling. The most mature category on this list.
The Delivery - Buying the software is easy, the challenge is clean operational data and planners trust.
The Innovation - Scenario run the franchised network. Stress test hundreds of network variants against the authorities’ likely specifications. Walk into the bus franchising conversation holding the map they don’t yet have.
Demand Responsive Transport
The Idea - DRT has proved that routing works. It hasn’t proved the demand.
The What - AI routed flexible services are most likely live already in the UK
The Delivery - The algorithm was never the problem. The problem was honesty about demand, that was baked in at the initial network design.
The Innovation - Run two way conversion tests, regularly. AI over BODS data, AI over your own data. Ask two questions at the same time. Which fixed routes should become DRT and which DRT zones should become fixed routes? This is how you create a network that breathes.
Autonomous Public Transport
The Idea - The UK should put its feet up and let Oslo pay the tuition fee
The What - Autonomous is now real, but the learning curve is long and expensive
The Delivery - The economics can be someone else’s expensive experiment, we learn from their results
The Innovation - Prepare on paper, not on roads. Write AV readiness into bus franchise arrangements and procurement specs today. Covering data access rights, safety elements, depot power provision and break clauses tied to technology step changes. This is a 2030 reality, but the contracts being signed today will still be running in 2030. The cost of readiness is a clause with a mechanism.
My challenge to the software suppliers
Firstly, if you are doing anything noted above, I would love to hear from you! However, my point is that software companies fail transport operators and authorities on AI and I say that having done so myself. We plaster AI across our marketing, we shout about machine learning, but we never state what the AI does or how it’s operating.
Coming from tech businesses myself, we enjoy a luxury that most others don’t. We have daily stand ups with engineers who can explain the nuts and bolts, and product managers who are excellent at translating. We have the latest software at our fingertips. Operators and authorities often don’t have such luxury, as their focus is on operational delivery, not software building. So when a supplier says ‘we have AI in our product’. It lands as a word, not an action.
My challenge to every supplier in this sector, don’t be scared to state what the AI does, how it’s built and what models are under the hood.
I do so on my own website AI Web Page
Pressure test everything on this page
I built an Independent Advisor Agent so you can challenge any idea above AI Advisor Agent
I haven’t tuned it to agree with me. I want challenges, conversation and competition. I want to be wrong, so that we can get the right answer for everyone faster.