From FMS Data to Operational Intelligence: AI Agents for Haulage Performance

Why did truck productivity decrease during the last shift? A Fleet Management System may contain much of the required information, but the answer is rarely captured by a single KPI. Cycle time, queue time, loading time, travel time, payload, equipment availability, dispatch assignments, delays, shovel utilization, road conditions, and mine-plan execution can interact to create the observed result.

An AI-enabled operational intelligence workflow can help mine operations investigate those relationships across FMS and supporting systems, while keeping dispatchers, supervisors, engineers, and operational leaders responsible for validation and action.

Start with the operational decision

OPEIM would first define the question the operation wants to answer. A practical example is: What were the main contributors to the truck-productivity loss during the night shift, and where should the morning team focus its investigation?

The required KPIs, data sources, business rules, thresholds, and operational context are then defined around that question.

More than FMS data may be required

Depending on the mine, the investigation may connect information from:

Fleet Management System / Dispatch. Assignments, cycle components, queue time, hang time, spot time, loading time, travel time, delays, equipment states, and dispatch parameters.

Payload systems. Load distribution, underloading, overloading, payload variability, and data-quality status.

Maintenance / Asset Management. Availability, downtime, equipment events, work orders, and recurring asset constraints.

Mine planning and short-interval control. Planned versus actual movement, mining areas, routes, destinations, and plan-execution gaps.

Operational context. Road works, weather, shift changes, congestion, temporary restrictions, and supervisor or dispatcher observations where those data are governed and available.

How the AI Agent investigates

The agent can compare the shift against plan, recent baseline, and operating targets, identify the KPIs with the largest deviations, and then drill into the equipment, route, shovel, or assignment patterns that explain the result.

A mine manager might ask: Why did haulage productivity finish below plan last night?

A structured response could identify, for example:

Production gap. Material movement finished below the shift plan.

Primary contributors. Truck availability was below target, cycle time increased on a specific route, and queue time increased at one loading unit.

Secondary observation. Payload remained within the expected operating range and was not a material contributor to the loss.

Recommended focus. Review the availability losses on the highest-impact trucks, investigate the route-time increase, and review queue and assignment behavior at the affected shovel.

Why this is different from another dashboard

A dashboard shows the operating condition. The AI-enabled workflow is designed to help investigate why that condition occurred by moving across related KPIs and authorized systems. The agent can also maintain a consistent investigation sequence so the team does not need to manually repeat the same search every morning.

Human validation remains essential

The agent may identify that travel time increased, but an operational supervisor may know that the cause was temporary road maintenance. That context determines the correct action. The technology therefore supports the decision process rather than replacing operational accountability.

The resulting loop is: Detect → Understand → Validate → Act → Measure → Improve.

The longer-term opportunity

Once the workflow is proven, the same architecture can be extended to other FMS questions: loading-hauling balance, payload improvement, queue management, equipment-time usage, mine-plan compliance, short-interval control, and recurring performance opportunities. This creates a scalable operational-intelligence capability rather than a one-off chatbot.