StructumAI Estimating & Billing

StructumAI needed one module to carry a construction project end to end: estimate → awarded contract → schedule of values → AIA-style pay applications → change orders → billing ledger, plus the full cost side: budgets, purchase orders, and vendor invoices. Avrixo designed, architected, built, and deployed it as a standalone Next.js application integrated natively into the multi-tenant platform.

The result is a production revenue-and-cost loop where a number entered at bid time stays traceable through every downstream document, accounting-ready through QuickBooks Online, QuickBooks Desktop, and Plaid bank feeds, with AI-assisted receipt capture feeding purchase orders automatically.

The estimating surface is an Excel-fast grid with seven line types (standard, allowance, alternate, unit-price, deduct, contingency, internal-only), per-line and rule-based markup, and internal-vs-owner visibility. Estimates carry versioning with lock/approve controls, import from Excel through a preview-then-commit flow with AI special-row detection flagging unusual lines for review, and clone from company-level estimate and line-item templates.

An awarded estimate converts into a contract and a balanced Schedule of Values. From there: AIA-compatible pay applications with retainage, stored materials, and a submit → approve → owner-approve → record-payment flow; owner change orders with full lineage back to the contract and SOV; a cross-contract billing ledger; and PDF + XLSX export that balances exactly to AIA math.

On the cost side: budget baselines from the awarded estimate, commitments and purchase orders, vendor invoices with approval and posting, and reports for profit fade, change-order gap, and billing-vs-cost. An AI receipt pipeline extracts vendor, total, line items, CSI codes, and date from an uploaded receipt and auto-aggregates it into a per-project purchase order.

The module ships as its own Next.js 16 application on a dedicated subdomain, but feels native inside StructumAI: single sign-on via a shared JWT cookie scoped to the platform domain (no second login, logged-out users redirect to the platform login), and a shared multi-tenant PostgreSQL database that reads the platform's real projects and membership, the same projects shown across Plans and Schedule. Where the module's tables would have collided with the platform's own finance tables, they were namespaced so each side stays isolated.