The AI Buyer's Advocate, Explained: What It Is, What It Does, and What Stays Human

An AI buyer’s advocate is software that does the buyer’s side of a property purchase (research, due diligence, valuation, offer strategy, negotiation support, settlement tracking) as an autonomous agent, rather than as advice you act on. It’s the workload a human buyer’s agent charges $8,000–$25,000 per deal for, run as code. The term is new; the workload it describes is thirty years old.

This page defines the category: what qualifies, what doesn’t, what stays human, and how to evaluate any product that claims the label, including Heyward, which is built as one.

The test for whether something is an AI buyer’s advocate is simple: does it do the work, or does it tell you about the work?

A buyer’s advocate (the human kind) is paid to execute: find candidates, verify each one, compute what it’s worth, structure the offer, run the negotiation, chase the gates to settlement. Software earns the same label only if it executes the same stages itself and hands you decisions, not homework.

That draws a hard line against three adjacent things that get marketed with the same vocabulary:

• Not a chatbot. General AI assistants answer property questions convincingly, but they hold no data feeds, run no checks against the actual title or hazard layers for the address in front of you, and carry nothing forward from your last deal. Advice, not execution.
• Not a data dashboard. Suburb-analytics platforms surface medians, yields, and growth series. Genuinely useful. But the interpretation, the valuation, the offer, and the deal are still all yours. Information, not execution.
• Not a course. Education transfers the framework and leaves you the ~40 hours per deal of applying it. Knowledge, not execution. (We’ve written up what that trade actually costs.)

A handful of products now market themselves near this category. Most are chat wrappers or dashboards under the hood. The label is earned at the execution line, not in the tagline.

Strip the offices and the licensing and a buyer’s advocate’s labour decomposes into roughly ten things per transaction: identifying candidates, verifying each one (flood, bushfire, easements, encumbrances, comparables), determining fair value, structuring the offer, negotiating, coordinating the conveyancer, interpreting building & pest, managing the finance gate, reviewing contract revisions, and coordinating settlement.

For most advocates that’s 20–40 hours of labour across 30–60 days, and the industry prices it at $8,000–$25,000 fixed or 1.5–3% of purchase price plus GST, Sydney averaging around $14,500, with a $1,000–$3,000 engagement fee up front. The full decomposition, with a worked stage-by-stage example, is in our buyer’s agent fee guide.

Two structural facts about the human model matter for understanding why the AI category exists:

1. The cost restarts on every deal. Five properties over five years is five fees.
2. The methodology varies. Different advocates weight comparables differently, negotiate differently, tolerate risk differently. Deal 1 with one advocate isn’t the same shape as deal 3 with another.

Software attacks exactly those two properties: a subscription doesn’t restart per deal, and code applies the same methodology on deal 1 and deal 50.

A category definition you can trust has to say what the software doesn’t do. The split, as it stands in 2026:

For a worked example of what the software side looks like in practice: Heyward prices offers from weighted comparables (five data sources, typically 14 comparables, ±4.8% confidence band on a typical residential property). A 20-property Sydney pilot in 2025–26 found asking prices averaged 9.5% above comparable-supported fair value; the software uses that pilot-derived calibration conditionally. It also runs 50+ due-diligence signals across six categories with source and severity on each, and advances a 13-gate lifecycle from offer to settlement as documents and emails arrive. Mean pricing error in the initial nine-sale validation sample was 6.1%. Both figures are early samples, not market-wide benchmarks or accuracy guarantees. Your active time per deal: minutes, not hours.

One rail matters more than any feature, and it belongs in the category definition: irreversible steps stay human. In Heyward’s case, a walk-away recommendation never auto-sends. Every step that ends or commits a deal requires your confirmation. Any product that removes the human from irreversible decisions has left the advocate category and become a liability.

Whether you’re evaluating Heyward or anything else wearing the label, the same six questions separate an advocate from a wrapper:

1. Can every number be traced? Ask where a valuation came from. “Comparable 14 of 14, weighted by recency × similarity × distance” is an answer. “Our AI model” is not.
2. Does it check the actual property? Flood, bushfire, easements, encumbrances against this address, or generic suburb commentary?
3. Does it state uncertainty? A tool that shows “4 comparables · wide band” on thin data is safer than one that’s always confident. Confidence without a basis is the most expensive feature in property.
4. Does it execute, or advise? Does the deal actually move (offers structured, deadlines tracked, contract revisions diffed), or do you get a report and a to-do list?
5. Who holds the pen on irreversible steps? You should. Anything that auto-sends walk-aways, offers, or signatures is mispositioned.
6. What happens to your data? Portfolio and deal history should be yours: exportable, deletable, not training material for someone else’s model by default.

A vendor that fails question 1 or 5 isn’t an AI buyer’s advocate regardless of the marketing.

The human model: $8,000–$25,000 per property, every property (decomposed here). Five deals over five years ≈ $72,500 at the Sydney average.

The AI model: subscription-priced rather than a fee on every purchase. Heyward is A$0 for Start, A$129 + GST a month (or A$1,290 + GST a year) for Watch, and A$349 + GST a month for Buy. Buy includes Watch, carries up to five purchases at once and can be cancelled when the buying window ends. There is no per-purchase or success fee. The full allowance table is on the access page. Plans are live; you can start free and upgrade in the app.