AI Image Enhancement for Property Listings
Carousell
A case study exploring the intersection of Generative AI, monetization strategy, and marketplace trust
Role
Product Designer
Timeline
Platform
Beta launch in Carousell app/website
User problem
Property agents on Carousell faced mounting costs and time pressure managing multiple listings. Professional photography and staging were cost-prohibitive when handling 10-20+ properties simultaneously, forcing agents to either absorb costs or publish suboptimal photos that reduced listing performance.
How does this affect our platform?
Property listings on Carousell also faced several challenges:
Quality inconsistency: User-generated photos often showed cluttered spaces, poor lighting, or distracting personal items
Competitive disadvantage: Professional agents had access to staging and professional photography; individual sellers did not
Conversion impact: Lower quality images correlated with reduced inquiry rates and longer time-to-sale
The proposal
A beta program offering AI-powered image editing, enabling 85-95% cost savings for agents while democratizing access for individual sellers as a secondary benefit.
The objective
To understand the demand for such a tool & the probable monetisation potential with a tool like this.
Provide a professional-seller tool for property agents to be able to enhance the quality of their listing images.
Context of Singapore Property Market
Agent-centric markets
Singapore's property market is dominated by professional and part-time agents who face unique pressures:
High-value transactions (HDB: S$500K-S$700K; Private: S$1M+) requiring professional presentation
Competitive marketplace where listing quality determines inquiry rates
Growing proptech adoption among forward-thinking agents
Carousell's unique position
While platforms like PropertyGuru and 99.co primarily serve established agencies with dedicated marketing budgets, Carousell attracts a significant base of:
Part-time agents building their portfolios
Independent agents without agency support infrastructure
Emerging agencies seeking cost-efficient tools
Individual sellers (secondary benefit) who list occasionally
Discovery & Research
We surveyed 14 property sellers where each evaluated 5 property examples, each showing 4 versions:
Original
Empty
Style 1
Style 2
They selected which version(s) they'd use for each property.
Some examples of properties shown. Respondents chose which version(s) they'd publish.
What agents told us
User Preferences for AI processing features (n=60)
Empty + Style
40.0%
Style only
21.7%
Empty only
20.0%
Original
18.3%
If this image tool was available on Carousell at a fee, would you use it?
Will use either or both tools
85.7
Will not use
14.3%
Price acceptance (10 coins/image)
Will use either or both tools
64.0%
Will not use
18.3%
What this means: Agents wanted flexibility and options, not a single feature. The 34% who chose "both" across examples suggests demand for a comprehensive toolkit.
Additional Insights from Open Responses
Agents who supported the tool cited:
"I may use it if it is still tenanted and the place is suppose to be cleaned up for next tenant so new tenant will not see the untidy house"
"It's more convenient if I can edit the photos and do the listing on a single platform at a go."
"I would use the tool as first impressions on the platform count."
The 16% Who Rejected Enhancement
Not everyone wanted AI editing. Across all 5 examples, some agents consistently chose "Original only". Quotes include:
"I will show my client the true image of the house they are going to buy or rent"
"In styling a photo, I prefer minimum editing to retain integrity of my property."
Pricing & Monetization strategy
Research validated 78% would use the tool and 64% found 10 coins per image acceptable. Now we needed to turn that validation into a sustainable business model.
Why 10 Coins per image
Our pricing balanced three priorities:
Affordability
Low enough for agents to process 10-20 listings per month without budget strain
Sustainability
High enough to cover API costs and generate platform revenue
Sustainability
85-95% cheaper than professional alternatives
Research validated 78% would use the tool and 64% found 10 coins per image acceptable. Now we needed to turn that validation into a sustainable business model.
Cost comparison: Traditional vs AI tool
Professional staging
S$2,000 - S$4,000
Professional photography shoot
AI Tool (~10 images)
Pay-per-usage model
✓
Lower barrier to trial: Process 1 image to test before committing
✓
Flexible volume: Works for agents with variable listing loads
✓
Flexible volume: Works for both one-time sellers and frequent listers
This "generate free, pay on publish" approach built trust where agents could review AI outputs before committing payment.
Risk assessment
This is where business strategy met ethical considerations. Since this is a new beta launch feature, we allow cheaper API costs to let user run watermarked previews for free. This trust building would hopefully net greater user usage in future and offset profit margin
User flow (MVP)
Despite survey preference for virtual staging, we built empty room first for speed and cost. Here's how I designed the MVP to maximize trust and adoption.
Upon curation of images, we immediately surface "Declutter your listing" as a seller tool with visual preview of before/after
Prevent surprise costs and abandonment at payment with disclaimer
Notification to prompt users to check results
Side-by-side comparison lets sellers reject bad outputs with no charge. Only pay for images they actually publish.
Mandatory "AI processed" labelling. Both versions are always visible to build transparency among sellers/buyers
Next step
Current buyer view shows processed images sequentially in the gallery. Cognitive load is high and buyers must mentally compare across multiple images. I would propose a better optimized view such as a slider or toggle once engineering effort is justified from this MVP experiment. We can also track data of rejection rate of AI processed images. Knowing which type of images let us know the limitation of such 3rd party tools and explore more alternatives.
Articles that reflect such issue
https://www.straitstimes.com/life/property-agents-are-turning-to-ai-generated-images-to-sell-homes-in-singapore
https://www.scmp.com/week-asia/lifestyle-culture/article/3327881/how-ai-redefining-singapores-home-renovation-landscape-its-so-easy