AI Transformation: Building a Smarter Service Charge and Contract Review Foundation

The Context

Industry: Commercial Real Estate and PropTech

Our Services

AI Strategy and Roadmap Automated Document Extraction (LLM-based) Rule-Based Validation Engine Human-in-the-Loop (HITL) Architecture EU Data Residency Compliance

The Mission

"Replace manual, tedious, and time-consuming document review with a scalable AI foundation that identifies unallowable charges and protects the bottom line."

The Challenge: The Unstructured Data Ceiling

In commercial real estate, the most valuable intelligence (lease clauses, service charge limits, and billing milestones) is often trapped inside thousands of unstructured PDFs and scans.

For a scaling management platform, this created a "complexity ceiling." Manual review of service charge statements against original contracts was too slow to scale, leading to a high risk of unauthorized billing items slipping through. The business didn't just need a digitizer; they needed a system that could "reason" between a contract and a bill to find financial discrepancies.

The Solution: The "AI Foundation" Approach

We engineered a cloud-native, AI-driven workflow that moves beyond simple OCR to create a fully auditable, rule-governed extraction pipeline.

1. Intelligent Extraction

We deployed an ensemble of Gemini models, evolving from Gemini 2.5 to the latest Gemini 3 (and 3.1) family. The system converts raw PDFs into structured JSON data with high precision, handling diverse invoice layouts and complex line items automatically.

2. Automated "Dispute" Discovery

By building a dedicated Rule Validation Layer, the system compares extracted billing data against predefined contractual "Hard Rules" as well as AI-extracted rules from the original leases. The system automatically identifies contractual inconsistencies and flags unallowable charges, VAT issues, or calculation errors, enabling a proactive and intelligent dispute process.

3. Human-in-the-Loop (HITL) Efficiency

We designed a bespoke UI where AI confidence levels dictate human intervention. Experts only review documents where the AI flags a high-severity violation or low-confidence extraction, drastically reducing manual overhead.

4. EU Data Sovereignty and Security

The entire infrastructure was architected within Google Cloud (Vertex AI) European regions. This ensures all sensitive financial and contractual data stays within the EU, meeting strict GDPR and residency requirements.

The Tech Stack

Google Vertex AI (Gemini 3 Pro and Flash) Python 3.12 / FastAPI AWS (S3/SQS) GCP (Vertex AI) Agno AI Agents PostgreSQL (pgvector)

The Results Dashboard

60%

HITL Load Reduction

AI handles high-volume templates, freeing experts for complex cases.

95%+

Extraction Accuracy

High-fidelity data capture across diverse document families.

Direct

Financial Savings

Automated detection of unauthorized billing items and math errors.

AI-Ready

Foundation

Structured data lake ready for predictive trend analysis and LLM chat.

Ready to Transform Your Document Workflows?

Don't let valuable data stay trapped in unstructured PDFs. Contact our team to learn how we can help you build a smarter, AI-ready foundation for your business.