Cost estimation is a core function of construction planning, yet it remains one of the most time-consuming and fragmented tasks in the industry. Even with better tools and more data, estimators deal with recurring challenges like inconsistent quantity formats, scattered historical data, and manual subcontractor reviews. These challenges slow down the process and increase the risk of errors, especially early in a project when decisions matter most. This study maps how estimation is actually done in practice, pinpoints where the main burdens occur, and proposes a generative AI-based framework to help streamline and support the process. Using a qualitative approach, ten semi-structured interviews were conducted with industry experts, including estimators, BIM/VDC managers, data analysts, and project managers. Thematic analysis of these interviews led to the identification of key burden areas, which were grouped into three categories: conceptual estimation, subcontractor evaluation, and change management. The results show that current workflows are heavily dependent on manual data handling, subjective judgment, and disconnected tools. In response, the study proposes a targeted LLMs-enabled framework designed to assist with quantity standardization, historical cost referencing, subcontractor bid evaluation, and version control.