Case Study
Revenue Leakage Detection
AI-assisted revenue leakage system for HVAC businesses that detects underbilled labor, materials, fees, and technician-note work, with automated review, recovery tracking, and Power BI reporting.
- Client
- Regional HVAC Company
- Industry
- Field Services
- Role
- Data Analyst, AI Automation Developer, and Power BI Dashboard Developer
- Date
- July 2026
01 / Business Problem
HVAC service businesses can lose revenue when completed work is not fully reflected on customer invoices. Technician labor, installed parts, refrigerant, diagnostic fees, and after-hours charges may be missed because operational records, invoices, and technician notes are stored separately.
Manually reviewing every job is slow and inconsistent, allowing leakage to remain undetected and making it difficult for managers to recover lost revenue or identify recurring billing problems.
02 / Goals
- 01
Detect underbilled labor, materials, refrigerant, and service fees.
- 02
Extract potential billable work from technician notes using AI.
- 03
Validate findings against operational, pricing, and invoice records.
- 04
Provide a human review workflow for confirming or dismissing cases.
- 05
Track recovered and outstanding revenue leakage.
- 06
Automate recurring detection and case assignment.
- 07
Visualize leakage trends and recovery performance in Power BI.
03 / Process
From fragmented information to a usable system.
04 / Solution
Built an end-to-end revenue leakage detection system that compares completed HVAC work against invoices and approved pricing. SQL and Python identify deterministic billing gaps, while OpenAI extracts potential billable work from technician notes for controlled validation.
Detected cases are automatically assigned through n8n, reviewed in Streamlit, tracked through recovery, and summarized in Power BI for operational and executive reporting.
05 / Results
Evidence over decoration.
- 502
- Completed jobs evaluated
- 511
- Technician notes analyzed
- $11,255
- Potential leakage detected
- 83
- Leakage findings generated
- 75
- Leakage cases created
- 28
- Automated tests passed
- $0.0016
- Controlled OpenAI test cost
- Evaluated 502 completed jobs and 511 technician notes.
- Identified $11,255 in potential revenue leakage across the synthetic dataset.
- Detected underbilled labor, refrigerant, after-hours fees, and note-only parts.
- Processed 196 structured note items, with ambiguous AI outputs routed for review.
- Automated recurring detection and case assignment through n8n.
- Enabled case confirmation, dismissal, recovery tracking, and executive reporting.
- Passed 28 automated system and integration tests.
- Completed the controlled OpenAI test for approximately $0.002.
06 / Project Gallery
The system in practice.
Have a similar problem?