Semantic Conversational AI for Construction Cost Analytics

Abstract

Construction companies generate large volumes of project data. Costs, labor hours, equipment usage, and productivity records, yet this data remains under-utilized due to inconsistent activity descriptions and spreadsheet-dependent workflows. We present a semantic conversational analytics framework powered by GPT-4 via a Microsoft Teams bot, combining fuzzy string matching for cost code identification with a deterministic Python analytics backend. Raw records are exported from Heavy Job into Azure Blob Storage; computed output files are written back to the same store. Evaluated against Microsoft Copilot Studio across 50 test queries, the system achieved 48 of 50 formal pass/fail trials (93%). Results demonstrate that semantic constraints and execution control are architectural pre-requisites for reliable enterprise conversational analytics.