How Proppi Builds Property Memory from Documents (The Technical Bit)

Why Property Documents Specifically?

Property documents are a surprisingly good fit for AI. Here’s why: they’re highly structured (leases follow patterns, insurance policies have standard sections, rates notices look similar every year), but they arrive in messy formats (email attachments, scanned PDFs, photos of paper, stuff your property manager forwards).

That combination — structured content in unstructured formats — is exactly what modern AI processing can turn into useful property memory.

A typical landlord in New Zealand or Australia with a few properties accumulates dozens of document types — tenancy agreements, bond lodgement forms, inspection reports, insurance certificates, rates notices, Healthy Homes assessments in New Zealand, Annual Fire Safety Statements and state-by-state compliance records across Australia, and more. Across a portfolio, the volume gets real.

Traditional approaches — Google Drive folders, spreadsheets tracking key dates, that one folder on your desktop called “Property Stuff 2024” — work fine until they don’t. They rely on you doing the sorting, you doing the searching, and your memory to track what’s due when.

Proppi changes the equation by building the memory layer first. Here’s how each step works.

Step 1: OCR — Reading the Document

When you upload a property document — PDF, scanned image, photo from your phone — the first thing that happens is OCR (Optical Character Recognition).

What good OCR handles:

• Layout analysis — headers, paragraphs, tables, signatures, page numbers
• Handwriting — annotations, margin notes, signed sections
• Multi-page documents — maintaining structure across pages (a 30-page lease isn’t treated as 30 separate documents)
• Dodgy quality — faded faxes, phone photos at weird angles, that scan your property manager made on a printer from 2009

The output isn’t just a flat wall of text. It’s structured: the OCR knows which text belongs to which section, which numbers are in which table column, and what the reading order is. That structure is critical for the next step.

Step 2: Classification — What Kind of Document Is This?

Once the AI can read the document, it classifies it. Is this a tenancy agreement? An insurance policy? A rates notice? A building inspection report?

For New Zealand and Australian property documents, the categories look something like this:

For the full breakdown, check our New Zealand property document types reference.

The key insight: classification isn’t just labelling. It tells the system what to look for next. A tenancy agreement needs lease dates and rent amounts. An insurance policy needs expiry dates and coverage types. A rates notice needs payment due dates. Classification determines which extraction rules apply.

Step 3: Entity Extraction — Pulling Out What Matters

This is where the real value kicks in. After the AI knows what kind of document it’s looking at, it extracts the specific pieces of information that actually matter:

• Dates — lease start/end, insurance expiry, next inspection due, Healthy Homes compliance deadline, ATO record-keeping windows
• Amounts — weekly rent, bond ($), premium, rates payable
• Parties — tenant names, landlord details, agent contacts, insurer
• Properties — street addresses, legal descriptions
• Clauses — break clauses, renewal options, special conditions, exclusions

The extraction is context-aware. The AI knows that “$550” in a tenancy agreement is probably a weekly rent, while “$550” in an insurance document is probably an excess. Same characters, different meaning — and the classification step gives the AI the context to get it right.

Step 4: Source Linking — Preparing Source-Linked Work

The final layer is what makes all of this usable day-to-day. Instead of navigating folders or remembering file names, you tell Proppi what work needs to move, and the system links the source pages it needs:

• “Prepare the rent and expiry summary for my Grey Lynn flat in Auckland.”
• “Check whether 15 Oak Avenue, Brisbane has current building-inspection records.”
• “Prepare the Australian Taxation Office substantiation pack for my Sydney property in FY25.”
• “Prepare the special-conditions note for the Smith tenancy.”

Under the hood, source linking matches on meaning, not just keywords. “When does the lease expire?” and “tenancy end date” point to the same kind of property memory.

Proppi links the relevant passages, prepares the work, and keeps the source document attached so you can approve it with context. No blind trust required.

Why This Matters More Than You’d Think

The obvious benefit is speed — finding the right records faster. But the bigger win is what happens when your documents become usable as a connected source record:

Compliance visibility. “Which of my properties have current Healthy Homes assessments?” (or for Australian landlords: “which properties have a valid Annual Fire Safety Statement?”) isn’t a question you can answer quickly with a filing system. With extracted entities and classification, Proppi can prepare a source-linked status view for approval.

Tax time. Your accountant — or you, preparing Inland Revenue records in New Zealand or Australian Taxation Office substantiation in Australia — asks for last year’s interest statements, repair invoices, insurance premiums, and rates notices across all properties. That’s an afternoon with folders. It becomes tax-prep work for approval with source links attached.

Deadline safety. Proppi doesn’t just extract dates — it watches them. Lease renewals, insurance expiries, compliance deadlines. No spreadsheet to maintain. No calendar reminders to set up and hope you don’t accidentally delete.

Cost awareness. When financial documents are connected to properties, dates, and categories, Proppi can prepare a source-linked view of which properties cost the most to maintain, where insurance premiums are climbing, and whether your yield calculations still stack up.

The Limitations (Being Honest)

Property memory is not magic. A few things to keep in mind:

• Garbage in, garbage out. If the scan is completely unreadable, OCR can’t save it. Phone photos work, but a blurry snap of a crumpled document taken in bad lighting will give poor results.
• Edge cases. Unusual document formats, non-standard templates, or very short documents (a one-line email forwarding an attachment) can sometimes trip up classification.
• Not legal advice. Proppi can show what a document says and prepare work from it. It can’t tell you what it means in a legal context. For that, you still need your solicitor.
• Only as complete as what you upload. If you haven’t uploaded your insurance policy, the AI can’t tell you when it expires. Sounds obvious, but the system only knows what it’s been given.

Getting Started

The barrier to entry is genuinely low. Upload your existing documents — even the messy scanned ones — and Proppi handles classification, extraction, and source linking automatically. No tagging taxonomy to design, no training period.

Start with the critical stuff: current tenancy agreements, insurance certificates, and anything with an upcoming deadline. You’ll see the value immediately.

If you want to try this with your own portfolio, Proppi reads your documents into the property file and prepares source-linked work for your approval — free while we build it with early members.

Keep Reading

• Property Document Types in New Zealand: Complete Reference — every document type you’ll encounter and what’s in them
• New Zealand Healthy Homes Standards: Compliance Guide — the compliance documents you’ll definitely need to track
• Australia Landlord Compliance Checklist 2026 — the state-by-state obligations and records Australian landlords need
• ATO Property Data Matching 2026: Records Australian Landlords Need — what the Australian Taxation Office cross-references and the documents that defend your return
• Hidden Costs of Property Investment — where all those documents (and costs) come from
• Rental Rule Changes 2026: What New Zealand and Australia Landlords Need to Track — the current rule-change series built around source documents and watched dates

Further Reading

If you’re investing in property across New Zealand or Australia, our Property Investment Gotchas 101 series covers the tax rules, hidden costs, and compliance traps that generate all those documents in the first place — from the bright-line test and Healthy Homes compliance to stamp duty and hidden costs across the Tasman.

For a timely compliance cluster, the Rental Rule Changes Watch 2026 series shows how tenancy changes, pet bonds, Inland Revenue records, bright-line dates, and Australian Taxation Office deduction guidance all become source-document work.