Visual Search for Tile Suppliers in Singapore: A Practical Guide
Most people renovating an HDB flat or condo do not shop for tiles by product code. They shop by look. They arrive with a Pinterest pin, a Xiaohongshu screenshot, a blurry photo from a showroom visit, or a picture their interior designer forwarded on WhatsApp. Then they try to find that look on your website with a search box that only understands "porcelain 60x60 matt".
Visual search closes that gap. A buyer uploads a photo, and your site returns the closest-looking tiles from your own catalogue, ranked by similarity. It does not identify the exact SKU from a picture, and anyone who promises that is overselling. What it does well is turn "I want something like this" into a shortlist of five or ten real products in seconds.
This guide explains how it works in plain English, what your catalogue needs, where it falls short, and how to test it on a slice of your range before committing to anything bigger.
Why tile and stone suppliers lose look-first buyers
Search results for "tile supplier Singapore" are dominated by listicles ranking the big showrooms. They help a homeowner decide which showroom to visit. They do nothing for the moment that matters to you: when a buyer is on your site, holding a photo, trying to find a match.
Here is how that moment usually goes wrong.
- ▸The words do not match the picture. A homeowner sees a warm, veined stone in a café in Tiong Bahru. Is it Calacatta-look? Travertine-look? Honed or matt? They do not know your vocabulary, so they type something vague, get 400 results, and leave.
- ▸Filters assume expertise. Size, finish, body, PEI rating, rectified edges. These filters are essential for your trade customers and meaningless to a first-time BTO owner.
- ▸The photo ends up in a WhatsApp thread. The buyer sends the picture to your sales line. Your staff scroll through the catalogue from memory, reply with three options, the buyer says "not quite", and the thread stalls. Multiply that by every enquiry during the peak renovation months after key collection.
- ▸Interior designers move faster than you reply. An ID juggling several projects will shortlist from whichever supplier gets them to "yes, that one" fastest. If your competitor's catalogue is easier to browse by look, you lose the specification before a price is ever discussed.
Some large tile brands overseas already offer "search by photo" on their sites, so buyers are starting to expect it. Most Singapore suppliers still rely on showrooms, PDF catalogues and WhatsApp. That is an opening, not a crisis.
How visual search works, in plain English
You do not need to understand the maths to make good decisions about visual search. You do need to understand three ideas.
1. Every image becomes a fingerprint
The system reads each catalogue photo once and records its look — colour, veining, texture, pattern scale, contrast — as a kind of visual fingerprint. Two photos of similar-looking tiles produce similar fingerprints, even if one is a close-up and the other is a bathroom wall.
Your whole catalogue, fingerprinted, becomes what the system searches.
2. The buyer's photo gets a fingerprint too
When a buyer uploads their Pinterest pin, their photo is fingerprinted the same way, in a moment.
3. The system ranks by closeness
The system compares the buyer's fingerprint with every product fingerprint and ranks them by how close they are. The top results are the tiles that look most like the photo. You can then apply normal business rules on top: hide out-of-stock items, show only floor-rated tiles, or boost a current promotion series.
That is the whole mechanism. The quality of the result depends far less on clever algorithms and far more on the quality of your photos and data, which is where most of the real work sits.
We have built a working tile visual-search prototype: photograph a tile and it returns the closest matches from a catalogue. It is a prototype, not a shipped product, and it is what taught us most of the lessons in this guide. You can read more on our visual search page.
What your catalogue needs first
Visual search is only as good as what it searches. Before anyone writes code, audit two things.
Photo standards
Your product photos are the fingerprints buyers are matched against. Inconsistent photos produce inconsistent matches.
| Standard | Why it matters | Common problem |
|---|---|---|
| Flat-on face shot for every SKU | The model compares like with like | Only room-set images available |
| Even, colour-neutral lighting | Stops warm or cool casts skewing colour matches | Showroom spotlights turning white tiles yellow |
| Several faces for variable designs | Porcelain with multiple face variations looks different tile to tile | One face photographed, the others never shown |
| Consistent crop, no watermark or logo overlay | Overlays pollute the fingerprint | Supplier-branded images from the factory |
| At least one installed or room shot | Helps match buyer photos of finished spaces | Only swatches, no context |
| Real slab photos for natural stone | Every slab is different; the swatch may not represent stock | One generic marble photo used for years |
If your images came straight from overseas factories in different styles, expect some reshooting or colour correction. It is usually the single best investment in the whole project, because it also improves your normal product pages.
SKU data
Visual search finds the look. Your data decides whether the result is usable. For each SKU you want, at minimum:
- ▸Size and thickness
- ▸Finish (matt, polished, honed, textured, lappato)
- ▸Body (glazed porcelain, full-body porcelain, ceramic, natural stone)
- ▸Application (wall, floor, outdoor, wet area)
- ▸Slip rating where relevant
- ▸Series or collection name, and colour family
- ▸Stock status, or at least "available" versus "indent"
Without this, a buyer can find the perfect-looking tile for their bathroom floor and discover later it is a polished wall tile. That is worse than no match at all, because it burns trust.
Honest limits: what visual search cannot do
Be clear about these with your team and your customers. Overselling visual search creates exactly the frustrated WhatsApp threads you were trying to avoid.
- ▸It matches look, not the exact SKU. Many tiles from different factories look nearly identical in a photo. Visual search will surface the closest ones in your range, which is useful, but it cannot confirm "this is the same tile as the one in that café".
- ▸Size and scale are hard to read from a photo. A 600x1200 slab and a 300x600 tile with the same pattern can look alike in a tight crop. Grout lines help, but not always.
- ▸Physical properties are invisible. Slip resistance, water absorption, thickness, rectified edges and body type do not show up in a picture. Your filters and staff still handle these.
- ▸Photo quality matters. A dark, angled, heavily filtered Instagram shot gives the model less to work with. Results from clear, flat-on, daylight photos are noticeably better.
- ▸Lighting shifts colour. Warm downlights in a condo bathroom, or an Instagram filter, can make a cool grey tile look beige. Good systems let the buyer crop to the tile area, and your staff should expect the occasional off-colour shortlist.
- ▸Finish is tricky. Reflections on polished surfaces can confuse matching, and matt versus honed is often a judgement call even for humans.
- ▸Natural stone varies. A marble slab in your yard may look quite different from the photo in your catalogue. For stone, treat visual search as a way to find the right type and colour family, then show real slabs.
None of these kill the idea. They just mean visual search is a shortlisting tool, and the confirmation still happens with samples, showroom visits and a conversation.
Is visual search worth it for your business?
Use this table as a rough filter before you spend anything.
| Signal | Good fit | Probably not yet |
|---|---|---|
| Catalogue size | Hundreds or thousands of SKUs | A few dozen products staff know by heart |
| How enquiries arrive | Lots of "do you have something like this?" photos | Mostly trade orders by product code |
| Customer type | Homeowners and IDs choosing by look | Contractors re-ordering the same lines |
| Photo quality | Consistent, or you are willing to fix it | Scattered factory images, no plan to improve |
| Data quality | Size, finish and application recorded per SKU | Specs live only in PDFs or staff memory |
| Sales process | Online shortlist then WhatsApp or showroom | Walk-in only, no online catalogue at all |
If most of your answers land in the right-hand column, fix the basics first: a clean, searchable online catalogue with proper filters and good photos. A well-structured website with SKU-level pages will also help you get found on Google and in AI answers, which our SEO and AI visibility work covers.
How to run a small pilot before committing
The worst way to adopt visual search is to try to index your entire range at once. The best way is to prove it on one slice.
Step 1: Pick one category
Choose the category where look-first enquiries are most common. For many Singapore suppliers that is marble-look porcelain, terrazzo, wood-look planks, or feature mosaics for bathrooms. A few hundred SKUs is plenty.
Step 2: Clean the photos and data for that slice only
Apply the photo standards and SKU fields above to this category. This keeps the effort contained and shows you what a full clean-up would involve.
Step 3: Build a real test set
Go back through past WhatsApp and email enquiries and collect genuine customer photos for that category. Ask permission where needed, and remove anything that identifies a person or a home, such as faces, unit numbers or addresses. Customer photos are personal data in many cases, so handle them in line with your usual PDPA practices; PDPC publishes guidance for organisations.
For each test photo, have an experienced salesperson note which tile they would recommend.
Step 4: Measure honestly
Run each test photo through the prototype and check one simple question: does the tile your salesperson picked appear in the top five results? Also note where it failed, and why. Was it lighting, a bad crop, a missing photo, or a genuine lookalike from another series?
Step 5: Decide
| Pilot result | What to do next |
|---|---|
| Staff pick usually in the top five | Put it in front of real buyers on that category and watch enquiries |
| Mixed results, failures explained by bad photos | Fix photos, rerun the test set before expanding |
| Poor results even with good photos | Stop, or narrow the use case, for example staff-only lookup |
A useful middle option: launch it internally first. Let your sales team use visual search to answer WhatsApp photo enquiries faster. If it helps them, it will help customers.
How visual search fits with WhatsApp enquiries
In Singapore, the tile sale usually closes in a conversation, not a checkout. Visual search should feed that conversation, not bypass it.
A practical flow looks like this:
- ▸The buyer uploads a photo on your site and sees a shortlist of closest matches.
- ▸They filter by application, for example "floor, wet area".
- ▸Each result has an "Enquire on WhatsApp" button that opens a chat with a pre-filled message listing the SKU codes they are looking at.
- ▸Your sales team replies knowing exactly which products are in play, and can suggest samples or a showroom visit.
That simple click-to-chat link turns a vague "do you have something like this?" into "I'm interested in these three SKUs for my kitchen floor". It saves several rounds of messages per enquiry and gives you a record of which looks people are hunting for. That data is useful for purchasing and for your marketing, including the ads and content we discuss in our guide for renovation businesses.
On the staff side, the same tool can sit behind the counter. When a buyer sends a photo straight to WhatsApp, a salesperson drops it into visual search and has a shortlist in seconds instead of scrolling the catalogue from memory.
Mistakes to avoid
- ▸Indexing everything on day one. Start with one category and prove it.
- ▸Skipping the photo audit. Bad photos in, bad matches out. No model fixes a yellow-tinted swatch.
- ▸Promising exact matches. Say "find similar looks", not "find this exact tile".
- ▸Ignoring stock. Showing a perfect match that was discontinued two years ago is a fast way to annoy a buyer.
- ▸Hiding the WhatsApp step. The shortlist should always lead to a human.
- ▸No measurement. Track uploads, shortlist clicks and WhatsApp enquiries from the tool, or you will never know whether it paid off.
Thinking about visual search for your catalogue?
If you supply tiles, stone or surfaces in Singapore and your team spends hours answering "do you have something like this?" photos, visual search may be worth a small pilot. It may also turn out that tidying your catalogue and product pages is the better first step, and we will tell you that honestly.
Have a look at how we approach it on our visual search page, then drop us a WhatsApp message with a rough idea of your range and how enquiries come in. We will tell you whether it is worth doing, and what a sensible first slice would look like.
Shawn Liu
LinkedInCo-Founder & Technical Director
Shawn Liu is Co-Founder and Technical Director of Rampage Connect, a Singapore digital studio. He builds the studio's products and client systems hands-on — from conversion-focused websites and web apps to AI agents, operations software, visual search and generative-AI video. His builds include Onz (onz.sg), a live agentic events site for Singapore whose "My Day" planner turns a one-line request into a day plan of real events, meals and travel times grounded in live NEA weather and OneMap data; a workforce portal for a Singapore SME covering GPS-geofenced check-in, shift rostering and leave approvals; a tile visual-search prototype that matches a customer's photo to the closest catalogue products; and a generative-AI luxury spec ad produced end to end with AI footage, compositing and grading. Shawn's working rule is simple: AI is only useful when it is grounded in real data and solves one specific job — so every build starts small, is tested on real cases, and earns the next one.
Ready to Play?
Let's Build Something Worth Playing
Turn your next campaign into an experience your audience chooses.
Let's Play →