Understanding LLM Limitations in Tactical Mission Planning

Outlet Title

Social Science Research Network

Document Type

Article

Publication Date

12-6-2025

Abstract

While AI tools have been integrated at the general staff level of the US Department of Defense, a significant technological gap persists at lower tactical echelons, where automated systems remain challenging to implement at scale. One area ripe for automation is the Military Decision-Making Process (MDMP), a systematic, iterative approach designed to analyze complex environments and formulate operational plans. In 2024, the U.S. Government explicitly recognized this opportunity by soliciting project proposals via the Small Business Innovation Research program, seeking technological solutions to automating the Course of Action (COA) development step of the MDMP for battalion-level planning. This paper investigates whether and how one commercially available AI tool, Palantir's AIP platform, could be tailored to effectively automate COA development. Developing this proof of concept involved defining operational requirements, fine-tuning an AI model on operational manuals and military documents, and evaluating its performance against real-world tactical scenarios drafted by senior military professionals. While the application demonstrates potential in generating effective operational plans for specific scenarios, it also exposes critical limitations inherent in current LLMs, including data interpretation issues, systemic integration challenges, and inconsistent outputs — most notably, the lack of robust visual graphic integration. These findings suggest a cautious approach to deploying commercial AI planning tools in the MDMP context. Future research should focus on domain-specific model training and visual data integration to enhance AI effectiveness in tactical environments.

Link to Full Text

Share

COinS