Methodology
How LeaseLens builds your report
LeaseLens reads the documents, extracts key facts, applies a fixed set of risk checks, and then runs a staged AI drafting workflow. At launch, those AI stages run in a deterministic sample mode and the paid reports are supported by concierge review.
The current methodology is designed for England & Wales leasehold packs. It uses a rule library for the main risk categories so important findings are not driven only by open-ended AI responses.
How the report is put together
- Ingestion and file metadata capture
- PDF text extraction with OCR fallback for low-density files
- Document classification into core pack categories
- Extraction of charges, reserve figures, insurance details, and risk indicators
- Rule-based flagging and confidence handling
- AI pack triage, financial analysis, issue prioritisation, report drafting, and QA review
- Final QA pass to reduce overclaiming and highlight contradictions
Current AI step contract
- Each AI stage has a named prompt and a required JSON output format
- The current live product uses deterministic sample responses for those stages
- The report is built from those structured step outputs rather than free-form AI text alone
- Every serious issue still needs evidence or a verification prompt
Guardrails
- No overclaiming or unsupported conclusions
- No “buy / don't buy” recommendations
- Medium/high findings should carry evidence context
- Contradictions are converted into verification prompts
- Missing documents reduce confidence instead of being ignored
Launch operating model
The launch product is concierge-backed rather than fully automated. The paid report is meant to support human review, not replace it.
