Unifying Transportation, Refinery, and Inventory Decisions with Mathematical Optimization

In large-scale energy operations, decisions about how to move, store, and process product are deeply interconnected, but they were often being made in isolation.

Every day, planners responsible for supply and operations execution had to balance an enormous number of moving parts: which routes to ship product along, how to schedule processing at each facility, how much inventory to hold at every location, and how to allocate constrained resources; all while respecting facility capacities, product output limits, quality specifications, and required inventory levels.

Because each of these domains was planned separately and largely by spreadsheet-based methods, locally reasonable choices routinely added up to globally expensive outcomes. The result was high logistics costs, suboptimal pipeline and refinery utilization, and frequent bottlenecks. What the organization lacked was not data or expertise, but a unified way to weigh every constraint at once and identify the most cost-effective, resilient plan across the whole network.

Aimpoint Digital began by partnering closely with the client's business and technology teams to understand the physical processes and data landscape across the enterprise, including how product actually flows, where constraints truly bind, and what data was available to drive decisions.

From that foundation, we designed and built a multi-tiered, multi-period mixed-integer optimization model that unifies the full supply chain in a single formulation. The model simultaneously optimizes network flow (transportation along lanes between locations), production and processing (the transformation of materials at facilities), inventory management, resource capacity, operational mode configurations, and product quality and composition limits. Its objective is to minimize total system cost, spanning transportation, processing, inventory holding, and deviations from planning targets, so that every recommendation reflects the true, end-to-end economics of the network.

The engine supports two types of planning: taking into account the entire supply chain at once, and drilling into the operations of a single facility. To make the solution fast enough for interactive, real-time use at enterprise scale, we engineered it with a two-phased feasibility verification procedure, allowing planners to run what-if scenarios and get answers in the flow of their daily work. Inputs flow directly from the planning system planners already use, and the solution was deployed robustly within the client's existing cloud infrastructure.

To keep results transparent, we also built a fast, AI-powered diagnostic layer. When a scenario has no feasible solution, it automatically pinpoints the minimal set of conflicting constraints and uses a large language model to translate that conflict into a plain-language explanation, telling planners exactly which input to adjust rather than returning a cryptic solver error.