Ertas for Construction BOQ Extraction

The Challenge

Construction projects rely on Bills of Quantities (BOQ), the detailed lists of materials, labor, and equipment needed for a project, as the foundation for cost estimation, tendering, and project management. BOQs are typically embedded in complex documents: tender packages with hundreds of pages, engineering specifications with nested references, and price schedules in inconsistent formats from different contractors. Extracting structured BOQ data from these documents is a manual process that takes quantity surveyors days or weeks per project.

The challenge is compounded by the diversity of BOQ formats across different standards (NRM, SMM7, POMI), countries, and trades. A civil works BOQ uses different measurement units and item descriptions than an MEP (mechanical, electrical, plumbing) BOQ. Subcontractor quotes often use their own item numbering and description conventions that do not map directly to the project BOQ structure. Generic data extraction tools cannot handle this domain because construction terminology is highly specialized: terms like 'preliminaries,' 'provisional sums,' 'prime cost items,' and 'daywork rates' have specific meanings that determine how line items should be classified and priced.

The Solution

Ertas lets construction firms and quantity surveying practices fine-tune extraction models on their own BOQ formats, measurement standards and trade conventions. In Ertas Studio, teams train on annotated BOQ pages, each paired with the structured data a quantity surveyor extracted from it, so the model learns to identify line items, quantities, units, rates and specifications across the formats they actually receive.

The fine-tuned model learns the real-world messiness of BOQ text: multi-line item descriptions, nested sub-items, cross-references to specification clauses, and provisional sums that need special treatment. Exported from Hub and run on the firm's own servers, it takes the text of incoming tender documents (after the firm's usual OCR or PDF-to-text step) and returns structured BOQ data in the firm's standard format, ready to import into cost estimation software. For tender comparison, it maps different subcontractor formats onto a common structure for side-by-side review. Training runs on Ertas's cloud GPUs, so the annotated examples are uploaded to Ertas. Live tender pricing processed after export stays on the firm's infrastructure.

Key Features

Example Workflow

A quantity surveying firm processes 40 tender packages a month, each with 100 to 500 pages of BOQ documents across several trades. The team annotates 3,000 BOQ pages from completed projects with structured extractions (item numbers, descriptions, quantities, units and rates), leaves out projects still under confidentiality, and uploads the set to Ertas Studio. They fine-tune a model for their main trades of civil works, structural steel and MEP, run several base models in parallel, and compare each run's extractions against a held-out set of pages checked by a senior surveyor. The chosen model is exported as GGUF and served on an office server. When a new tender arrives, the firm runs its usual OCR step and sends the text to the model, which returns structured BOQ data in the firm's template. Quantity surveyors review the output, correct errors and handle items that need professional judgment, such as provisional sums and alternative specifications. Corrected extractions go back into Data Craft for the next training run, and the team tracks review time per tender against its manual baseline.