Singapore agency AI tools: what ERA and PropNex have announced

In this article

Written by

Kooky

Builder of Shaka, the payment router that pays every agent their commission on closing date.

About Kooky and Shaka →

Singapore's estate agencies have been announcing artificial intelligence tools for more than three years. The announcements are of very different kinds: a press release, a slide in a results presentation, a feature in the regulator's own newsletter, a trade press report. Each says what a tool is meant to do. Few say how well it does it, and that gap is the subject of this guide.

The guide describes the position as read in October 2026. It sets out what ERA and PropNex have said in documents that could be read for it, what was and was not found for OrangeTee, Huttons and the property portals, and how the stated capabilities sit beside the advertising guidelines of the Council for Estate Agencies (CEA) and the personal data guidelines of the Personal Data Protection Commission (PDPC). Every capability below is the company's own claim, with its date. None was tested for this guide, and nothing here is a view on whether any tool works or is worth using.

ERA Singapore release of 18 August 2026; PropNex business update for the six months ended 30 June 2026. Company figures, not independently checked.

Which documents this guide rests on

Five company or company-related documents were read. For ERA: a press release published by ERA Singapore on 18 August 2026, a report by EdgeProp Singapore dated 7 March 2023, and a feature CEA published for the industry, marked accurate as at 24 May 2023, which reports ERA's own estimates. For PropNex: the business update for the six months ended 30 June 2026, an attachment to an announcement the listed company broadcast on the Singapore Exchange on 13 August 2026. For OrangeTee: an EdgeProp Singapore report dated 18 December 2020.

Two regulator documents complete the set: CEA's Practice Guidelines on Ethical Advertising, numbered PG 2/2011, in version 2.0 dated 10 December 2020, and the PDPC's Advisory Guidelines on use of Personal Data in AI Recommendation and Decision Systems, issued on 1 March 2024.

The list is short, and the limits matter. No annual report was read. No announcement by Huttons was found. The newsroom of PropertyGuru Group, a portal operator, was read as an index only. Where a company is absent from a section below, the reason is that nothing was read, which is different from saying nothing exists.

March 2023: ERA puts GPT-3 into its agent app

The earliest deployment in the documents is ERA's. EdgeProp Singapore reported on 7 March 2023 that ERA Realty Network had integrated GPT-3, a language model from OpenAI, into SALES+, the agency's own app for its agents, and that the function was available to ERA agents using the app from that day. The report gives the app's own launch as July 2022 and says the integration took ERA's development team about one month.

According to the same report, the function came in two modes. One, called Guided Creation, produced set kinds of text: property listing descriptions, listing content written for search engines, open house invitations, profile introductions, translations, tenancy reminders and personalised festive greetings. The other, called Advanced, let an agent converse with the model more directly for a wider range of responses.

The report also carries money and headcount figures, all ERA's. The agency had spent S$3 million on property technology in 2022, mainly on developing and rolling out the app, and planned to invest a further S$5.2 million in 2023. EdgeProp writes the amounts with a plain dollar sign; they are given here in Singapore dollars. Added together, the two years come to S$8.2 million, of which the second was a plan at the time of the report. ERA had about 8,600 agents registered, and more than 10,800 attendees had completed training modules on the app, a count that ERA said included repeat attendance.

What ERA told the regulator's newsletter two months later

CEA's feature of 24 May 2023 adds figures the EdgeProp report does not have, and one difference of wording. Where EdgeProp states that GPT-3 is what was integrated, CEA's feature says ERA integrated GPT-3 and later ChatGPT into the app, and dates the announcement to March 2023. Both are shown here as written.

The feature reports three ERA estimates. A salesperson could draft a property listing in a few minutes, where the task had taken 30 minutes to an hour. Use of the app had risen by more than 35 per cent since the launch, mainly for general queries and content generation. And the function was to be extended to all ERA offices in 11 Asia Pacific countries by the third quarter of 2023. The feature repeats the S$5.2 million pledge for 2023 and adds that ERA's technology team had grown threefold since the third quarter of 2022.

Two cautions belong beside those numbers. The 35 per cent rise is a rise in use of the app as a whole, as CEA words it, and no baseline is given. The time saving is an estimate by ERA of its own tool; neither document describes how it was measured. Whether the regional extension happened on schedule was not established from the pages read.

The same feature names one other agency. VIP Realtors, working with a local financial technology start-up, had used an AI chatbot since the middle of 2021 to process invoices and organise documents, and productivity was reported to have risen by more than 50 per cent after almost two years. That is back-office automation, a different thing from a tool that writes for the public. The feature does not mention PropNex, Huttons or OrangeTee.

August 2026: ERA-GPT, a pilot

ERA Singapore's release of 18 August 2026 announces a product called ERA-GPT and a training agreement with NTUC LearningHub. Both were presented, the release says, at the agency's mid-year leaders and achievers conference at Marina Bay Sands.

The release describes ERA-GPT as a conversational assistant built into the same app, a single point through which an agent can reach the app's other functions by asking a question. Its examples of requests are property or market information, guidance on ERA's own tools and workflows, and operational and compliance information. Support for 24 languages is to be rolled out progressively.

Three points of status are worth reading closely.

1. ERA-GPT is in a pilot phase. The release gives no launch date.
2. Several functions are described as possibilities, not as features: proactive recommendations, document analysis and the automation of selected workflows.
3. The release says the app holds more than 30 AI-enabled applications and over 500 capabilities, and does not name them.

The third point limits what can be said about the functions listed in this guide's brief. The release, as read, does not mention valuation estimates, listing writing, lead matching or virtual staging. Listing writing is documented for ERA only by the 2023 sources. Whether any of the 30 applications values a home or stages a room is not something the release answers.

On training, the release says the agreement with NTUC LearningHub covers 8,780 advisers and gives access to more than 200 AI training programmes. Marcus Chu, named in this release as chief executive of ERA Singapore, is quoted: "AI will not replace the trust between an adviser and a client."

ERA's usage figures and how far they go

The August 2026 release is the only document read that gives usage figures for AI functions specifically. ERA says more than 75 per cent of its agents already use AI capabilities within the app, and cites approximately 590,000 AI queries and up to 113,000 AI-generated property reports.

A little arithmetic shows the scale, with its assumptions stated. Taking the 8,780 advisers named in the same release as the base, 75 per cent is 6,585 people. Spread across all 8,780, the query figure is about 67 queries per adviser and the report figure about 13 reports per adviser. This is a worked example and nothing more: the release, as read, gives no period for either count, so the figures cannot be turned into a monthly or yearly rate, and the release does not say that the 8,780 is the base of its percentage.

What the figures do not show is equally plain. A query count says an agent asked something; it does not say the answer was right or was used. The phrase "up to" before the report figure is the release's own. The term "property report" is not defined in what was read: the release does not say what such a report contains, whether it includes a price estimate, or whether it goes to clients. ERA's agent count, for comparison, was about 8,600 in the EdgeProp report of March 2023 and is "nearly 9,000" in the release of August 2026.

PropNex: a leads platform and three suites

PropNex's statements come from a different kind of document. A business update is a presentation to investors, and its technology content is a few slides among financial ones. For the first half of 2026 the update reports revenue of S$603.0 million, against S$598.9 million a year earlier, a rise of 0.7 per cent, and profit for the period of S$43.5 million, against S$45.5 million.

On technology, the update says the Leads Engagement Suite was officially launched on 24 July 2026 at an event it calls the Singapore PropTech Convention 2026, attended by more than 1,400 PropNex salespersons. The document describes it as an AI-driven platform; the label is the company's. The listed functions are capturing leads, follow-up, which covers reminders, follow-up alerts and ready-to-use engagement content, and conversion. The update does not say what the AI in the platform does, which model it uses or what data it learns from.

The same slides describe three other suites without calling them AI. The Investment Suite is presented as market intelligence and location insight, with real-time trends, historical insights, property analysis, price comparisons and market indicators. The Business Suite includes a feature named SPOT Units, financial calculators and a builder for landing pages. The Agent Suite covers submissions, training bookings and company information. Price comparison is the nearest thing in the document to a valuation function, and the document does not describe it as an estimate of value.

Two sets of numbers in the update call for care. It gives PropNex 14,555 of Singapore's 38,162 registered salespersons as at 1 July 2026, citing CEA, which is 38.1 per cent. A separate slide says "over 18,000 salespersons" without a date or scope. The document does not reconcile the two, and both are shown here. On the first figure, the 1,400 at the convention were about 9.6 per cent of the salesforce. The update gives no usage figure for any app, no amount invested in technology, and no statement on accuracy. It does list awards it names as the SBR Technology Excellence Awards: a proptech award in 2024 and 2025 and an analytics award in 2026.

OrangeTee, Huttons and the portals: what was found

For the other large agencies, the documents read contain no AI announcement at all.

The one OrangeTee source is older and concerns a different technology. EdgeProp Singapore reported on 18 December 2020 that OrangeTee & Tie had first launched its Property Agent Review platform in 2016, that it held more than 28,000 client reviews, and that daily traffic had risen from an average of 150 unique visitors to between 650 and 700. The report says a client's identity is screened and verified and that agents cannot edit a review. It names no AI. The same report describes PropNex, ERA, OrangeTee & Tie and Huttons as the four largest agencies, each with at least 3,000 agents at the time.

For Huttons, nothing was read. For the portals, the newsroom of PropertyGuru Group showed, when read on 10 October 2026, six press releases from 2026, none of which carried AI, valuation, search or virtual staging in its title. No announcement by any other listing portal was located.

This is a finding about the research, with its stated budget of pages, and it settles nothing about the companies. Portals and agencies describe product features in places that were not read for this guide: product pages, help centres, app store notes, annual reports. Virtual staging in particular does not appear in a single company document read here, and so no company's virtual staging claim is reported.

What the companies say about data and accuracy

Statements about where a tool's information comes from are brief, and statements about how accurate it is are absent.

ERA's 2026 release says ERA-GPT draws on ERA's proprietary data, official government sources and relevant property information, and uses multiple AI models, none of them named. It adds that the assistant can take account of an adviser's profile, activity and previous interactions to personalise its help. The 2023 EdgeProp report names OpenAI's GPT-3 and says nothing about the data behind it. PropNex's update cites, as sources for its figures, the Urban Redevelopment Authority's REALIS system, the Housing and Development Board, Singapore's open data and its own data; it also cites its Investment Suite as the source of a table of bank interest rates updated on 27 July 2026. Those are sources for slides, not a description of what feeds an AI function.

No document read gives an error rate, a test result, a comparison with human work or a description of how output is checked before a client sees it. The nearest thing to a performance figure is ERA's 2023 estimate of drafting time, which measures speed.

How the stated functions sit with CEA's advertising rules

CEA's Practice Guidelines on Ethical Advertising were issued on 29 June 2011 and applied from 1 August 2011; the version read does not mention artificial intelligence, virtual staging or computer-generated images. Its rules are about the advertisement, whoever or whatever wrote it. Paragraph 3.1 says advertisements must not be inaccurate, false or misleading, and that claims about market trends are to be supported by statistics from reliable sources. Paragraph 3.2 says photographs must not be altered or enhanced in a way that misrepresents actual aspects of the property, and that a photograph used for illustration carries a qualifier beside it. Paragraph 6 requires the source of publicly available data used in an advertisement to be indicated, and paragraph 7 applies the rules equally online, with paragraph 7.1 asking salespersons to affirm that a listing's information and photographs are actual and not misleading. Read against the tools above, a listing description drafted by ERA's 2023 function is held to paragraph 3.1 like any other, and "ready-to-use engagement content" of the kind PropNex describes is advertising material when it promotes a property. The guidelines place the checking with people: the preamble requires estate agents to vet their salespersons' materials before publication, and paragraph 5 requires key executive officers to approve all advertisements or to have a structured process for doing so. None of the company documents read describes how its AI output passes through that vetting. How the guidelines apply to a given advertisement depends on its content, and CEA states that non-compliance may result in disciplinary action.

How they sit with the personal data rules

The PDPC's advisory guidelines of 1 March 2024 say, in paragraph 2.1, that the Personal Data Protection Act covers all collection and use of personal data by an organisation, including to develop, test and monitor AI systems or as part of deploying them; paragraph 2.2 adds that the guidelines themselves are advisory and that the Act prevails. Two of the stated functions touch personal data on their face. A leads platform, by its description, handles information about enquirers, and ERA says its assistant can use an adviser's profile, activity and previous interactions. Paragraph 9.1 of the guidelines says consent is generally required to collect and use personal data to provide recommendations, predictions or decisions, unless deemed consent or an exception applies, and paragraph 9.2 recalls the duty to tell individuals the purposes on or before collection. Paragraph 9.5 lists four items for organisations to explain to individuals; paragraph 10.3 asks for written policies proportionate to the risk. Where a supplier builds a bespoke system and processes personal data for the agency, paragraph 11.2 treats it as a data intermediary, and paragraph 11.8 leaves primary responsibility with the organisation that chose the system. The guidelines do not mention generative AI by name, and whether a given text-drafting function falls within their description of an AI system is not settled by the text read. None of the company documents read says on what basis client or enquirer data is used, whether it is used to train a model, or where it is processed.

What none of the documents discloses

Set side by side, the announcements leave the same questions open. The table lists them; "not stated" means not stated in the document read, and a company may say more elsewhere.

Five gaps stand out.

• Accuracy. No error rate, test or audit appears in any document, for text, for reports or for lead handling.
• The period behind the counts. ERA's query and report figures carry no dates in what was read, and its 2023 usage figure has no baseline.
• Client data. No document says whether clients' or enquirers' personal data enters a model, is used for training, or leaves Singapore.
• Human checking. No document describes who reviews AI-drafted text or reports before they reach the public, or how the agency's advertisement approval process treats them.
• Disclosure to the public. No document says whether a client is told that a description, a report or a message was produced with AI.

A company's count of queries shows that a tool is used. It does not show that the answers were checked.

Two further points concern the record itself. No annual report of PropNex or of ERA's group was read here, and an annual report may say more than a release or a slide. And several statements in this guide are plans as announced: the regional extension of 2023, the 24 languages, the future features of ERA-GPT. The documents that would show whether they were delivered are the companies' later announcements.

Kooky, from Shaka

Kooky edits Agents Estate and builds Shaka, the payment router he made for real estate professionals. One payment comes in, and every agent, agency and party in the deal receives their signed share on closing date.

About Kooky and Shaka →