Farul Haziq · case study
PropertyGuru listings needed a location you could trust.
turning listing creation from form filling into search and checks.
the brief
why it existed
The problem
property agents, PropertyGuru's customers, post listings through AgentNet. creating one was a long form: free-text property names, no location check and no duplicate detection. wrong listings had to be corrected by hand, and buyers lost trust in what they saw.
The goal
treat listing creation as a search and validation problem, not form filling, so every listing starts from accurate data.
My part
- led the end-to-end redesign of the listing creation flow.
- improved the property search, so agents pick a verified property.
- designed the AI description step, with limits that stop scams.
- facilitated about 20 agent interviews (run by a UX researcher) and the stakeholder workshop that set the direction.
- validated the design with agents, through to a beta with 200+ of them.
- worked with a PM, a design manager and the dev team, and added new components to design system 3.0.
the journey
how it actually went
-
Step 1 · the old flow
Where the bad data came from
across Malaysia and Singapore, custom locations were often fake or scams. our team had to check each one by hand, which was slow and hard.
-
Step 2 · interviews
Hearing from agents first
about 20 interviews in a month, run by a UX researcher, with me facilitating. one thing stood out: agents race to list first, so speed comes before everything. we designed around that.
-
Step 3 · workshop
Four rules for every screen
the stakeholder workshop cleared up the direction, and gave us four rules: accuracy over convenience, guide instead of restrict, use the white space, build confidence through clarity.
-
Step 4 · the new flow
Fewer pages, less typing
we merged pages, cut fields that no longer mattered, and let a verified property fill in its own details. one click writes the description.
-
Step 5 · beta
200+ agents before everyone
real use showed edge cases no prototype could. short feedback loops fixed them before the full rollout.
the mess in between
what we were up against
-
when agents couldn't find a property, a banner let them create a custom location. it fixed a real usability issue, and opened a bigger one.
-
some agents used it to make up locations for more visibility, so buyers saw listings in the wrong place.
-
titles and descriptions were free text. some agents typed gibberish to get past the step, others added their own contact details.
the work
what actually shipped
-
before: the old flow, on design system 2.0
Before the redesign
- a long run of pages and fields, with agents clicking "Next" again and again.
- agents typed a property name as free text, with no check that it was real.
- if the property wasn't found, a banner offered a temporary location instead.
- nothing caught duplicate listings, so wrong ones were fixed by hand later.
- the quality score and tips only came at the very end, on the summary.
-
a verified property on a map, and a report link if it's wrong
The new location step
- a new step narrows the search first, so less data loads while agents look.
- agents pick a verified property, and its state, city, township, street and postcode fill in.
- the map shows the pin, so agents can see it's the right place before moving on.
- two banners became one clear "report an issue" link, in place of custom locations.
-
one click for a headline and description, from verified data
One click for the description
- "Auto-fill with AI" writes the headline and description from PropertyGuru's data on the property and location.
- agents can edit the result, so the listing still sounds like them.
- descriptions are capped at 2,000 characters, and phone numbers are blocked to stop scams.
- no more gibberish to get past the step, and less room for contact details.
the calls
the decisions behind it
-
Report it, don't create it
we removed custom locations. agents now report a missing property, and PropertyGuru checks it first. fewer unverified locations meant fewer scam listings for buyers to worry about.
-
Fill it in, but keep it editable
agents race to list first, so we merged pages, cut fields and pre-filled details from verified data. they can still change the text, so the listing stays theirs.
-
AI that only writes what we know
the AI writes from PropertyGuru's own data on the property and location. i capped it at 2,000 characters and blocked phone numbers, to stop scams.
the outcome
what changed
- 80%+ fewer incorrect listing locations in the 3-month beta
- 200+ agents in the beta, across Singapore and Malaysia
measured over the 3-month beta with 200+ agents, against the 3 months before. those results gave leadership the confidence to roll it out to every agent.
reflection
what I'd tell past me
"i'd get everyone clear on the direction first."