Vision
Accelerating Project Delivery with Smarter Estimation
A multi-billion-dollar infrastructure and construction enterprise set out to explore how AI could transform the operational workflows that are central to their business, including job estimation, safety programs, risk management, and proposal response. Operating as a highly decentralized organization with more than 270 independently run operating companies, they recognized both the opportunity and the complexity of adopting AI across environments with different tools, norms, levels of digital maturity, and risk tolerances. Their goal was to demonstrate clear business value within a high-impact operating company while also establishing an enterprise governance framework capable of safely scaling AI across a safety-critical, SOX-impacted, federated organization.
Challenge
Proving AI Value While Building Trust at Scale
The organization’s highly decentralized structure meant that much of the operational expertise, particularly in job estimation, had been built over decades and varied widely across operating companies, teams, and individual estimators. This created inconsistent tools, workflows, document formats, and expectations, all of which made it difficult to introduce AI in a way that was both accurate and adoptable.
They faced several obstacles:
- Estimators spent one to two hours building each bid (often processing 10–12 bids per day) relying on deep, experience-based judgment and working with highly variable, unstructured documents, making automation extremely challenging.
- The organization needed to show measurable, operational value through a high-impact use case (not just experimentation) while also responding to growing pressure to establish responsible, auditable AI practices.
- Governance needed to function across more than 270 independent operating companies, balancing strict safety and compliance requirements with the need to support innovation and avoid adding friction to existing workflows.
- Any new AI workflows required systems capable of interpreting technical, unstructured documents and integrating with diverse identity, security, and enterprise environments across the federated organization.
Breakthrough
Production AI Delivered, Ready to Scale
The engagement combined AI engineering within a key operating company and the development of an enterprise AI governance framework. The estimation workstream modernized a complex, manual process that required interpretation of diverse project documents. In parallel, the governance workstream defined the principles, guardrails, and operating model needed to guide responsible AI adoption across the broader organization. Together, these efforts delivered tangible value while creating the foundation for future AI expansion.
Automated Estimation System
KUNGFU.AI delivered a production-grade automated bid estimation platform that processes project documents through OCR and AI pipelines to extract, analyze, validate, and produce estimate outputs with minimal human intervention.
Key technical breakthroughs included:
- Novel OCR-LLM hybrid architecture that combined OCR bounding boxes with LLM reasoning to extract data from highly variable document formats with higher accuracy and lower cost than VLM-only methods.
- End-to-end multi-cloud integration using Azure OCR, AWS Gateway, S3, Cognito, Bedrock, Entra ID, and Dockerized compute to build a secure, scalable full-stack system.
- Intelligent cost optimization through checkpointing, regex-based snippet extraction, and dynamic model selection to control token usage during batch document processing.
The web application can process batches of documents, create Excel estimate sheets, and generate KMZ files for location-based jobs.
AI Governance Framework
The organization needed a governance approach that balanced responsible AI use with continued innovation, offering practical, easy-to-follow guidance that teams could adopt independently across a federated environment. KUNGFU.AI developed a lightweight, self-service governance framework that included:
The framework:
- A rapid risk assessment and classification method to help teams quickly understand the appropriate level of oversight for different AI use cases.A detailed risk and controls matrix outlining expectations for high, medium, and low-risk categories, aligned with existing enterprise processes.
- Comprehensive guidance for the full AI development lifecycle, including ideation, model evaluation, deployment, and ongoing monitoring.
- Templates, checklists, and self-service materials that made it straightforward for teams across the organization to apply governance consistently without relying on centralized support.
Safety Workflow Generator (Pilot)
KUNGFU.AI also developed and piloted a retrieval-augmented workflow generator that used historical safety and incident reports to create consistent alerts in standard formats. While the pilot was not deployed to production, it validated new opportunities for downstream workflow automation.
Outcome
Measurable Operational Gains, Built to Last
The estimation system is now fully deployed and operated by the organization’s internal cloud team, with estimators reporting more than 80 percent time savings per job. This reduction in manual effort enables teams to respond to more proposals, increase throughput, and pursue additional revenue opportunities.
The AI governance framework has been distributed across every IT leader, giving teams clear, practical guidance for exploring and implementing AI responsibly while maintaining local autonomy. Work is underway to expand oversight for higher-risk use cases and to establish an enterprise steering structure.
Together, these efforts have created several lasting outcomes:
- A production-grade applied AI system that delivers measurable efficiency gains in a mission-critical workflow.
- A practical, self-service governance foundation that enables responsible AI exploration across a federated enterprise.
- A repeatable model that pairs high-impact applied AI with clear guardrails for scale.
- A foundation for future AI use cases in estimation, proposal workflows, legal processes, safety, and other operational domains.
The organization now has both the technical infrastructure and the operating model required to continue expanding AI in a safe, consistent, and value-driven way.
