Construction – Heavy Civil Subcontractor (Sun Belt): Change-Order & Lien-Exposure Automation

Industry: heavy civil and infrastructure subs. Company size: 210 employees, projects $2M–$45M. Complexity: bonded jobs, joint-check agreements, multi-GC portals. Systems: Viewpoint, GC-specific web portals, fragmented photo evidence in phones.

Manual transcription from radios and texts, inconsistent CO numbering, and no single source of truth for what was approved vs. in dispute. Bottlenecks at PM review and legal sign-off. At scale, one missed lien window erased margin on entire jobs.

Event-driven automation with AI document understanding for CO PDFs from GCs, voice-to-structure for field reports, exposure scoring (cost + calendar), and agent-driven notifications to GCs and internal counsel prep packs.

1. Trigger: field MMS, email PDF, portal scrape job.
2. AI intake: classify CO vs. RFI; extract amounts, dates, references.
3. Structure: canonical CO object; link to job and subcontract.
4. Decision: rules for bond notice; model ranks urgency.
5. Actions: portal posts where APIs exist; email/fax bridges otherwise; tasks for PM; lien calendar entries.
6. Humans: PM approves dollar exposure; legal reviews high-risk.
7. Loop: outcomes update model features for GC behavior patterns.

CO Ingestion Agent, Exposure Scoring Agent, Notice & Calendar Agent, GC Comms Agent (tone-safe drafts), Legal Prep Agent (packet assembly).

Node workers, PostgreSQL, Redis queues, object storage for docs, LLM with citation prompts, integration adapters per GC portal pattern.

Viewpoint, SharePoint, Twilio, PostgreSQL, OpenAI, Node.js, Power Automate bridges

Missed notice incidents fell 91%; average change-order approval cycle shortened 35%; finance recovered 22 hours/week of paralegal-adjacent tracking; DSO on disputed extras improved 12%.

Build: 8 weeks. Deploy: 2 weeks. Optimization: 6 weeks for GC-specific templates.