Tenant Screening: AI Document Verification on GPU

The Challenge: 2,200 Applications and 72 Hours to Verify Each One

A letting agency managing 6,500 rental properties across London and the South East processes approximately 2,200 tenant applications monthly. Each application requires verification of identity documents, proof of address, employment references, three months of bank statements, and three months of payslips. The referencing team of eight spends an average of 72 hours per application — not because individual checks take long, but because the queue means each application waits 2-3 days before a team member begins reviewing documents. During peak letting season (June-September), the backlog stretches to five days, costing landlords an average of 3.5 void days per property at £35 per day — totalling approximately £80,000 per month in landlord dissatisfaction and competitive disadvantage against faster agencies.

Tenant applications contain highly sensitive personal data: bank statements showing spending patterns, payslips with salary details, passport scans, and visa documentation. Routing these through third-party verification services creates GDPR compliance risk and potential data breach exposure.

AI Solution: Automated Document Verification Pipeline

Document AI and PaddleOCR running on a dedicated GPU server extract structured data from uploaded documents: names, dates, amounts, employer details, and account information from payslips and bank statements. A vision model checks document authenticity — detecting common forgery indicators like inconsistent fonts, modified amounts, and template mismatches. An LLM via vLLM cross-references extracted data across documents: does the salary on payslips match deposits in bank statements? Is the employer name consistent? Does the stated address match utility bill evidence?

The pipeline processes a complete application (8-15 documents) in under 5 minutes, flagging inconsistencies for human review while approving clean applications automatically.

GPU Requirements

Document verification processes multiple pages of OCR, fraud detection, and cross-referencing per application. Peak throughput during letting season reaches 150 applications per day.

An RTX 5090 handles peak daily volume seven times over. Private AI hosting ensures all tenant personal data remains within GDPR-compliant UK infrastructure.

Recommended Stack

• PaddleOCR for extracting text from payslips, bank statements, and utility bills.
• Document fraud detection model (fine-tuned vision classifier) for identifying tampered documents.
• vLLM serving a 7B model for cross-referencing extracted data and generating verification reports.
• Structured extraction pipeline mapping OCR output to standardised fields (name, salary, account number, dates).
• CRM integration (Goodlord, Rightmove Referencing, or in-house) for automated workflow.

For handling Right to Rent checks, add a vision model for passport MRZ reading and document authenticity verification. Deploy an AI chatbot for tenant enquiry handling during the application process.

Cost Analysis

The eight-person referencing team costs approximately £240,000 annually. AI verification handles 75% of applications without human intervention, enabling the team to focus on complex cases (self-employed applicants, overseas income, guarantor arrangements). Rather than reducing headcount, the team processes the same volume 4x faster, eliminating the void-day cost of £80,000 per peak month.

Reducing average verification time from 72 hours to 2 hours means landlords’ properties let faster. Over 6,500 managed properties with an average of 1.2 tenancy changes per year, saving 3.5 void days per changeover saves landlords a combined £273,000 annually — a powerful retention metric for the agency’s property management contracts.

Getting Started

Digitise 500 historical applications with verified outcomes (approved, rejected, flagged for fraud). Train the OCR extraction pipeline on the document types you most commonly receive (UK payslips, major bank statement formats, utility bill layouts). Validate by running the AI alongside manual verification for 200 applications, comparing flag rates and accuracy before enabling automated approvals.

GigaGPU provides UK-based dedicated GPU servers for document AI workloads with full GDPR compliance. Scale capacity during peak letting season.