VertOS MCP: AI in Ground Transport Operations

Artificial intelligence is everywhere in software right now, but in transport operations, most implementations still stop at the surface. They can write an email, summarise a document or answer a question, but they often remain one step removed from the business they are supposed to help.

For the past three months, we have been using AI differently inside VertOS. We have connected VertOS directly with models from Claude and OpenAI, giving them controlled access to live operational information and, where appropriate, the ability to carry out actions inside the platform.

This is done through a custom MCP, or Model Context Protocol. Put simply, it is the secure bridge that allows AI to work with approved data and tools inside VertOS rather than waiting for somebody to manually feed information into it.

It changed how quickly we could act, but it also changed how much of the business we could see.

We Already Had the Data

A passenger transport operation generates huge amounts of raw data every day.

Bookings, prices, supplier costs, chauffeur payments, flight movements, amendments, cancellations, waiting time, vehicle availability, client activity, payments and dozens of other variables are constantly being recorded.

At our scale, that matters. Vert has now handled tens of thousands of rides, works with thousands of chauffeurs globally and maintains a near perfect service rating.

That produces a substantial operational footprint. The problem was never whether the data existed. The problem was extracting useful information from it.

Which routes are becoming less profitable? Where are supplier costs creeping upwards? Which vehicle classes repeatedly run short? Which completed journeys do not reconcile correctly? Is a client growing in revenue while declining in margin?

Those answers may already exist inside the system, but traditional software usually expects the person using it to know where to look.

With direct AI access, instead of beginning with the report, filter or spreadsheet, we can begin with the question.

From Reporting the Business to Interrogating It

A dashboard is designed around questions somebody anticipated in advance.

Revenue this month. Number of bookings. Driver utilisation. Outstanding payments.

Useful, but fixed.

What has changed for us is the ability to investigate questions that were never built into a dashboard in the first place.

If margin falls, we can look across revenue, supplier costs, waiting time, cancellations, vehicle utilisation and client mix together. If a cost line begins moving in the wrong direction, we can investigate what sits behind it rather than just acknowledging the figure.

The distinction is important.

Traditional business software tells you what happened.

Connecting AI to the underlying operation helps us ask why it happened, what is connected to it and what deserves attention next.

We have found that particularly useful because some of the most expensive problems in a business are not dramatic enough to announce themselves.

A supplier cost increasing gradually.

A client sending more work but worse work.

Owned capacity sitting unused while third-party supply is being purchased elsewhere.

A recurring shortage at the same time every week.

Individually, none may look serious.

Viewed together, they can tell a very different story.

AI Has Become Another Pair of Eyes on the Operation

Operational businesses are naturally good at noticing loud problems.

A late passenger pickup demands attention. A complaint arrives. An invoice is overdue. A chauffeur becomes unavailable. Someone deals with it.

The quieter problems are easier to miss because nobody is necessarily looking for them.

That is where the MCP has become more valuable to us than we initially expected.

Rather than using AI only to retrieve something we already know we need, we can use it to examine relationships across the operation and surface patterns that deserve investigation.

It’s not replacing management judgement. It is widening the field of view.

A person running a business can only hold so many variables in their head at once.

The system does not have that limitation.

Dispatch Proved the Model

Dispatch was the obvious starting point because the result is immediate.

There are a lot of moving parts and changing variables: flight delays, booking changes, chauffeur cancellations and more.

Resolving the issue often requires context from several different parts of the operation before anyone can decide what happens next.

Through the MCP, we can bring more of that context together in one interaction.

Relevant booking information, available supply and surrounding operational data can be surfaced together and, where authorised, the next action can be taken from there.

That cuts out a lot of unnecessary movement between screens.

But dispatch ultimately proved something bigger.

If AI can understand the context around one live booking, the same connection can be used to understand patterns across thousands of them.

That distinction is where the business value starts to compound.

The useful question stops being:

“How do we cover this job?”

and becomes:

“Why do we keep having this problem?”

Finance Became Less About Finding the Needle in the Haystack

Passenger transport finance is rarely as clean as the original booking suggests.

A client pays one amount. A chauffeur or supplier receives another. Parking, waiting time, amendments, cancellations and commissions change. The list goes on.

Reconciliation is where the planned version of the journey meets the outcome of payment.

Before, investigating something unusual could mean working through individual bookings or rows until the discrepancy appeared.

We can use VertOS and AI together to surface journeys that do not reconcile as expected, inspect unusual cost movements, compare expected against actual payments or understand where outgoings have changed.

As we work with a larger variety of job providers and suppliers, this has become increasingly important to automate and cut total manual input.

The simple unit economics of this has saved us a pretty penny.

Instead of searching through the haystack for the needle, we can ask the system to help identify where the needle is likely to be.

Coverage Became Something We Could Analyse, Not Just Solve

We have seen the same shift around capacity.

Being short of a Business Class chauffeur tomorrow morning is an operational problem.

Somebody needs to find a car for the job.

Being short of Business Class chauffeurs every Tuesday morning is a business signal.

That requires recruitment, fleet planning, supplier relationships, pricing or understanding where additional capacity needs to be built.

VertOS already connects supply and demand through the VertOS Network, allowing operators to send work they cannot fulfil and receive work where they have capacity, including between different systems without manually re-entering bookings.

AI helps us reveal where pressure repeatedly builds, where we rely too heavily on external supply or where spare capacity could be turned into additional demand.

One solves the fire.

The other asks why that corner of the building keeps catching fire.

The Biggest Change Is How We Use the Software

After three months, the most important outcome is not one AI feature.

It is that the starting point has changed.

Traditionally, using operational software looks something like:

Find the screen → find the data → interpret it → decide → act.

Increasingly, we can start with the decision, question or problem itself.

The system can retrieve the relevant context, compare it, organise it into something useful and, where permitted, help take the next step.

That matters because the same booking means something different depending on who is looking at it.

To dispatch, it needs fulfilling.

To finance, it is revenue, cost and reconciliation.

To fleet management, it consumes capacity.

To us as business operators, it contributes to a much bigger picture of margin, growth, utilisation and risk.

Building It Inside a Real Operation Matters

MCP is an open standard.

Building something we are willing to trust inside a live passenger transport operation is another matter.

VertOS deals with passengers, commercial information, bookings, chauffeurs, payments and actions with real-world consequences.

Permissions matter. Validation matters. Authentication matters.

Some actions can happen directly, while others should require confirmation.

That is why we developed and tested our MCP inside our own operation first.

For the prior seven months before introducing it to our live operations, we had been testing it against a large dummy backup of our dataset and running it through various simulations to stress test it.

Now we have successfully proven over the past three months that, with more exposure to real-world use and pressure, it has had a measurable impact on how we conduct our operations and has been tested against real edge cases.

That matters because there is a difference between an impressive AI demonstration and something genuinely useful on a Monday morning when the operation is live.

What This Has Shown Us

For us, it has meant easier investigation of financial discrepancies, better understanding of capacity, less friction in dispatch and a greater ability to uncover patterns that may previously have remained buried inside the operation.

There will still be dashboards and spreadsheets, and people will still make the most crucial decisions.

But after using this inside VertOS, we no longer see AI as another feature sitting on top of the software.

We see it as a whole new landscape for growing our operations and business relationships.

Want to See How This Works in Practice?

We are already using VertOS and its AI capabilities inside our own operation, and we are now opening that infrastructure up to more of the industry.

If you are a passenger transport operator, aggregator, technology provider or potential partner, we would like to hear from you.

Whether you want to:

• see a demonstration of VertOS

• connect your operation to the VertOS Network

• send or receive work through the network

• explore an integration

• discuss how AI could work across your own operation

Book a VertOS demo and we will show you how VertOS works in a live operating environment.