AI Around Optimization: Better Inputs, Decisions and Learning Loops

Alex Bokov argues that the most practical role for AI in supply chain planning is not to replace planners or established optimization engines. Instead, AI can improve the data entering the plan, help users interpret planning outcomes and capture learning from previous decisions.

At the center of the architecture, mathematical and constraint-based planning remain responsible for generating feasible plans. Poor results often originate in weak master data, missing business rules, unrealistic capacity assumptions or misaligned parameters rather than shortcomings in the optimization method itself.

AI can add value before a planning run by detecting abnormal lead times, capacity assumptions and inconsistencies between master data and execution history. After the run, it can translate constraints, bottlenecks and trade-offs into clearer business language, supporting faster decision support.

The most important longer-term opportunity is the feedback loop. Accepted and rejected recommendations, planner overrides, inaccurate assumptions and execution outcomes can become governed inputs for future cycles. This requires explicit ownership of planning rules, validation criteria, version control and a manual override process. Without those controls, the system risks learning from inconsistent or poorly documented decisions.

The resulting model is AI around optimization: established planning logic generates the plan, AI improves inputs and interpretation, and planners retain accountability for scenarios, exceptions and operational trade-offs.