When people talk about AI in transport, the conversation often jumps to the big use cases. But some of the most useful opportunities are much closer to home, in the repetitive work happening every day in depots, control rooms, customer service and back-office teams.
In fact, 86% of UK operators are now integrating AI in some form, yet manual paperwork alone still costs the average business £62,000 a year. At the same time, McKinsey’s latest research shows why, in any single business function, no more than 10% of organisations have scaled AI beyond the pilot stage.
So while more operators are experimenting with AI, much of the everyday work is still done manually.
That doesn’t mean handing whole processes over to AI. In transport, operational knowledge and judgement are too important for that. A more useful place to start is the repetitive work around those decisions: finding information, processing data, producing reports and dealing with routine tasks.
Here are eight everyday transport tasks where AI can already help.
In this article, you’ll learn:
|
1. Answering routine passenger enquiries
Passenger contact centres deal with a predictable flow of questions every day, such as timetables, fares, delays, cancellations, refunds, lost property, and journey information. Many are routine, but they still take adviser time, and demand can rise quickly when services are disrupted.
Where AI comes in
Voice and conversational AI can handle routine enquiries without an adviser, including outside contact centre hours and when demand increases during disruption, while passing more complex cases to the team.
Let’s look at an example. Norwegian bus and ferry operator Reis Nordland introduced an AI travel assistant and saw enquiries reaching its customer service team drop by 55% within six weeks, with response times cut nearly in half.
What needs to be in place
The strongest solutions go beyond answering questions from a fixed knowledge base. By connecting AI to timetables, traffic updates, service policies and real-time transport feeds, operators can give passengers information that reflects what is happening across the network. That’s particularly useful during disruption, when contact volumes increase at the same time as information changes quickly.
For CIOs, that makes integration the deciding factor: connecting to the operational systems already in place, updating information in real time, and recognising when an enquiry needs to move to an adviser. Get that right, and considerably more routine demand comes away from the contact centre without compromising the passenger experience.
Where people stay involved
Complex complaints, accessibility requirements, compensation and unusual disruption still benefit from an adviser who can understand the wider circumstances. We find that the strongest model is one where AI handles the predictable volume and advisers concentrate on the conversations that genuinely need them.
The CIO test: Can the solution connect to the information passengers need and handle routine demand without creating more work for the contact centre?
2. Building and adjusting driver schedules
Driver scheduling is already an optimisation problem. Planners have to balance routes, duties, driver availability, working-time rules, depot requirements, absence and disruption, while keeping costs under control and ensuring services are covered. Industry research points to a global shortage of around 2.4 million transit workers, which leaves many depots short-staffed and driving up unscheduled overtime, fatigue and turnover.
Where AI fits
AI-assisted scheduling and optimisation tools can work through large numbers of possible combinations far faster than a planner could manually, helping teams build efficient duties and respond when the plan changes.
Let’s look at an example. Scottish operator West Coast Motors now uses AI for around 60% of its scheduling and timetabling. Operations and systems supervisor Paul Borthwick sees the impact land with drivers directly: “The day of the longer shifts and longer schedules are over. A lot of drivers want better family life, and I think that’s important to try and get a better balance.”
That becomes particularly useful during disruption. A driver calls in sick, a vehicle becomes unavailable or services need to be adjusted, and planners can assess alternative options quickly rather than rebuilding parts of the schedule by hand.
What needs to be in place
A schedule can look perfect on a screen and still fall apart on a Tuesday morning, because the software didn’t know that one depot only has two relief drivers, or that a particular route always runs late past the school run. Industrial agreements, fatigue and rest requirements, route knowledge, depot constraints and local working practices have to go into the model alongside the driver, vehicle and duty data. Leave any of it out, and the tool will hand planners a schedule that sounds like it might be optimal, but in reality, it would be useless if deployed.
We find that integration decides whether the tool is working from today’s reality or yesterday’s. Connect it to live workforce and operational data, and it knows who’s available right now. Leave it running on a static export, and every schedule it builds is already out of date by the time someone acts on it.
The thing is, a feature lists won’t tell operators any of this. And that’s why AI applications fail. A platform built around single-depot, urban routes won’t automatically know how to handle a rural network with split shifts and long relief runs. The features look the same, but the results won’t be.
Where people stay involved
Planners still bring the operational knowledge that comes from running the network every day. AI gives them the ability to assess more options, faster, while they retain control over exceptions, workforce implications and the final plan.
The CIO test: Can the technology model your real operating constraints and give planners better options when conditions change?
3. Spotting vehicle faults before they become failures
Engineering teams already have multiple ways of understanding vehicle condition, from inspections and servicing schedules to diagnostic codes and reported faults. The difficulty can sometimes be in knowing what needs attention before it affects service.
Where AI fits
Predictive maintenance systems analyse vehicle telemetry, diagnostic data and maintenance history to flag early indicators of component deterioration. Instead of waiting for a fault to develop or relying solely on fixed maintenance intervals, engineering teams can get earlier warning of where intervention is likely to be needed.
Let’s look at an example here, too. The Metropolitan Transportation Authority in New York increased maintenance productivity by 75% and cut material costs by 24% after introducing an AI-powered predictive maintenance system across its bus fleet.
What needs to be in place
The good news is, most modern fleets already generate large volumes of data. They just need to bring enough of it together and use it consistently.
Mixed fleets can mean different manufacturers, vehicle generations, telematics platforms and maintenance systems. The right predictive maintenance solution needs to work across that mix and turn those different data sources into information engineering teams can use.
Ultimately, a prediction on its own doesn’t do much. The value sits in the warning time it buys, enough for engineering to plan the intervention, secure the parts and allocate workshop capacity before the vehicle is lost from service.
Where people stay involved
Engineers remain responsible for diagnosing the vehicle, prioritising work and deciding what action should follow. Predictive maintenance gives them another source of information and, crucially, more time to act on it.
The CIO test: Can the solution turn the fleet data you already have into reliable warnings early enough for engineering to act?
➡️Further reading: Predictive maintenance in bus fleets: The £6 million opportunity in engineering data
4. Producing operational and management reports
Weekly and monthly reporting can mean drawing on punctuality, lost mileage, fleet availability, engineering performance, customer complaints, safety, staffing and financial data before anyone gets to the job of understanding what the numbers are saying.
Where AI fits
AI can reduce much of that preparation. It can consolidate agreed information, identify changes or exceptions that need attention and produce first-draft commentary for operational and management reports.
In fact, across UK businesses, improving business operations is already the single most common reason for using AI, reported by close to 60% of firms, according to the ONS. For transport operators, reporting and management information sit squarely in that category.
That gives managers more time to investigate performance rather than compiling it. AI can flag that lost mileage has increased by 8%; an experienced operator can establish why it happened, whether it is likely to continue and what needs to change as a result.
What needs to be in place
Imagine two reports landing on the same desk on a Monday morning: engineering says fleet availability was 94% last week, operations says 89%. Both are right, by their own definition of “available.” One counts a vehicle as available the moment it leaves the workshop; the other only once it’s back on its scheduled route.
AI can’t fix that on its own. Connect scheduling, ticketing, telematics, engineering, customer and finance systems, and it has more raw material to work with, but if those systems don’t agree on what “available” or “on time” means, it will write a confident report built on numbers that don’t reconcile.
Get that alignment right first, and the rest follows: less manual reconciliation, and less doubt about which source is authoritative for a given measure.
So, for CIOs, the real test of a solution is whether it can show its working. Where each figure came from, and which system’s definition it used to get there.
Where people stay involved
Management information only becomes useful when someone applies operational context. Managers still need to challenge anomalies, understand the causes behind performance and decide what action follows.
The CIO test: Can the solution work reliably across your existing data sources and give managers information they can trace back to the underlying operation?
➡️Further reading: Transport AI ROI: 2026 results
5. Finding engineering information
An engineer who’s fixed the same fault for years knows things that were never written down. Which sound usually means a bearing, not a belt. Which fix actually works, even if it’s not the one in the manual. When that engineer retires, all of that goes with them.
Where AI fits
An internal AI assistant can give engineering teams a faster way into that information. Engineers can ask a question in everyday language and retrieve relevant guidance from approved technical sources, rather than searching across documents and systems individually.
That becomes particularly useful on mixed fleets, where engineering teams may be working across different vehicle types, manufacturers and generations of technology.
What needs to be in place
Somewhere in most workshops is an engineer who’s seen a particular fault on a particular vehicle type a dozen times, and knows within minutes what a junior colleague might spend half a shift chasing through manuals. That knowledge doesn’t show up on an org chart or in a job card. It walks out the door the day that engineer retires, and it doesn’t come back.
And that’s a problem across industries. The IET’s 2025 skills survey found that while 80% of UK engineering employers say they share knowledge effectively between employees, only 57% feel they do so effectively with staff who are retiring.
Bringing approved technical information and accumulated maintenance knowledge into a controlled system doesn’t replace that experience, but it stops it from disappearing entirely. A newer engineer can find relevant guidance without waiting for the one person in the workshop who’s seen it before, and that person’s knowledge stays useful long after they’ve moved on.
The source material still needs to be tightly managed. Current manuals, approved procedures and verified maintenance records should take priority, with engineers able to see where an answer has come from.
Where people stay involved
The assistant helps engineers find and interpret the information available to them. It doesn’t take responsibility for the vehicle. Diagnosis, repair decisions and anything affecting whether a vehicle is safe to return to service remain with qualified engineering teams.
The CIO test: Can the solution make your engineering knowledge easier to find and share, while keeping its answers grounded in current, approved technical information?
6. Processing invoices and operational paperwork
Transport operations generate a steady flow of paperwork across parts, fuel, fleet, depots, contractors and other suppliers. For finance teams, that can mean extracting invoice data, matching purchase orders, entering information into finance systems, routing approvals and resolving anything that doesn’t match. That’s a major resource drain.
In fact, a recent Road Haulage Association survey found that manual data entry is one of the biggest sources of daily inefficiency for 43% of UK operators, with more than nine in ten still working across disconnected systems. So the paperwork is a symptom of that, rather than a separate problem finance happens to own.
Where AI fits
AI-based document processing can take on much of the repetitive handling. It can read invoices in different formats, extract the relevant information, match it against purchase orders or other records and identify discrepancies for review.
That means finance teams no longer have to treat every invoice the same way. Predictable transactions move through the process on their own. Exceptions get routed to the right person instead.
What needs to be in place
The value increases when the solution connects properly with existing finance, procurement and operational systems. That way, invoices get checked against information the business already holds, rather than creating another standalone process.
Ultimately, the controls should reflect the transaction. A routine parts order that matches the purchase order and agreed tolerances needs little more than a glance. A fuel invoice that’s suddenly 20% higher than the same supplier charged last month, or a payment request against a purchase order that’s already been closed, needs a person to look at it before it goes anywhere near payment.
That’s the real test of a system here. It needs to tell those two invoices apart. A tool that treats every invoice the same either buries finance in exceptions or waves through the ones that needed a second look. Either way, the operator ends up doing the checking anyway, just later and with less context.
Where people stay involved
Finance and operational teams still handle exceptions, disputed charges, unusual transactions and approvals that require judgement. Instead of checking every document manually, they can focus their time on the transactions that need attention.
The CIO test: Can the solution automate routine transactions within your existing controls and reliably identify the exceptions that need someone to step in?
➡️Further reading: The 2026 operational mismatch in transport
7. Monitoring safety and compliance
Transport operators already collect large volumes of safety information through CCTV, incident reports, vehicle and station data, and frontline reporting. Reviewing that information takes time, particularly when teams are looking for one event that needs immediate attention, or a trend that only becomes visible across hundreds of records.
Where AI fits
Computer vision and AI analytics can monitor those information streams continuously, flagging potential hazards, unusual behaviour, crowding or incidents for staff to review.
TfL has been testing exactly this. A year-long AI trial at Willesden Green station generated 44,000 alerts, with 19,000 sent through to station staff in real time, an average of 55 a day. A Rail Accident Investigation Branch report later found staff acknowledged only 60% of them. That’s the trade-off CIOs need to weigh before they buy in.
And TfL has kept building on this since. Its DLR platforms at Custom House already use AI to detect people or objects on the tracks, and British Transport Police has just expanded a live facial recognition trial from mainline stations onto the London Underground itself.
What needs to be in place
For a CIO, analysing CCTV is the easy part. Fitting those alerts into an established safety process is a little trickier.
Someone has to decide what gets flagged, who receives the alert and what happens next, and set that out before go-live. Accuracy counts for something different here too. Too many false alerts, and staff stop trusting the system. Once that happens, a real alert can get missed too. You may remember this example. A Rail Accident Investigation Branch report into a fatal accident at Ickenham found that during a year-long AI trial at Willesden Green station, staff acknowledged only 60% of the 19,000 alerts sent to them, an average of 55 a day. Four in ten went unactioned, whatever the reason behind it.
Privacy, access to footage and audit trails belong in that same conversation, especially where passenger or employee data is involved. Facial recognition trials are moving from railway stations onto the Tube network right now, while operators are still working out where the lines sit.
Where people stay involved
This is a good example of AI supporting a safety-critical function without taking the decision away from the people responsible for it. The technology can watch more information and flag potential problems earlier. Trained staff still investigate, interpret and decide how to respond.
That gives safety teams another way to spot risk, without expecting people to monitor every feed or review every record manually.
The CIO test: Can the solution identify useful safety information early enough to support your teams, with the accuracy and controls needed for where it will operate?
8. Turning operational meetings into actions
Daily operational calls, depot performance meetings, engineering reviews, safety meetings and disruption debriefs all generate decisions and actions that someone then has to capture, circulate and follow up.
Where AI fits
AI meeting tools can transcribe discussions, summarise key points, extract decisions and produce action lists immediately after a meeting. For regular operational meetings, that removes a repetitive administrative task and gives teams a consistent record of what was agreed, who owns each action and what needs to happen next.
It also makes previous discussions easier to search, which helps when an issue runs across several meetings or teams.
What needs to be in place
The technology is the easy part to get working. What’s more challenging is knowing what it’s processing, and where that information goes afterwards.
Let’s put this into perspective. Microsoft-commissioned research found 71% of UK employees have used unapproved AI tools at work, with over half doing so every week. A meeting assistant is one of the easiest to slip in unnoticed. Anyone on a call can add one in seconds, and IT may never know it happened.
For a transport operator, that covers a lot of sensitive ground. A depot review can include employee information. An engineering meeting can cover incident and fault detail. A disruption debrief can carry operational or passenger information. Staff using third-party meeting assistants without approval could be sending any of it outside the business’s agreed controls.
Operators don’t have to leave that to chance. A defined policy, paired with an approved tool built to their own requirements for data storage, access, retention and security, closes off the risk without banning the technology outright.
Where people stay involved
AI can produce the first record quickly. But meeting owners still need to check the context, decisions and actions before anything becomes an official record, or gets circulated more widely.
Done well, this strips out the note-taking and admin. Accountability for what was agreed doesn’t move an inch.
The CIO test: Do you know which AI meeting tools are being used across the operation, what information they can access and where that information is stored?
Start in the depot
Every example in this piece came from watching how the operation runs day to day.
Spend time with a controller during disruption. Sit with engineering. Listen to passenger calls. Look at how planners build duties and what depot managers are still compiling manually every week.
Do that, and it doesn’t take long to see where time is being lost: repetitive work, bottlenecks, information spread across different systems, and experienced people spending hours gathering what they need before they can make a decision.
Those are good places to assess what AI could improve and which solutions are suited to the job.
For transport operators, progress will come from understanding those operational problems in detail, knowing which systems and data sit behind them, and giving teams technology that genuinely makes the day-to-day operation easier to run.
If you’re deciding where AI could improve your transport operation, we can help you find the use cases worth pursuing and the solutions best suited to them.
Our advice comes from experience on both sides of the table. Having spent decades working in transport, including at Group CIO level at Stagecoach, we understand the operational and technology decisions behind getting new systems into use.
Talk to us about where AI could work in your operation.
Contact us: info@mindtheaigap.com | Call us free: 0800 009 6408