Predictive maintenance in bus fleets: The £6 million opportunity in engineering data

A 100-bus fleet can lose over $6.2 million a year through unplanned breakdowns, and around 78% of that downtime is preventable.

The thing is, most operators already have the data to help prevent these breakdowns, and it has been building for years. Telematics, diagnostics and maintenance history are well established, giving a detailed view of how vehicles are performing day to day. Even so, servicing still follows fixed intervals, vehicles come off the road on schedule whether they need to or not, and early warning signs go unaddressed.

Meanwhile, the way decisions are made has changed very little, which is why these issues persist.

But it doesn’t need to be like this. Leading firms are using predictive maintenance capability to interpret existing data, which informs when vehicles are taken out of service, how work is prioritised and how risk is managed across the fleet. And this blog looks at exactly how predictive maintenance improves operations, across the full fleet.

The data is there. Decisions lag behind

The data is there. The challenge is turning it into outputs that influence decisions at the right point in the operation. In fact, we find that most fleets already operate with a high level of visibility.

Vehicle performance, fault codes and component behaviour are tracked continuously, often across multiple systems. But that data rarely feeds into day-to-day decisions. Maintenance planning still follows fixed intervals in many depots. Vehicles are booked in based on mileage or time rather than current condition, and workshop capacity is allocated accordingly.

Meanwhile, emerging faults develop between those intervals, only to be addressed once they begin to affect service.

And that creates a problem. Work is carried out on vehicles that could have remained in service, while others fail between planned intervals and disrupt operations. Over time, availability becomes less predictable, recovery activity increases, and cost per mile goes up.

What industry data tells us

• Over 90% of buses manufactured after 2015 broadcast diagnostic data through factory telematics (engine temperature, fuel burn, vibration, brake wear, battery voltage, coolant pressure, transmission data) streaming continuously.
• Only 27% of fleets have operationalised predictive maintenance. 65% plan to adopt it by 2026, but haven’t moved.
• A typical AI predictive model catches 75% of failures 2–4 weeks before breakdown; long enough to schedule the work into a planned slot rather than a roadside recovery.
• What it changes: McKinsey research finds predictive maintenance cuts unplanned downtime by 30–50%, lowers maintenance costs by 18–25%, and extends asset useful life by 20–40%.
• What it costs to ignore: unplanned vehicle downtime in the UK runs £360–£610 per vehicle, per day.

Downtime: the cost is higher than reported

Downtime is usually tracked through engineering metrics (mean time between failure, vehicle off-road time, workshop backlog). But that’s not where the cost ends up. Once a vehicle fails in service, what starts as a technical issue becomes an operational and financial one, with cost spread across multiple teams.

Reactive repairs cost 3.5–4.5x more than preventive repairs.

So, engineering sees the failure, operations manage the disruption, and finance absorbs the cost later. And these are rarely brought together in a way that informs day-to-day decisions. For instance, a typical in-service failure carries cost across:

• Lost mileage and revenue
• Service recovery and control room activity
• Spare vehicle and driver deployment
• Overtime and disruption to planned duties
• Performance penalties in contracted environments

Because each of those costs is owned by a different team, the full financial impact rarely sits in one place. That’s why attention needs to move earlier in the decision cycle, to when intervention should take place before service is affected. When that happens, downtime gets treated differently. It becomes a cost driver that links operational risk, timing of intervention and financial exposure across the business. That means work gets scheduled against service risk rather than mileage alone, fewer in-service failures follow, recovery activity reduces, and cost per mile becomes easier to manage.

Parts inventory: cost without control

Parts inventory is set to protect availability. Critical spares are held to avoid delays, particularly where lead times are long or failure impact is high, and stock levels are often based on historic consumption and worst-case assumptions rather than current component condition or failure risk. Over time, that leads to excess inventory sitting across depots, with limited visibility of how it aligns to actual demand.

Spare parts inventory carrying costs reduced by 20–30% with predictive maintenance.

Needless to say, shortages still occur. Components with lower usage frequency but high service impact are not always available when needed, which leads to delays, emergency procurement and extended vehicle downtime. So both conditions exist at once, with capital tied up in slow-moving stock and operational disruption caused by missing critical parts.

This comes back to how demand is defined. When parts planning is based on elapsed time or historic usage, it struggles to reflect how assets are actually performing. Failure patterns vary across vehicle types, routes and operating conditions, but inventory policies do not always adjust accordingly.

When that changes, planning becomes more precise. Component-level demand is aligned to expected failure rates, procurement reflects lead times and service criticality, and stockholding begins to match operational risk more closely. As a result, inventory turns improve, emergency orders reduce, and workshop delays linked to parts availability fall. Working capital comes under tighter control, without increasing risk to availability.

Further reading: The 2026 operational mismatch in transport

Safety: intervention happens too late

Safety performance is still largely managed through post-incident review. Events are recorded, investigated and reported, with actions taken once patterns become visible, and by that point the underlying behaviours have already been in place for some time.

And this still happens despite the fact that, across most fleets, the relevant signals are already available. Telematics data highlights driving behaviour, engineering data shows component wear and fault recurrence, and operational data reflects route conditions and service pressure. The thing is, those signals rarely get used together in a way that supports earlier intervention. So coaching takes place after incidents, maintenance checks are triggered after failure, and route or duty changes happen once disruption has already occurred.

Equipment failure-related accidents reduced by ~40% with predictive systems.

But when those signals are used earlier, risk can be identified at component, vehicle or route level before it results in an incident, and engineering and operations can act before claims accumulate or performance begins to deteriorate. The impact tends to be visible quite quickly:

• Fewer preventable incidents
• Lower vehicle damage and repair cost
• Greater stability in claims frequency
• Reduced regulatory exposure

From reactive review to active risk management. That’s what earlier intervention makes possible.

Fleet availability: where it shows up

Fleet availability is where engineering decisions become visible to the rest of the organisation. At board level, performance is tracked through availability, cost per mile and contract delivery, and engineering activity feeds directly into those outcomes, whether it is measured that way or not.

25–35% reductions in unplanned downtime within 6–12 months of AI predictive maintenance implementation.

Small changes at vehicle level scale quickly. A marginal reduction in in-service failures, or a slight improvement in repair turnaround time, compounds across the fleet over a twelve-month period. When failure rates reduce and intervention is better timed, vehicles spend more time in service and less time moving between breakdown, recovery and workshop cycles. And it’s easy to see how that influences day-to-day operations:

• More consistent service delivery, with fewer gaps caused by vehicle loss
• Reduced reliance on spare vehicles, particularly at peak
• More predictable workshop activity, with less disruption from unplanned work
• Stronger contract performance as reliability becomes easier to sustain

Availability reflects all of that combined, which is why it carries weight at board level. It shows how consistently engineering, operations and planning decisions are working together over time. Because when availability improves in a controlled way, cost per mile stabilises and service delivery holds.

Further reading: Transport operations 2026: Meeting the next two years’ challenges

Why results are inconsistent

Progress is uneven across the sector, even where similar data and tools are in place, and it usually comes down to how decisions are coordinated. As we’ve touched on already, engineering, planning and depot teams are often working to different priorities, and those decisions are rarely aligned in real time.

This means bus fleet operations are impacted daily:

• Workshop capacity is filled before the highest-risk vehicles are prioritised
• Parts are ordered based on historic usage rather than current failure patterns
• Vehicles remain in service despite known issues, because taking them out creates short-term pressure elsewhere

In many cases, predictive outputs exist but sit outside the systems and routines where decisions are made, which means they aren’t built into how work is planned, prioritised or scheduled across the operation.

But where results are more consistent, engineering, planning and operations work from the same view of risk, availability and workload, and that shared view changes how decisions get made. Vehicles are taken out of service at the right point, workshop time is allocated against service impact, and parts decisions reflect expected demand rather than past usage.

Further reading: Transport AI ROI: 2026 results

What We See In Transport

Across leading bus fleet operators, the difference is whether available data and predictive outputs are used at the point where decisions are made. Here’s what that tends to look like:

Where the £6 million comes from

The value does not sit in one area, it builds across the operation. Each change on its own looks small. A reduction in in-service failures, fewer emergency part orders, and more consistent workshop activity. But over a twelve-month period, across a large fleet, those changes accumulate.

Unplanned downtime reduces, which limits recovery activity and lost mileage. Workshop demand becomes more predictable, which improves utilisation and reduces disruption. Parts usage aligns more closely with expected demand, which lowers excess stock and emergency procurement. Safety performance stabilises as intervention happens earlier.

These effects sit across engineering, operations and finance, and they are often tracked separately, which is why the full impact is not always visible in one place. That is where the £6 million comes from. So, it isn’t a single initiative, but from decisions made earlier and applied more consistently across the operation.

If availability, cost per mile and performance are under pressure, we support transport operators in turning existing data and predictive capability into better operational outcomes.

Contact us: info@mindtheaigap.com | Call us free: 0800 009 6408