ARIZ-85C is a systematic algorithm for inventive problem solving, yet its complexity and need for expert facilitation have limited adoption. Transformer-based large language models (LLMs) offer structured reasoning and retrieval capabilities that may automate ARIZ substeps. In this work, we integrate a preliminary patent-context extraction stage, to guide LLM-driven generation of inventive solution concepts. We compare three modes—baseline, baseline_ariz, and simple_ariz—across twelve runs on a problem statement. Solutions are evaluated qualitatively on inventiveness, novelty, and feasibility, and token usage per run is recorded to assess cost-quality trade-offs. Results indicate that explicit ARIZ steps combined with patent context yield the most inventive and novel solutions, though at higher token cost. Implicit ARIZ—where the LLM is simply instructed to apply ARIZ logic without explicit substep outputs—underperforms explicit ARIZ even with context. Practical guidelines are provided to balance solution quality and token efficiency.