Sales Rental appraisal with Microsoft Azure and Open AI

About the Client

Bayleys Real Estate is one of New Zealand’s largest and most recognised real estate agencies, operating across residential, commercial, rural, and property management services throughout New Zealand and Australia (including McGrath Real Estate). With over 150 property management agents generating 1,800–2,000 rental appraisals per month, the volume and complexity of documentation demanded a step-change in how the business operated.

The leadership team — including the National Director and Financial Director — identified AI-driven automation as the strategic priority, with a clear mandate: reduce time spent on repetitive documentation tasks, improve consistency and accuracy, and free agents to focus on client relationships.

The Problem

Each rental appraisal traditionally required an agent to manually complete market analysis, write a property description, cross-reference compliance requirements, and produce a formatted report — a process taking 30 to 45 minutes per appraisal. At an average cost of NZD $50 per agent-hour, this represented a significant and measurable drag on productivity.

With 1,800 to 2,000 appraisals generated per month — and a target of 60,000 to 67,000 per year — the business needed automation, not incremental improvement. Key pain points included:

Inconsistent quality and formatting of appraisals across regions and agents
No centralised, AI-assisted way to access tenancy legislation, tribunal case law, or market data
Agents frequently leaving the platform to search external sites for comparable properties or school zone information
No automated mechanism to prevent appraisals older than 90 days from being shared with clients
Manual population of agent profiles, property features, and branding on every report

Solution Overview

Peritos Solutions designed and built a two-module platform hosted on Microsoft Azure, integrating with Bayleys’ existing technology ecosystem including Office 365, CoreLogic, and the Bayleys property API.

Part 1 — Automated Appraisal Engine

The appraisal engine automates the end-to-end production of a professional, branded rental appraisal report. A property manager enters or confirms a property address and the system handles the rest:

Property data (bedrooms, bathrooms, floor area, carparks) is automatically retrieved via the CoreLogic API integration
Property images are pulled from the Bayleys listing API where the property is actively listed for sale — ensuring brand consistency
Agent profile photos, contact details, and office information are automatically populated via Office 365 Single Sign-On — the system knows who is logged in
An AI-generated property description is produced in seconds, trained on thousands of historical Bayleys appraisals using sentiment analysis to match the tone and style of an experienced property manager
School zones and local amenities are automatically identified from the property address
A regional disclaimer is appended, with built-in logic preventing appraisals older than 90 days from being shared without review
Agents can select from multiple property types (house, apartment, unit, minor dwelling) and customise the back-page advert by region or user
For new or off-plan properties not yet in data systems, agents can manually enter property details — which feeds directly back into the Bayleys Data Lake, building proprietary market data ahead of public availability

Part 1 — EMV (Estimated Market Value) Engine

The EMV engine is the technical centrepiece of the appraisal platform. Peritos evaluated multiple modelling approaches and selected a Random Forest Regression model, trained on data from the Bayleys Data Lake, as the optimal fit for rental valuation.

The model evolved through three refinement stages:

Initial model (basic features only) — margin of error exceeded ±15%, predictions unreliable at extremes
After geometric mean aggregation — outlier influence dampened, margin of error narrowed to ±8–10%
Final model with full feature set and local correlation weighting — dynamic suburb/street-level weighting using Pearson’s R between property capital value and achieved rent — margin of error consistently within ±3% nationwide

Key technical components of the EMV model include:

Bootstrap aggregation (bagging) — many de-correlated decision trees built from different samples of the Data Lake, averaged to reduce variance
Geometric mean aggregation of individual tree outputs — dampens the influence of extreme outliers, particularly effective in large heterogeneous suburbs such as Remuera
Suburb and street-level Pearson correlation analysis — dynamically adjusts feature weighting based on the local relationship between capital value (CV) and achieved rent (e.g. R=0.93 in Epsom, meaning CV carries very high weight in that suburb’s regression)
The model resolves the classic valuation challenge of ‘best house on the worst street vs. worst house on the best street’ by varying feature weights by location rather than applying a national average

Part 2 — AskKen AI Chatbot

AskKen AI is a purpose-built real estate intelligence assistant, powered by OpenAI GPT-4O with a custom AI engine layered on top to control output quality and data sources. It is accessible via mobile and desktop and requires no training to use.

The architecture uses Retrieval-Augmented Generation (RAG) to ground the LLM’s responses in Bayleys‘ proprietary and controlled data sources rather than the open internet:

Proprietary documents — including the Residential Tenancies Act, relevant legislation, tenancy tribunal cases, suburb profiles, maintenance cost databases, and vendor checklists — are ingested, indexed in a vector store, and retrieved at query time
54,000 tenancy tribunal cases have been ingested to give the model deep contextual and interpretive capability, not just legislative knowledge
Fine-tuning and prompt engineering steer GPT-4O toward real-estate-specific tone, compliance obligations, and output style
A filtering and guardrails layer reviews all LLM outputs against compliance checklists, strips unsupported assertions, and flags uncertain answers for human review
A knowledge management layer tracks document version, source authority, and effective date — enabling rapid re-indexing when legislation or market data changes

AskKen AI is capable of handling queries across legislation and compliance, comparable market analysis, tenancy tribunal precedents, suburb and school zone research, rental market reports, maintenance cost estimates, yield calculations, and more.

Technical Architecture

Cloud Platform	Microsoft Azure — cloud-native, scalable infrastructure
AI / LLM Core	OpenAI GPT-4O with supervised fine-tuning and custom prompt engineering
Valuation Model	Random Forest Regression — bootstrap aggregation, geometric mean, Pearson R correlation weighting
RAG Layer	Vector store indexing of proprietary legal, market, and operational documents — real-time retrieval at query time
Data Lake	Bayleys proprietary property data, appraisal history, and manually entered new-build data
Integrations	CoreLogic API (property attributes), Bayleys API (listing images), Office 365 SSO (user profiles)
AI Copywriting	Sentiment-analysis-trained model on thousands of historical appraisals — generates descriptions matching Bayleys tone
Output	Professionally formatted, branded PDF appraisal report with agent profile, property image, EMV, description, and disclaimer
Platforms Supported	Web (desktop, mobile, tablet) — zero-training, no-manuals design philosophy
Markets	New Zealand (primary) — extensible to Australia (McGrath Real Estate)
Security	Internal vs external data segregation; only version-controlled approved data ingested into AI; external web search disabled by default

Key Challenges & Solutions

First AI project under scrutiny	As Bayleys‘ inaugural AI initiative, the project carried high visibility. Peritos delivered on time, within budget, and with quantifiable ROI — establishing the blueprint for future AI projects across the business.
EMV accuracy	Initial models had ±15% error. Through iterative refinement — geometric mean aggregation, Pearson R suburb correlation, and full feature weighting — accuracy was driven to ±3% nationwide, matching experienced agent assessments.
No-training UX requirement	The platform was designed to require zero manuals or training, following the design philosophy that a good product is intuitive regardless of technical experience. The interface works seamlessly across mobile, desktop, and tablet.
Data gaps for new properties	Off-plan and newly built properties often lack data in CoreLogic or the Bayleys API. A manual entry flow was built to capture this information, feeding it directly into the Data Lake to improve future model accuracy.
Preventing stale appraisals	Built-in logic prevents appraisals older than 90 days from being shared without a property manager review — protecting compliance and professional standards.
Cross-market compatibility	The platform was architected to support both New Zealand and Australian markets from a single codebase, with regional disclaimer management configurable by administrators.
RAG grounding vs hallucination	Rather than relying on the open internet, all AI responses are grounded in controlled, indexed proprietary documents. A guardrails layer strips unsupported assertions before they reach the user.

Financial Impact

Appraisal time reduction	From 30–45 minutes per appraisal to under 1 minute — saving approximately 40 minutes per appraisal
Appraisal volume	1,800–2,000 appraisals per month (60,000–67,000 annually)
Saving per appraisal	NZD $33.33 per appraisal (based on NZD $50/hr agent cost)
Annual appraisal saving	NZD $720,000–$800,000 per year from appraisal automation alone
AskKen AI research saving	10 hours of manual research saved per agent per month × 150 agents × NZD $50/hr = NZD $75,000/month
Annual AskKen saving	NZD $900,000+ per year in research and legal advisory time
Total annual savings	NZD ~1.75 million per year (combined appraisal + AskKen AI)
Microsoft contribution	NZD $20,000 from Microsoft recognising this as an industry-first Azure build

Benefits to the Client

Appraisal turnaround time reduced from 30–45 minutes to under 1 minute — freeing agents to focus on client relationships
Consistent, professional, branded appraisal reports across all regions and agents — no variation in quality
EMV accuracy of ±3% nationwide — matching or exceeding experienced agent assessments
AskKen AI gives every property manager instant access to legislation, case law, market data, and maintenance knowledge without leaving the platform
Agent profiles, contact details, and property images auto-populated via SSO and API integrations — zero manual formatting
New property data captured for off-plan and new builds feeds back into the Bayleys Data Lake — building proprietary market intelligence ahead of public availability
Compliance safeguards built in — 90-day appraisal expiry, regional disclaimer management, data source version control
Stress and cognitive load reduced for property managers — repetitive tasks automated, accuracy protected
Platform scales across New Zealand and Australia from a single codebase
Recognised by Microsoft with a $20,000 contribution as an industry-first innovation on Azure

Support & Next Steps

Peritos Solutions provided post-go-live support covering monitoring of AI output quality, EMV model tuning, and integration stability. The platform is designed for continuous improvement — as new tribunal decisions, legislation updates, and market data arrive, automated pipelines re-index the RAG knowledge base daily or weekly.

Planned next-phase enhancements include:

Extension to additional Australian markets and McGrath Real Estate agent base
Expansion of the Bayleys Data Lake with manually entered new-build and off-plan property data
Additional AskKen AI modules covering insurance policy guidance and financial modelling
Business intelligence dashboard tracking appraisal volumes, lead conversion, and referral rates from sales to property management
Further fine-tuning of the EMV model with expanded suburb-level correlation data

Looking for a Similar AI or Property Technology Solution?

Peritos Solutions specialises in AI-powered applications, ERP integrations, and cloud-native platforms across New Zealand, Australia, USA, and India. We are a Microsoft partner with hands-on experience delivering first-of-kind solutions on Azure.

info@peritosolutions.com | +64-212579909 | www.peritossolutions.com