Case Study

Mobility · AI Fleet Planning

How We Built an AI Co-Planner That Schedules Coach Fleets and Keeps Planners in Charge

A Norway-based coach software company wanted daily trip planning to stop depending on hours of manual cross-checking. We built an auto-planning engine that assigns trips by proximity, timing, capacity and cost, then hands every proposal to the planner to confirm.

How We Built an AI Co-Planner That Schedules Coach Fleets and Keeps Planners in Charge

Project Snapshot

The Solution at a Glance

4 checks

Overlaps, passenger capacity, fleet rest periods and availability validated on every auto-plan

3 cost factors

Fixed daily, per-hour and per-kilometre costs weighed in each assignment

2 chaining rules

Closest garage for the first trip, shortest hop to the next, built to limit dead kilometres

1 final say

The planner confirms selected trips, confirms all or rejects; nothing is committed automatically

01

Business Context

The client is a Norway-based software company whose cloud platform helps coach operators run their day-to-day business, from customer bookings and planning to driver apps and live vehicle tracking. By its own account, its customers range from fleets of three coaches to operators running more than 200.

Planning is where that work comes together. Each day, someone has to decide which vehicle takes which job. That decision touches three groups:

In the platform’s visual planner, trips sit on a timeline and planners assign them to vehicles by hand. The client wanted the software to produce the first draft.

02

Challenges and What Was at Stake

Coach planning looks like a calendar problem. It is really a distance and cost problem. Every assignment decides how far a vehicle drives empty before it earns anything, and one poor choice early in the day can leave the wrong coach in the wrong place for the next job.

Six pressures shaped the work.

What was at stake was planner time and dead kilometres: the distance a coach covers without passengers, which uses fuel and driver hours and earns nothing. Driver hours are scarce. The IRU’s 2025 driver shortage report puts unfilled bus and coach driver positions at 7% in Germany and 8% in Spain, with coach tourism operators reporting higher rates.

Still planning fleet management manually?

Bring it to a 30-minute call. We will come back with a proposed approach and a scoped estimate.

03

Our Approach

Three decisions shaped the build.

04

What We Delivered

We delivered an auto-planning engine with a cost-optimisation layer and a planner-facing AI chat. It runs on Python, Django and Go services with React and Vue.js interfaces, PostgreSQL and MongoDB for data, AWS for hosting and Google Maps for travel times.

How one auto-plan runs

  1. SelectThe planner picks the trips to plan
  2. FilterEligible vehicles are found from filters and availability
  3. StartThe first trip goes to the vehicle with the shortest garage-to-pickup time
  4. ChainEach next trip minimises the gap from the last drop-off
  5. ValidateCapacity, overlaps and vehicle events are checked
  6. ProposeAssignments go to the planner to confirm or reject
I

Intelligent auto-assignment

The planner selects a set of trips. The engine identifies eligible vehicles from the active filters and availability, then matches each trip to the most feasible vehicle using distance, timing and capacity.

II

Proximity-first starts

The first trip of the day goes to the vehicle with the shortest travel time from its garage to the pickup point, so the day begins with as little empty running as possible.

III

Smart trip chaining

Each following trip is chosen to minimise the distance between where the previous trip ends and where the next one begins, linking jobs into an efficient sequence for each vehicle.

IV

Real-time feasibility checks

Before anything is proposed, the engine validates timeline overlaps, passenger capacity, fleet rest periods and vehicle events such as maintenance windows.

V

Cost optimisation engine

Administrators set a fixed daily cost, a cost per hour, and a cost per kilometre for each vehicle type. The engine calculates the operating cost of each candidate assignment and selects the lowest-cost feasible plan.

VI

Objective packages and planner control

From a panel beside the auto-plan button, planners choose a bundled objective package, for example one that minimises unassigned trips, maximises value and minimises cost. They then confirm selected trips, confirm all, or reject the proposal.

VII

Map view and AI chat

A map view shows planned trips visually, and an AI chat lets planners ask questions about vehicle selection while they review a plan.

How chaining limits dead kilometres

Garage
Shortest garage-to-pickup time
Trip 1Pickup to Drop-off
Shortest hop to the next pickup
Trip 2Pickup to Drop-off
Empty legs are the dashed lines. The engine keeps them as short as it can.

How each assignment is costed

Fixed costonce per vehicle per day
+
Trip duration × cost per hourdriver-related cost
+
Distance × cost per kilometrefuel, wear and maintenance
=
Assignment costcompared across feasible plans

Assignment cost is the sum of the three. With costs configured, the lowest-cost feasible plan is selected.

05

How AI Is Used in the Product

The AI in this project lives in the product and works for the planner. That is different from AI-assisted software development, which this case study does not cover.

It works at two levels.

Optimisation

The planning engine applies the chosen objectives, generates candidate solutions and selects the optimal plan, balancing cost, distance, time and value so the planner does not have to weigh them trip by trip.

Conversation

An AI chat gives planners real-time support by answering questions about vehicle selection. GPT-4 is part of the project’s technology stack.

In both cases, the output is a suggestion. A person reviews it before it becomes the plan.

06

Quality Built into Every Plan

A plan is only useful if planners can trust it, so the rules that protect plan quality are stricter for the engine than they are for a person.

These safeguards are part of the product itself. Together they mean the engine drafts, and a person decides.

07

Results and What Changed

The following section details what the co-planner now does in the planning workflow as well as the outcome the client has reported publicly.

Operational Outcomes

The first draft of a plan is now generated, not assembled. The proximity, chaining, capacity, overlap and availability checks that planners used to work through by hand are applied automatically to a selected batch of trips. Dead kilometres are addressed by design, through closest-garage starts and shortest-hop chaining. With costs configured, an economical plan can be told apart from a merely workable one, because each assignment carries a calculated operating cost.

Client-Reported Outcome

In its public launch material, the client reports that early-adopter operators have cut planning time from hours to minutes, and that the co-planner is in use with customers across Scandinavia and the UK.

Coach planning will always need a planner’s judgment. What changed is where that judgment is spent: on reviewing a validated, costed proposal instead of building one from scratch.

Ready to Take the Manual Work Out of Fleet Planning?

You might run a coach or bus fleet that is still planned manually. You might build transport software and want optimisation inside your own product. Or your planning rules might live in one experienced person’s head. In every case, the first step is the same.

Book a 30-minute call with Brain Station 23. Bring your current planning workflow, and we will come back with a proposed approach, a delivery plan, and a scoped estimate.

Still planning fleet management manually?

Bring it to a 30-minute call. We will come back with a proposed approach and a scoped estimate.

08

Related Case Studies