RFI Pre-emption: Detecting Drawing-Specification Conflicts at Ingestion, Before the Field Finds Them

Requests for Information (RFIs) are a pervasive and costly aspect of construction projects, frequently stemming from inconsistencies and contradictions embedded within project documentation—drawings, specifications, and schedules. This paper introduces RFI Pre-emption, a novel, ingestion-time conflict detection system designed to identify and flag these discrepancies proactively, before they escalate into field-level issues. The proposed architecture leverages advanced document AI techniques to extract typed claims from diverse project documents, resolve element identity across disparate document families, and join these claims by element-attribute. By normalizing units and tolerances, the system deterministically flags incompatibilities and emits structured conflict records with dual-source citations. These conflicts are then ranked by their potential trade and schedule impact, enabling project teams to address critical issues early. The system is designed for incremental re-execution upon document revisions, ensuring continuous conflict detection. This proactive approach aims to significantly reduce RFI volume, mitigate costly delays, and improve overall project efficiency and predictability.

Construction Tech, Document AI, Knowledge Graphs, Conflict Detection, RFI Management, Cross-Document Reasoning, Entity Resolution, Semantic Conflict Detection, Proactive Project Management

The construction industry is notoriously complex, characterized by fragmented information, diverse stakeholders, and tight deadlines. A significant source of inefficiency and cost overrun in construction projects is the Request for Information (RFI) [1]. RFIs are formal queries raised by contractors to clarify ambiguities, resolve discrepancies, or obtain missing information from design teams or owners. While essential for project execution, the sheer volume and late discovery of RFIs often lead to project delays, cost escalations, and rework [2].

Traditional RFI management systems primarily focus on optimizing the workflow *after* an RFI has been submitted, aiming to expedite responses and track resolutions. However, this reactive approach fails to address the root cause: many RFIs arise from contradictions that are already present, but undiscovered, within the vast and often disparate project documentation—architectural drawings, structural plans, mechanical specifications, electrical schematics, and various textual documents [3]. These inconsistencies, if not identified early, can lead to critical issues during construction, such as material procurement errors, on-site clashes, and significant schedule impacts [4].

This paper proposes RFI Pre-emption, a novel system designed to proactively detect and highlight these cross-document conflicts at the earliest possible stage: document ingestion. By shifting conflict detection from a reactive, field-driven process to a proactive, AI-driven analysis of project documentation, RFI Pre-emption aims to significantly reduce the incidence of costly field-generated RFIs, thereby improving project predictability, efficiency, and overall quality.

2.1. RFI Management and Impact in Construction

RFIs are a well-documented challenge in construction. Studies by organizations like the Navigant Construction Forum have consistently highlighted the substantial costs associated with RFIs, estimating an average cost of over $1,000 to process a single RFI, including review and response time [5]. Furthermore, a significant percentage of RFIs are deemed unjustifiable, indicating that they could have been avoided with better upfront coordination or document quality [5]. The timing of RFI discovery is critical; conflicts found during construction lead to idle labor, delays, and expensive rework, whereas those identified during design or pre-construction phases are significantly less costly to resolve [2]. This underscores the need for proactive conflict detection rather than reactive management.

2.2. Cross-Document Reasoning and Entity Resolution

The ability to analyze, connect, and synthesize information across multiple, disparate documents is known as cross-document reasoning [6]. This capability is crucial in complex domains like construction, where project information is distributed across numerous document types (e.g., CAD drawings, BIM models, textual specifications, schedules, contracts). Key techniques enabling cross-document reasoning include entity linking (connecting references to the same real-world object across documents, even if named differently), coreference resolution (identifying when different terms or pronouns refer to the same entity), and contradiction detection (flagging conflicting claims) [6].

Entity resolution is a foundational component of cross-document reasoning, aiming to identify and link records that refer to the same real-world entity across different data sources [7]. In construction, this means recognizing that 'Wall Type A' in an architectural drawing refers to the same physical element as 'Partition System A' in a specification document, despite variations in nomenclature. Challenges in this area include handling synonyms, abbreviations, implicit references, and variations in data representation across different document formats [8]. Recent advancements in Document AI and knowledge graph technologies offer promising avenues for robust entity resolution and semantic linking in complex document sets [9].

2.3. AI for Conflict Detection in Construction

Automated conflict detection has traditionally been a feature of Building Information Modeling (BIM) software, primarily focusing on geometric clashes between 3D model elements (e.g., a pipe clashing with a beam) [10]. While effective for spatial coordination, BIM-based clash detection does not typically extend to semantic conflicts embedded in textual specifications or discrepancies between textual requirements and graphical representations. Emerging AI approaches, particularly those leveraging Natural Language Processing (NLP) and knowledge graphs, are beginning to address this gap by extracting structured information from unstructured text and identifying logical inconsistencies [11]. However, a comprehensive system that integrates multi-modal document analysis (drawings, text), robust entity resolution, and proactive, ingestion-time conflict detection across the entire project documentation lifecycle remains an active area of research.

RFI Pre-emption operates as an ingestion-time conflict detection system, processing all project documentation to construct a unified knowledge representation and identify inconsistencies. The methodology comprises several sequential and iterative steps:

3.1. Document Ingestion and Claim Extraction

All project documents—including architectural, structural, MEP drawings (in formats like PDF, DWG), textual specifications (PDF, DOCX), schedules (XLSX), and any revisions or addenda—are ingested into the system. Advanced Document AI techniques, combining Optical Character Recognition (OCR), Layout Analysis, and Natural Language Processing (NLP), are employed to extract structured claims. A claim is defined as a specific, verifiable piece of information about a building element or its attribute (e.g., 'Wall W1 has fire rating 2-hour', 'Pipe P1 diameter is 4 inches', 'Concrete strength is 4000 psi'). For drawings, this involves interpreting annotations, dimensions, and symbols. For specifications, it involves parsing technical requirements and material properties. The output of this stage is a collection of structured triples or quads (e.g., `(Element, Attribute, Value, SourceDocument)`).

3.2. Cross-Document Entity Resolution

This is a critical step to establish a consistent understanding of project elements across all documents. The system employs entity resolution algorithms to identify and link references to the same physical or logical building element, even when they are named differently or referred to implicitly across various document types. For example, 'Wall Type A' in an architectural drawing, 'W-A' in a structural drawing, and 'Partition System A' in a specification must be recognized as referring to the same underlying entity. This involves:

• Lexical Matching: Identifying exact or near-exact string matches, accounting for abbreviations and common aliases.
• Contextual Matching: Using surrounding text, document type, and spatial proximity (for drawings) to infer relationships.
• Knowledge Graph Integration: Building and leveraging a project-specific knowledge graph where entities (e.g., walls, doors, pipes) and their relationships are explicitly defined. This graph serves as a canonical reference for element identities.

3.3. Claim Normalization and Joining

Once entities are resolved, the extracted claims are normalized to a common schema and unit system. This involves:

• Unit Conversion: Standardizing units (e.g., converting imperial to metric or vice-versa where appropriate).
• Tolerance Handling: Incorporating specified tolerances into attribute values to allow for acceptable variations.
• Semantic Alignment: Mapping diverse terminologies to a unified ontology of building attributes (e.g., 'fire resistance rating' and 'fire rating' are aligned to the same attribute).

Normalized claims are then joined by element-attribute. For instance, all claims pertaining to the 'fire rating' attribute of 'Wall W1' from all ingested documents are grouped together.

3.4. Conflict Detection and Ranking

Within each group of joined claims for a specific element-attribute, the system performs conflict detection. A conflict is identified when two or more claims for the same attribute of the same element present incompatible values (e.g., 'Wall W1 fire rating: 2-hour' from Drawing A vs. 'Wall W1 fire rating: 1-hour' from Specification B). The system emits structured conflict records, each detailing:

• The conflicting element and attribute.
• The incompatible values.
• Dual-source citations (identifying the exact documents, pages, and even specific locations within documents where the conflicting claims originate).
• A timestamp of detection.

Conflicts are then ranked based on their potential impact on project cost, schedule, and safety. This ranking can be informed by predefined rules (e.g., structural conflicts are high impact) or by machine learning models trained on historical project data. Factors considered include the trade involved, the stage of construction, and the criticality of the element.

3.5. Incremental Re-execution

Construction projects are dynamic, with frequent revisions and addenda to documents. RFI Pre-emption is designed for incremental re-execution. When new document versions or addenda are ingested, the system efficiently re-processes only the changed information, updates the knowledge graph, and re-runs conflict detection. This ensures that new inconsistencies introduced by revisions are promptly identified.

RFI Pre-emption is a powerful tool for identifying and citing contradictions, but it operates within clearly defined boundaries. The system's primary function is to detect and highlight inconsistencies in project documentation. It does not auto-decide design intent or governing authority. The resolution of identified conflicts—determining which claim takes precedence or requires modification—remains the responsibility of human project stakeholders (e.g., architects, engineers, contractors). The system provides the necessary evidence (dual-source citations) to facilitate informed decision-making, but it does not automate the decision itself. This clear separation ensures that human expertise and contractual agreements retain ultimate authority over design and construction decisions.

To validate the effectiveness and quantify the benefits of RFI Pre-emption, a comprehensive evaluation plan is proposed:

• Retrospective Replay Against Historic RFI Logs: The system will be applied to a dataset of completed construction projects, using their initial documentation sets. The conflicts identified by RFI Pre-emption will be compared against the actual RFI logs from those projects to determine how many of the field-generated RFIs could have been pre-empted. Metrics will include the percentage of RFIs detected, the average lead time of detection, and the types of conflicts identified.
• Precision, Recall, and Reviewer Effort Metrics: For a subset of projects, human experts will manually review the documents to identify conflicts. This ground truth will be used to calculate the precision (proportion of flagged conflicts that are real) and recall (proportion of real conflicts that are flagged) of RFI Pre-emption. Additionally, the effort required by human reviewers to validate and resolve conflicts identified by the system will be measured and compared against traditional RFI processing times.
• Live Pilot Tracking RFI-per-$1M Trend Versus Portfolio Baselines: A live pilot will be conducted on ongoing construction projects. The key performance indicator will be the RFI-per-$1M construction cost metric, tracked for projects utilizing RFI Pre-emption and compared against historical baselines and similar projects within the same portfolio that do not use the system. This will provide a real-world measure of the system's impact on RFI volume and associated costs.

Despite its potential, RFI Pre-emption is subject to several limitations that warrant consideration:

• Extraction Quality Varies by Document Fidelity: The accuracy of claim extraction is highly dependent on the quality and format of the input documents. Poorly scanned drawings, low-resolution PDFs, or inconsistent drafting standards can lead to errors in OCR, layout analysis, and NLP, consequently affecting the reliability of extracted claims.
• Entity-Resolution Errors: While advanced, entity resolution algorithms are not infallible. Ambiguous naming conventions, lack of explicit identifiers, or complex relationships between elements can lead to false positives (incorrectly linking distinct entities) or false negatives (failing to link identical entities), which can propagate errors through the conflict detection process.
• Tolerance Tuning is Domain- and Trade-Sensitive: Defining appropriate tolerances for numerical attributes (e.g., acceptable variations in dimensions or material properties) requires domain expertise and can vary significantly between different construction trades and project types. Mis-tuned tolerances can lead to an excessive number of false positives or, conversely, miss critical conflicts.
• Value Depends on Completeness/Timeliness of Document Sets: The system's effectiveness is directly tied to the completeness and timeliness of the project documentation provided. If critical documents are missing, outdated, or ingested late, the system's ability to detect conflicts comprehensively will be compromised.
• Interpretation of Design Intent: RFI Pre-emption identifies factual contradictions but does not interpret design intent. Resolving a conflict often requires understanding the designer's original vision, which is a human cognitive task beyond the scope of automated detection.

1. 1.

   DPR Construction. (2026, February 18). Q1 2026 Market Conditions Report. Retrieved from dpr.com

2. 2.

   Deloitte. (2025, November 13). 2026 Engineering and Construction Industry Outlook. Retrieved from deloitte.com

3. 3.

   Skanska. (2026, February 10). 2026 Winter Construction Market Trends. Retrieved from interactive.usa.skanska.com

4. 4.

   Navigant Construction Forum RFI analyses. (2013, April). Impact & Control of RFIs on Construction Projects. Retrieved from cmaanet.org

5. 5.

   Gordon, E., Rawlinson, K., & Dabhade, N. (2023). Assessing Quality Performance Outcomes & the Relationship With RFI and PCI Processes: A General Contractor Case Study. International Group for Lean Construction (IGLC) Conference Proceedings, 2023. academia.edu

6. 6.

   Frontiers in Built Environment. (2026). Improving construction efficiency through lean techniques and digital tools: case of a real-time implementation in an institutional building construction. Retrieved from frontiersin.org

7. 7.

   Cross-Document Reasoning. (2026, June 23). LlamaIndex Glossary. Retrieved from llamaindex.ai

8. 8.

   Navigant Construction Forum RFI analyses. (2013, April). Impact & Control of RFIs on Construction Projects. Retrieved from cmaanet.org

9. 9.

   Cross-Document Reasoning. (2026, June 23). LlamaIndex Glossary. Retrieved from llamaindex.ai

10. 10.

    What Is Entity Resolution? How It Works & Why It Matters. (n.d.). Senzing. Retrieved from senzing.com

11. 11.

    Cross Document Co-Reference. (n.d.). Intel Labs. Retrieved from intellabs.github.io

12. 12.

    DocSage: An Information Structuring Agent for Multi-Doc Multi-Entity Question Answering. (2026). arXiv preprint arXiv:2603.11798. [Reference Needed]

13. 13.

    Clash-of-Leges: A bilingual dataset for conflict detection and ... (n.d.). ScienceDirect. Retrieved from sciencedirect.com

14. 14.

    Evidence-Constrained Document Intelligence for Traceable Green-Attribute Review in Power Equipment Procurement. (n.d.). IEEE Xplore. Retrieved from ieeexplore.ieee.org