I TRIED TO DO MY OWN TAX WITH AI.
IT NEARLY BROKE ME.

The problem I lived

Three investment properties. Two part-time incomes. A side business. One SMSF. Every financial year, the same story: weeks of downloading bank statements, hunting for receipts in my email, cross-referencing depreciation schedules, and trying to remember which expenses were repairs and which were capital improvements.

I'm the kind of person who wants to understand their own numbers. I don't want to hand a shoebox of receipts to an accountant and hope for the best. I want to know what I'm claiming, why I'm claiming it, and whether I'm missing anything.

The AI promise — and the reality

When ChatGPT arrived, I thought I'd found the answer. I pasted in bank statements. I asked about negative gearing rules. I got confident-sounding answers. But then I noticed the cracks.

Every conversation started from scratch. It forgot my properties between sessions. It couldn't process a CSV. It mixed up state-specific rules. And the worst part: it sounded certain about things that were wrong.

I tried Claude, Gemini, ChatGPT. They're all brilliant reasoning engines. But managing property tax isn't a conversation problem — it's a data problem. You need a system that remembers your portfolio, processes your documents, and builds an evidence trail that survives an audit.

What I actually needed

After months of trying to make general-purpose AI work for property tax, I realised what I actually needed wasn't a smarter chatbot. It was a dedicated system.

• Ingest actual documents — bank CSVs, loan statements, QS reports, PEXA settlements
• Remember my portfolio — properties, purchase dates, loan structures, cost bases
• Classify expenses correctly — using ATO rental schedule categories, not generic guesses
• Track depreciation properly — Div 40 and Div 43, parsed from QS reports, updated annually
• Build an evidence pack — every claim matched to a receipt, gaps flagged automatically
• Protect my data — redact TFNs, ABNs, Medicare numbers before anything reaches an AI model

No product did this. So I built one.

A ledger, not a chatbot

Lodgey is the product I wished existed when I was sitting at my kitchen table at 11pm on a Tuesday, three browser tabs deep in ATO rulings, trying to figure out whether a $4,200 plumber's invoice was a repair or a capital improvement.

It uses the same AI models that power ChatGPT and Claude — but wraps them with specialised tools that property tax actually requires. Document pipelines. ATO-sourced knowledge files. Depreciation calculators. A persistent database of your properties, loans, and deductions.

The result is something that looks like a chat but works like a ledger. Your data doesn't disappear when you close the tab. Your evidence pack grows with every document you upload. And when EOFY arrives, you export a structured workbook your accountant can actually use.

Where we're going

Property tax in Australia is unreasonably complex. Eight states and territories, each with different thresholds, rates, and surcharges. Federal CGT rules that interact with state land tax in non-obvious ways. The ATO found errors on 9 in 10 rental returns last year.

This complexity shouldn't be the investor's problem. AI is finally good enough to absorb it and present simple, actionable outputs. But only if you build the right system around it — not just a chat window with a prompt.

Lodgey will become the system of record for Australian property investors. Every document, every deduction, every cost base adjustment, and every evidence trail — organised, classified, and ready.

The era of doing your property tax with a general-purpose chatbot and a spreadsheet is ending. Lodgey is what comes next.