Enterprise AI in Transportation: What’s Working, What Isn’t, and Where to Start

Enterprise AI is moving fast. Transportation is starting to catch up.

This post works through the key issues: why the gap between AI’s potential and what transportation businesses are actually achieving exists, what the integration challenge really looks like, and how to think about agents in an operational environment.

1. The Gap Between Tech and Transportation Operations

Anyone who has spent time both in Silicon Valley and in a transport control room will tell you the same thing: these are two different worlds, and the gap is not primarily about ambition or budget.

Engineering roles in tech have a set of advantages that make AI adoption almost automatic. The work is verifiable — you can check whether the code works. The tools are self-selected. When something breaks, the person using it can debug it. And the AI models themselves are exceptionally good at code.

Transportation operations are the inverse. Users are less technical. Data is fragmented across TMS, depot management, telematics, driver hours systems, maintenance platforms, and finance — often with no clean line between them. When something goes wrong operationally, the consequences are immediate and physical.

And it is why the AI tools that work effortlessly for a software startup do not simply transplant into an airport operation or a rail maintenance team.

The diffusion from what works in tech to what works in transport will take time — but the direction of travel is clear. And our customers in transportation adopting AI understand that.

2. Why Enterprise AI Keeps Failing in Transportation

The MIT statistic that “95% of enterprise AI efforts fail” gets quoted a lot. It is probably right, but it is also slightly misleading — because plenty of individuals inside transport businesses are already using AI tools effectively every day. 

The problem begins top down – atleast at the moment. The board demands AI. The CEO commissions something — a consultant, a vendor, a centralised pilot. The IT team runs the project. The operations team is consulted briefly and then left out of the loop. Nobody with genuine accountability for the outcome is driving it from the business unit side.

There is a second, underappreciated reason things stall: architectural paralysis.

Which tool / vendor / LLM / AI stack will still be the right bet in two years? Operations and IT teams have been burned before — they went down an AI path in 2021 or 2022 that is now deprecated, and they are not in a rush to do it again. That caution is rational. It is also, in many cases, causing organisations to miss genuinely useful near-term wins while waiting for a market that has not fully settled yet.

3. The Architectural Shift: Stop Adding Features, Start Thinking About Access

So far, the dominant approach to AI in transportation is to add AI features into existing products. A chat interface predominantly.

That model is giving way to something more fundamental. Instead of thinking of AI as software you integrate, think of it as a new type of user you need to give access to.

Can data from your systems be consumed by an agent the same way a human user consumes it. Can an AI agent log in, read the right records, query the right data, and take the right actions — with appropriate access controls — without needing a screen to click through?

This is a significant mental shift, and it has real consequences. It means the value of your existing systems is partly determined by how accessible they are to agents, not just to humans. 

For transportation businesses evaluating new technology or renewals: this is worth adding to your criteria. Not “does it have an AI feature?” but “can an agent use this system the way a human would, with appropriate access controls?”

4. The Integration Wall — and Why Agents Do Not Automatically Clear It

Agents hit the same integration walls humans do.

A dispatcher who needs to cross-reference a vehicle’s current location, its next scheduled maintenance, its driver’s remaining hours is navigating different systems — and probably knows which colleague to ring when one of those systems does not have the up-to-date information. That informal knowledge (“ask Gary what the actual position is in the yard”) is not documented anywhere. 

An AI agent given dispatcher-level access will encounter exactly the same walls. It will find that the maintenance record is in CMMS, the driver hours are in a separate compliance tool, the live vehicle position is in telematics. Unlike the dispatcher, it will not know to ring Gary. It will just stop, or — worse — proceed with the wrong information.

This is not an argument against agents. It is an argument for being honest about what needs to happen before agents can do useful work in your operation. You do not need to centralise your data or build a data lake. Transport businesses are already swimming in data. The question is much simpler: can an agent access the authoritative version of the information it needs, rather than a stale copy from a secondary system?

5. What Happens to Transport Jobs Once AI is Implemented?

Our take on this, backed by consistent historical precedent, is: transportation operations jobs are not going away.

IBM’s pitch in 1965 was that computers would eliminate accountants. There are more accountants now than at any point in history — because once you can process financial data at speed, the scope of what you can usefully do with it expands, and that expansion requires more people, not fewer. Computers made accounting more complex, more comprehensive, and more valuable. They did not replace accountants. They changed what accountants do.

The same pattern played out in law. Every lawyer you encounter today works in a computerised environment — citations from databases, documents in track changes, case research done digitally. The legal profession is substantially larger than it was before computers arrived. Automation changed the work, not the headcount.

Transport will follow the same trajectory. The planner who currently spends four hours a day compiling cross-fleet utilisation data will spend that time on the analysis and the decisions — because an agent will do the compilation. The coordinator who currently chases confirmation calls will be overseeing an agent that handles the routine confirmation loop and escalating only the exceptions. The maintenance engineer who currently reviews service histories one vehicle at a time will be managing an AI system that monitors every vehicle simultaneously and flags the ones that need attention.

These are better jobs. Not the end of jobs.

The concern about job displacement tends to come from a static view of what work is. The amount of useful work available expands with the capability of the tools available to do it. There is no serious historical evidence to suggest this moment is different.

6. Where to Start with AI in Transportation Projects?

The organisations getting this right share three characteristics.

1. They have a business unit sponsor — not a transformation team, not IT — who is accountable for the outcome.
2. They have identified which data sources agents actually need to access, and made sure those sources are authoritative and accessible.
3. And they have a human in the loop at the output stage, particularly early on, because the cost of a wrong AI-generated decision in transportation operations is real and immediate.

Start with read-only use cases.

Cross-fleet performance summaries. Pre-meeting customer account briefings. Anomaly detection across maintenance or utilisation data. These sit alongside your existing workflows without disrupting them, deliver visible value quickly, and build the trust — in the technology and in the team — that allows you to expand from there.

The acting agents — the ones that reschedule runs, adjust maintenance plans, reallocate resource — come later. When the data access is clean, the permissions are right, and the review process is established.

Frequently Asked Questions

Will AI replace engineers, planners, or operators in transportation?
No. The historical pattern is consistent — technology changes the nature of work, not the volume of it. Planners will shift from compiling data manually to directing agents and reviewing their outputs. That is a better job, not the end of one.

Why do most enterprise AI projects fail in transportation?
Usually because the project is driven from the top without a business unit sponsor who is accountable for the outcome. Pilots that sit with IT or a central transformation team rarely get adopted in operations.

Do I need to centralise my data or build a data lake before using AI in transportation?
No. Transportation businesses already have the data they need. The question is whether an agent can access the authoritative version of it — the live system, not a stale copy.

What are the best first AI use cases for a transportation operations?
Start with read-only use cases: cross-fleet performance summaries before the morning briefing, customer account overviews ahead of contract renewals, and anomaly detection across maintenance or utilisation data. These deliver visible value without touching core processes.

How do AI agents integrate with systems like a TMS or telematics platform?
Agents work best when they can access systems via API, the same way a human user would — but without a screen to click through. The key requirement is that the system has accessible data endpoints and that the agent has its own access credentials, not borrowed logins from a human user.

How much productivity improvement can a transport business realistically expect from AI?
In targeted workflows, 3–4x is a realistic and achievable number. Ten times is a headline figure, not one we’ve seen in transportation yet. The bottleneck is usually the review process — the time a human needs to check the output — not the AI itself.

What is the difference between adding AI features to existing software and using AI agents?
AI features are built into a product — a chat interface, an alert in your telematics dashboard. AI agents are separate entities that consume your existing systems the way a human user would, with their own credentials and a defined scope of action. Agents are more flexible and more capable, but they require more deliberate setup to deploy safely. Amygda has experience building and deploying agents into transportation operations.

How do you stop AI from making mistakes in a live transport operation?
Keep a human in the loop at the output stage, particularly early on. Start with read-only use cases where a wrong answer is informational rather than operational. As confidence in the data and the agent’s behaviour builds, you can extend to actions — but compliance-critical decisions around drivers’ hours and vehicle safety should retain human sign-off.