Why Specialized AI Beats ChatGPT for Tenant Screening

The £20,000 Question: Why pay for specialized forensic AI when you could just paste documents into ChatGPT for free?

1. The Metadata Blind Spot (The Digital Fingerprint)

The Failure: General LLMs cannot read the underlying digital structure of a file.

The Problem: Your most valuable feature—detecting if a PDF was edited in Photoshop or Canva—is physically impossible for a chat model to perform. They cannot access the "digital DNA" or metadata of an uploaded file; they can only read the text you extract and paste.

The Consequence: A fraudster who simply edits a real bank statement will pass the ChatGPT check every time, because the AI is oblivious to the forgery.

ProperLet Solution: We integrate PikePDF specifically to bypass the AI and read the file's raw data. Our tool is designed to be forensic first—checking document integrity before any reading even happens.

2. The Hallucination & Accuracy Failure

The Failure: General LLMs are prone to hallucination and lack the strict calculation focus required for finance.

The Conflict: These models operate with a high "Temperature" (creativity) setting by default to make conversation flow. When calculating complex financial data (like prorated rent, income averaging, or debt-to-income ratios), this creative guessing is catastrophic. One research study showed general LLMs failed to include major assets in basic financial calculations.

The Consequence: A single mathematical hallucination or misreading of a transaction date could lead you to reject a perfect tenant or accept a broke one, directly exposing you to liability and lost rent.

ProperLet Solution: We hardcode temperature=0 (zero creativity) into the Anthropic API call. We use Python code to perform all critical calculations (like Affordability Ratio and Liquidity Check), forcing the AI to act only as an extractor of data, not a calculator.

3. The Workflow & Liability Gap

The Failure: General LLMs stop at the analysis. They do nothing to protect the agent's professional workflow.

ProperLet Solution: Our Forensic Audit Certificate provides a complete, timestamped record of your decision-making process. This audit trail is legally defensible and ensures you have a paper trail if ever challenged by a tenant or regulator.

Affordability Risk Profiling

What We Detect: Rent-to-income burden across all UK tenant types.

ProperLet calculates the precise affordability ratio for each applicant, accounting for:

• Employment Income: Salary from payslips and employment letters
• Self-Employment: Averaged profit/SA302 tax returns (3-year lookback)
• Student Finances: Maintenance loan amounts and assumed parental support
• Pension Income: State pension, private pensions, and annuities
• Joint Applications: Combined household income with co-applicant share calculations

Traffic Light System: Green (<35%), Amber (35-50%), Red (≥50%) rent burden—instant visual feedback on affordability.

Lifestyle Risk Keywords

What We Detect: Risky financial behaviours buried in transaction descriptions.

Our system scans bank statements for keywords indicating:

• Gambling Transactions: Bookmakers, betting exchanges, casinos (signals impulse spending)
• Debt/Lending Activity: Payday loans, credit transfers to friends, money lenders (signals financial distress)
• Bounced Cheques/Overdraft Fees: Penalty charges (signals payment stress)

One or two flags = Amber warning. Three or more = High-risk red flag.

Why This Matters: A perfectly solvent applicant with a £50k salary might still be a tenant default risk if they're regularly borrowing from payday lenders. General LLMs cannot contextualize these patterns.

Ghost Deposit Detection

What We Detect: Unexplained large deposits indicating borrowed/gifted funds.

ProperLet analyzes all deposits over 50% of monthly income to flag suspicious patterns:

• One-off anomalies: Large deposits that appear once then disappear (signs of temporary borrowing)
• Suspicious timing: Deposits that arrive exactly before rent is paid (coordinated gifting)
• Cross-account transfers: Deposits from accounts named differently (possible third-party loans)

Why This Matters: A tenant might have £50k in savings, but if it's a one-time "loan" from family, they're still a default risk. This detection catches what spreadsheet analysis misses.

How the Fraud Repository Works

Step 1: Hash Calculation

Every file you upload receives a unique cryptographic fingerprint (MD5 hash). This fingerprint is impossible to fake and remains the same even if the file is renamed or moved.

Step 2: Database Check

Your file hash is instantly checked against the global repository of flagged documents. If your document matches a known fraudulent file:

• NETWORK HIT Alert: Immediate critical warning appears at the top of your results
• Automatic Rejection: Risk status is forced to RED—regardless of other analysis results
• Non-Negotiable: Network hits cannot be overridden; they represent community consensus on fraud

Step 3: Repository Auto-Population

When ProperLet detects a HIGH RISK case, all uploaded file hashes are automatically added to the shared repository. This means:

• Your analysis protects others: The next landlord who encounters the same fraudulent document will be instantly warned
• Exponential Protection: As more landlords use ProperLet, the fraud database grows stronger and faster
• Real-Time Detection: Fraudsters cannot recycle the same forged bank statement across multiple applications

Why This Beats General AI

The Problem with ChatGPT: It has no memory of past frauds and no ability to warn you about known fraudulent documents.

Every time you paste a document into ChatGPT, it analyzes it in isolation. A fraudster could use the same forged payslip a hundred times across a hundred applications, and ChatGPT would be oblivious.

The ProperLet Difference: Our system builds institutional knowledge. The community learns from each case, and every landlord benefits from that collective intelligence instantly. This is a true network effect that protects everyone.