Why AI Can't Manage Your Property Portfolio
- AI can't manage your portfolio because it has no access to your loan balances, outgoings, or fund deed — and no accountability when it's wrong.
- It's genuinely useful for summarising documents, explaining structures, drafting communications, and spotting patterns in data you provide.
- A miscalculated LVR or debt-to-income ratio from a confident guess can cost you unbudgeted LMI or a declined refinance.
- The real work is bookkeeping: structured data kept accurate and calculated the same way every time.
Ask a large language model what your interest-only period rolling over to principal-and-interest will do to your cash flow, and it will give you a confident answer. It might even be right. But it doesn’t know your actual loan balance, your fixed rate expiry, or that your Brisbane property’s body corporate just levied a special assessment for remedial concrete work. Confidence without your data is just a well-worded guess.
There’s a lot of noise about AI running your investments. It’s worth being precise about where the technology genuinely helps and where it quietly falls apart.
Where AI breaks down
The core problem is that portfolio decisions depend on facts a general model has no access to, and on judgement it cannot exercise.
Consider a lender serviceability assessment. To know whether your portfolio qualifies for the next loan, a lender needs your total debt across all properties, your gross income, and the result of shading your rental income to 75–80% and testing all repayments at 3% above actual rates. The critical number here isn’t a single property’s rent-to-repayment ratio — it’s your total debt-to-income ratio across everything you own, checked against the APRA 6x cap that limits high-DTI lending. A model that hasn’t seen your figures will construct plausible ones. Plausible is worthless when you’re deciding whether to proceed to formal approval.
The same applies to anything time-sensitive or jurisdiction-specific. CGT discount eligibility turns on the exact 12-month holding threshold and settlement dates. Depreciation depends on the quantity surveyor’s schedule for that specific building, its construction date, and whether Division 40 plant-and-equipment applies. SMSF compliance depends on your fund’s deed, the sole purpose test, and whether the property is subject to a limited recourse borrowing arrangement. These aren’t questions with a general answer. They have a correct answer for your situation, and it changes when your situation changes.
AI also has no accountability. If it tells you your LVR is 78% when it’s actually 82% because it misread a valuation, there’s no recourse. A miscalculated LVR that pushes you past 80% means LMI you didn’t budget for, or a refinance that gets declined at assessment stage.
Some AI tools now connect to external data sources via APIs — bank feeds, property data services. This narrows the gap slightly. But unless the tool holds a clean, consistent record of your portfolio that you’ve maintained over time, it’s still making inferences from whatever you’ve fed it in that session. An inference isn’t a record.
What it’s actually good at
Strip away the hype and there’s a real, narrow set of tasks where AI earns its place.
Summarising documents you already have. Feed it a 40-page strata report and ask what maintenance items are flagged for the next three years. Ask it to pull the rent review clauses out of a commercial lease. It’s genuinely useful at reading fast and surfacing what matters — provided you verify the specifics against the source.
Explaining unfamiliar structures. If you’re weighing a unit trust against a company as a holding structure, AI can lay out the general trade-offs on land tax thresholds, asset protection, and distribution flexibility. Treat this as a starting brief for a conversation with your accountant, not as advice.
Drafting and communication. Chasing a property manager about an overdue statement, writing to a tenant about a rent increase within notice periods, structuring a question for your broker. Low stakes, high time savings.
Spotting patterns in data you provide. If you give it twelve months of actual expenses across five properties, it can flag that one property’s maintenance costs are running 40% above the others. That’s a useful prompt to investigate. It’s not a diagnosis.
The common thread: AI helps with language and pattern-spotting on data you supply. It doesn’t help with holding the authoritative record of your portfolio, because it doesn’t have one.
The part that isn’t AI
The unglamorous truth is that most of what makes portfolio management work is bookkeeping done properly. Knowing the current LVR on every property. Tracking actuals against your forecast so you can see the negatively geared property that’s drifting further into the red than you modelled. Running a pre-purchase scenario before you commit, so you know what a sixth acquisition does to your consolidated DTI and whether the deposit and stamp duty actually clear your available funds.
That’s structured data, kept accurate, calculated the same way every time.
Use AI for the reading and the drafting. Keep the numbers somewhere they’re actually correct.
Track your own portfolio free → prop-insights.app