A C-SUITE VIEW ON AIWe speak with leaders about how their companies are putting artificial intelligence to work. In this edition, Tech in Asia talks to Yi-Wei Ang, chief product and technology officer at PropertyGuru. Ang, who joined the company around two months ago, shares how a nearly 20-year-old company is building toward AI adoption without losing its footing. The interview has been edited for brevity and clarity.  A lot of tech companies say AI has transformed how their teams work. What does that actually look like at PropertyGuru day to day? We have a three-part framework for how we think about AI. We call it defend, extend, scale. Defend is about how AI might disrupt our business model and what we need to do about that. Extend is how we use AI to deliver more value across our two-sided marketplace. And scale is the internal-facing one - how we change the way everybody works. One thing we've had to figure out is tooling: what to provide everyone in a safe, secure way where people can use company information without it being compromised. We also spent a lot of time on training. We have ambassadors across different functions - finance, legal, marketing - and there are usually two or three trailblazers in each who chart the path and bring everyone else along. In engineering, that's where you see the largest adoption. Right now, we have groups piloting Claude Code, Gemini CLI, and Codex. We're tool agnostic for the most part, but I can already tell that teams that have adopted AI have much higher throughput, and we're starting to measure that through pull requests. Has there been any occasion where AI disappointed you or your team? Fortunately, compared to a lot of the bad news I see online, I haven't been in that situation. That said, I think the biggest disappointment is the gap between the promise of AI and what you can actually achieve overnight. If you were a fresh company building from scratch, maybe you can move faster. But we have 20 years of history to work through. The biggest challenge is trying to match reality with what the world is talking about and convincing everyone around you, whether it's leadership or engineers, that we will get there, but we'll get there in stages. You mentioned that in order to really scale AI, you need to invest in the foundations first. What does that look like in practice? One very concrete example is we built a skills repository inside the company. In data analytics, we have a lot of Looker dashboards, but maybe 20 to 30 of them are the gold standard ones. We now have a common markdown file that says: if you're using Looker MCP to answer questions, point to these dashboards - this is the source of truth. Before that, every time you connected to Looker, it would search through thousands of dashboards and not know which one was right. We're doing the same in our repos - explaining and documenting things not just for humans, but for AI tooling. This folder does this. This repo does that. These are the five things it does. Investing in that foundation is what allows us to move faster. Where has AI actually moved the needle for your users? Because we're a multisided marketplace, I think about this for both consumers and agents. One area is video. When we look at property listings, most websites still show you 10 pictures and some descriptions. A video does a much better job. Property agents are very busy - a lot of them are one-person businesses - and they don't have time to open a video editor. So we let them automatically generate a listing video from the images and descriptions they already have, with voiceover on top. Adoption on that has been strong: double-digit percent of our agents have tried it. Another is image moderation. We have millions of images coming through daily, and the only way to check quality and content guidelines at that scale is AI. Our AI moderation engine handles about 8 million images today. And then there's pricing. Buying a property is one of the biggest purchases most people will ever make. We've built price estimate capabilities, and our ability to predict property values is getting much better as we invest in first-party data. That benefits both agents, who can price their units better, and consumers, who can make more informed decisions. What's your assessment of where model providers stand today and what matters most to you as someone building on top of them in Southeast Asia? A year ago, I would have said: stop giving me generic models - they produce hallucinations and generic outputs. But in the last six months, we've seen a rapid shift toward models fine-tuned for very specific tasks. Those wishes are starting to come true. The one thing I'm watching closely is Chinese models and their cost efficiency. To win in Southeast Asia, cost is something you can't ignore, and the ability of some of these models to be significantly cheaper is a real competitive advantage. How that plays out over the next year or two will be interesting to follow. |
