AI in Malaysia · Property development

33,094 unsold homes, and loan rejection is the top reason. Where AI helps a Malaysian developer

Developers blame end-financing rejection more than price for unsold units, sale and purchase agreements went digital in January 2026, and every defect starts a 30-day clock. The useful AI for a developer works on the loan file, the agreement data and the defect list.

Pikabyte Engineering 7 min read

33,094

completed homes unsold in Malaysia in the first half of 2026, worth RM17.78 billion, according to the National Property Information Centre

Malaysia had 33,094 completed and unsold homes in the first half of 2026, worth RM17.78 billion, according to the National Property Information Centre (NAPIC). Six months earlier the figure was 30,471. Unsold serviced apartments rose to 23,375 units worth RM19.33 billion.

The Real Estate and Housing Developers’ Association (REHDA) asked its members why. In its survey for the first half of 2026, covering 181 developers in Peninsular Malaysia, 59% had unsold completed units at 30 June. The most common reason given was end-financing loan rejection, at 30% of responses. High pricing came second at 21%.

In the same survey 77% of respondents faced financing issues, and homes priced between RM500,001 and RM700,000 had the highest reported rejection bracket, at 31% to 45%. The Edge reported in November 2026 that housing loan approval rates fell to 38.36% in September, from 40% to 41% between 2022 and 2025.

Costs moved the other way. Nine in ten of the developers who answered reported construction costs up by an average of 13% between March and June 2026.

A developer cannot change a bank’s credit policy or the price of earthworks. It can change how well it knows a buyer before taking a booking, how clean the agreement data is, and what happens after a buyer reports a defect. Those three are information problems, and AI is useful on all of them.

The agreement went digital in January 2026

From 1 January 2026 the housing ministry requires developers to prepare and sign sale and purchase agreements digitally through its Housing Integrated Management System (HIMS), according to Sime Darby Property’s guidance to its buyers. The requirement covers homes under construction that are governed by the Housing Development Act. The guidance says the electronic agreement has the same legal effect as the printed one and is stamped once all parties have signed. Buyers verify their identity through the iDsaya app.

Law firm Azman Davidson & Co noted in May 2026 that practice may stay hybrid for a while. Some land offices still want physical signatures, banks may ask borrowers to sign financing documents by hand, and large launches can strain the system.

Stamp duty is changing as well. The Inland Revenue Board’s self-assessment system began on 1 January 2026 with tenancies, securities and general instruments. Transfers of real property that do not involve a government valuation follow on 1 January 2027.

The practical effect is that a buyer’s name, identity number, unit and price now pass through several systems without anyone retyping them. A mistake made at booking travels all the way to the stamped agreement.

Where AI is useful today

1. The first enquiry. In April 2026 S P Setia said it had appointed an AI voice and messaging startup to handle initial customer enquiries, qualify leads and pass eligible prospects to its sales team. The work won the real estate category at the Malaysia Technology Excellence Awards. For a mid-sized developer the same idea is an assistant on the project’s WhatsApp number that answers from the approved price list and unit availability, and asks the questions a salesperson would. Meta’s rules for such assistants apply.

2. The loan file, before the booking. If loan rejection is the leading cause of unsold stock, the place to act is before the unit is taken off the market while a bank decides. A model can read payslips, EPF statements and bank statements into one record and list what is missing. Tested code works out the commitments against income. A mortgage officer then tells the buyer honestly which banks are worth approaching, or which unit fits. We describe the reading step in how an LLM reads a statement of account.

3. The agreement data. Before a file goes to HIMS, software compares the booking form, the identity document and the loan offer field by field. A model reads the scanned documents. Any mismatch in a name, a number or a price is fixed before signing, not after stamping.

4. The defect list. This is where the law sets the clock. The defect liability period is 24 months from vacant possession. According to the National Housing Department’s guidance, as summarised by EdgeProp, a developer must repair a reported defect within 30 days. If it does not, the buyer can notify it of the repair cost and, after a further 14 days, recover that cost from the 5% stakeholder sum.

Defect reports arrive as photographs and short messages, hundreds at a time after a handover. A model can sort each one by trade and location, spot the same crack reported twice, and draft the work order. The system assigns it to the contractor and shows which reports are on day 25. UEM Sunrise said in 2024 that its buyers mainly used its app for key handover and defect appointments.

5. Finding defects before the buyer does. S P Setia also described analysing images from helmet-mounted cameras with AI to flag potential defects during construction and before handover. CIDB’s QLASSIC scheme scores workmanship independently. In REHDA’s survey, 45% of respondents had QLASSIC-certified projects in the past ten years, and 51% agreed it should apply to all residential developments. Inspection results kept as structured records are what make either approach useful on the next project.

What it should not do

Write the advertisement unchecked. The housing development regulations require an advertisement to carry the licence and permit numbers, the tenure, the expected completion date, the selling price and more, and they make a misleading statement in an advertisement an offence. An image model will happily add a lake. Treat generated copy and renderings like any artist’s impression: checked against the approved plans by a named person.

Tell a buyer they will get a loan. A model can check that a file is complete. The bank decides. Guidelines issued by the Personal Data Protection Commissioner on 30 April 2026 say AI should not be relied upon as the sole basis for a decision about a person.

Calculate late delivery damages. The statutory agreements set vacant possession at 24 months for landed homes and 36 months for strata homes, with damages at 10% a year of the purchase price. That is arithmetic for tested code. It does not belong in a chat reply.

Promise a repair date or compensation. The assistant reports the status in the system. A person commits the company.

Buyers are checking

The housing ministry counted 303 “sick” private housing projects in Peninsular Malaysia at 30 June 2026, with 43,288 units, and 100 confirmed abandoned projects involving 14,961 buyers. It urges buyers to check a developer’s licence and project status on its TEDUH portal before buying.

When a deal goes wrong, buyers have somewhere to go. The Tribunal for Homebuyer Claims hears claims of up to RM50,000, and its awards in 2024 totalled RM26.4 million. A defect record that shows when each report was received, assigned and closed is the developer’s best answer in that room. A WhatsApp group is not.

The law is moving

In November 2025 the government announced a new law to replace the Housing Development Act and extend regulation to commercial developments. As of the end of July 2026 the ministry said it was in the final stage of drafting. Build systems so that periods, rates and document lists are settings, not code scattered through the application.

Buyers’ data

A developer holds identity documents, income records and bank statements for every applicant, including those who never completed a purchase. Decide how long the unsuccessful files are kept. Know where any AI provider processes them. The duties are set out in the amended PDPA for software teams.

An order of work

  1. Start with defects. The clock is statutory, the volume is high and the data is yours.
  2. Add the loan file check at booking, and track rejection rates by project and price band.
  3. Check agreement data against source documents before it reaches HIMS.
  4. Put an assistant on first enquiries, answering only from the approved price list.
  5. Keep inspection records from every project in one place.

None of this sells a unit that is priced wrong. It does stop a developer losing sales, and goodwill, to its own paperwork.

Sources

This article is general information and is not legal advice.

Written by

Pikabyte Engineering

Pikabyte builds custom business systems, applied AI and secure cloud infrastructure from Malaysia. Talk to us.

Start a conversation

Want this applied to your business?

Tell us what is slowing the business down. You will talk to an engineer, not a salesperson, and get a straight answer on whether software is the right fix.