AI appraisal preparation.

What appraisal preparation covers

Appraisal preparation is the desk work behind the written appraisal a vendor receives before signing an agency agreement: finding recent sales, deciding which are comparable, laying them out, and explaining how the property compares and how the range was reached. It is often done in a hurry before a listing presentation, which is where shortcuts show. REA’s appraisals guidance describes a complaints committee that saw “simply a presentation of data collected with no evidence of analysis”.

AI is useful for the gathering and the first draft of the analysis. REA already lets a non-licensed assistant “compile market data to support an appraisal”, adding “you must analyse the data and prepare the appraisal”. The same line works for software: the model compiles and drafts, the licensee analyses and decides.

How the workflow runs

Inputs

The subject property’s title, land and floor area, rooms and the agent’s condition notes and photos from the visit; recent sales and current listings from a data provider such as Property Guru; prior appraisals for the address; the fee schedule from the draft agency agreement.

Systems

The sales data subscription, the CRM, the agency’s appraisal template, a model API, and a small rules engine for filters and commission arithmetic.

Steps

Pull sales by rule (distance, date window, property type); filter by land and floor size bands; draft a one-line comparison for each sale against the subject; let the agent select; draft the reasoning for the range and the exclusions; calculate the commission estimate; draft a rule 10.3 note if evidence is thin; flag the word “valuation”.

Human review

The licensee views the property, chooses every comparable, sets the range and rewrites the reasoning in their own words. A branch manager samples appraisals each month, looking for one-sided evidence.

Outputs

A draft written appraisal with a comparables table, inclusion and exclusion notes, a commission estimate and a file record of the data pulled and the choices made.

Measures

Agent hours per appraisal, comparables the agent changed, appraisals revised before listing, and later sale prices against appraised ranges, tracked as a calibration check rather than a target.

Risks

Favourable sales only, stale or mismatched sales, missing renovation context, valuation language, and other owners’ names reaching the vendor’s document.

Illustrative example: a three-bedroom home in Hoon Hay, Christchurch

Illustrative example, synthetic numbers: an agent appraises a 1970s three-bedroom weatherboard house on 680 m² in Hoon Hay.

The agent drops two sales: one with a new kitchen and bathroom, one on a rear section with no street frontage. With an assumed fee of 3.5% on the first $400,000 and 2% on the balance, plus 15% GST, the rules engine gives an estimated commission of $22,770 at $690,000 (14,000 + 5,800 + 2,970) and $23,920 at $740,000 (14,000 + 6,800 + 3,120).

If the agency prepares 10 appraisals a month and each takes 150 minutes by hand against 75 with a draft, the saving is 12.5 hours a month. At about 30,000 input and 3,000 output tokens per appraisal on Claude Sonnet 5.5 (US$2 and US$10 per million tokens, as listed on 19 October 2026), model fees are about US$0.09 an appraisal.

New Zealand rules for appraisals

• Written and evidenced. Rule 10.2 of the Code of Conduct requires an appraisal in writing that realistically reflects current market conditions and is supported by comparable sales. Rule 10.3 requires a written explanation when there are none.
• Viewed, not estimated. REA says an electronic appraisal or market estimate “is unlikely to satisfy your requirements”, and that not visiting the property risks breaching rule 5.1. A renewed sole agency needs a fresh appraisal.
• Balanced evidence. REA reports a finding of unsatisfactory conduct where a licensee “considered the most favourable sales” and ignored the rest, and warns that an overinflated range may breach rule 6.4.
• Not a valuation. REA reminds licensees that valuing land for the public requires a registered valuer under the Valuers Act 1948.
• AI responsibility. REA’s generative AI guidance names appraisals as an area of “significant risks” and says an AI error is no defence.
• Privacy. Sales data often includes owner names. The vendor’s appraisal does not need them, so strip them before drafting. If owner data feeds prospecting, IPP 3A notice duties have applied since 1 May 2026. Check your data licence before sending provider data to an AI service.

Where appraisal drafts go wrong

• Confident adjustments. A model will happily say a sale is “about $40,000 better” because of a deck. Let it describe differences, never price them, and keep numbers in code, as our guide to structured outputs and validation explains.
• Wrong sales. Cross-lease flats, unit titles and lifestyle blocks slip through filters. Show title type in every row.
• Old markets. A six-month window can span a turn in prices. Show sale dates and let the agent weigh them.
• Copied prose. Three appraisals with identical paragraphs read as no analysis. Require the agent’s own reasoning.

A sound appraisal underpins the price wording used in listing description drafting and the price replies in buyer enquiry handling. The real estate sector guide covers the wider rules. Our AI document processing and AI reporting services cover data pulls, templates and review queues.

Written by the Levy Street team in Wellington. Last updated 19 October 2026.