Dispatch decisions before coffee

By 6am the plan is already wrong. A driver called in sick at 5:40. A truck failed its walkaround check. A customer moved a delivery window by two hours, and the neat schedule built yesterday evening is now a suggestion. Dispatch is not a planning problem. It is a re-planning problem, and the hardest hour of it happens before coffee.

We build decision support for dispatch teams, and this piece is the argument for why support, not full automation, is the right target. The figures are illustrative of the pattern we see in mid-size fleets, not audited results.

The morning scramble, measured

Watch a dispatcher for an hour on a bad morning and the striking thing is how little of the hour is dispatching. Most of it is assembling the picture needed to dispatch: ringing drivers for positions, checking three portals for trailer status, reading the overnight emails, retyping a change into the transport management system, then discovering the change has already been overtaken by another one.

Six minutes of the hour is the job. The other fifty-four are the price of getting to the job. That ratio is the whole opportunity, and it says something important: the dispatcher was never the bottleneck. The picture was. A person with twenty years of route knowledge is being used as a human integration layer between a phone, a tracking portal, and a transport management system that do not talk to each other.

What the system has to see first

Before any of this becomes decision support, it is an integration job. The telematics feed with live positions and remaining driving time. The transport management system with the plan and the windows. The driver messaging channel, because the truth about a delay usually arrives as a text. The customer portals where windows get moved. None of this is exotic, and all of it already exists in a typical fleet. What does not exist is the single screen where it agrees with itself. Building that screen is most of the project, and it pays for itself before a single recommendation is made, because the twenty-two minutes of chasing collapses the day the picture assembles itself.

Exceptions go to people, with options attached

The routine reshuffles are mechanical. A driver is out, so the system re-sequences the runs it can re-sequence within hours rules and windows, and shows its working. What it should not do is quietly commit anything that touches a promised window, a key account, or a driver's legal hours limit. Those go to the dispatcher as an exception, framed as a choice: here are three feasible reshuffles, here is what each one costs in late deliveries and overtime, here is the one we would pick and why. The person decides in thirty seconds instead of rebuilding the puzzle from scratch.

This framing matters because dispatch knowledge is deeply local. The system does not know that the receiving bay at one site is chaos before 8am, or that a particular customer shouts but always accepts a late slot, or that a particular driver knows the awkward site and the new driver does not. The dispatcher knows. Decision support keeps that knowledge in the loop instead of pretending it can be replaced by a distance matrix.

The confidence threshold deserves the same honesty as everything else. Start conservative. In the first weeks the system proposes and a person approves everything, which is slower but builds the evidence for where the system's judgement can be trusted. Widen its autonomy run type by run type, with the exception log as the referee, not on a date someone put in a project plan.

The paperwork that keeps a fleet legal

Underneath the scheduling sits the compliance layer, and it is relentless. Driver hours and tachograph records that must reconcile with the plan you actually ran, not the one you wrote. CMR consignment notes for cross-border loads. Dangerous goods declarations where they apply. Vehicle maintenance and defect reports with dates that inspectors read carefully. Proof of delivery that has to make it back from a cab to an invoice. None of this is optional, all of it is deadline-bound, and much of it is still assembled by the same person doing the 6am scramble.

This is quietly the best first target in most fleets. Drafting the compliance pack from data the telematics and transport systems already hold is lower risk than touching live dispatch, it removes the afternoon's drudgery rather than the morning's judgement, and it builds trust in the data before that data starts proposing reshuffles.

Why decision support beats full automation here

• The inputs are incomplete and late. Traffic, a driver's text, a gate delay. A person integrates missing information better than an optimiser that assumes its inputs are true.
• The costs are asymmetric. A slightly suboptimal route costs pennies. An automated decision that strands a key account's delivery costs the account. Keep the person on the asymmetric calls.
• The knowledge is local. Sites, drivers, and customers have quirks that live in the dispatcher's head, not in any dataset you currently have.
• Accountability is human. When the customer calls, someone has to own the decision, and owning a decision you made beats defending one you did not.

Full automation also fails worst on exactly the mornings that matter most. The clean Tuesday needs no help. The morning with a sick driver, a failed truck, and a moved window is the one where an automated planner either freezes or does something confidently strange. Decision support degrades gracefully: on the worst morning it is still doing the arithmetic on hours, windows, and options while the human does the judgement. That division holds up.

There is also a quieter benefit. When the reshuffles are logged with their options and their reasons, the fleet gets a record of its own decisions for the first time. Why runs get resequenced, which customers move windows most, which sites eat the mornings. That record is what makes next season's planning better, and no amount of heroic 6am improvisation ever produced it.

The dispatcher was never the bottleneck. The picture was. Give one person a complete picture and the decisions take minutes.