Challenge
Visual search tools that work everywhere else fail in jewelry
The client's retail partners face a problem every day: a customer walks in with a photo of a ring they love and expects to find it, or something close, in the store. The old answer was to flip through binders, call a rep, or shrug and hope the customer came back.
The tools that were supposed to help made it worse. Generic image search models are trained on the world at large. In jewelry, that means they fixate on the most visually dominant element, the center stone, while ignoring everything that actually defines a ring's design character: the shank profile, prong style, basket construction. A search that ignores those elements does not just return poor results. It returns confidently wrong ones.
- Focus on the Wrong Attributes: Generic tools prioritize center stones, ignoring shank, prongs, and basket design
- Lost Sales at the Counter: Dead-end results had no recovery path. No match meant no sale.
- Fragmented Buying Flow: Retailers fell back on manual catalog searches or rep calls, slowing every decision
- No Custom Design Capture: When inventory did not match, there was no structured way to capture the custom request
If a customer could not find an exact match, they hit a dead end. A dead end in retail is a lost sale. And usually a lost customer.
Solution
Cogent Labs built a visual search engine that thinks like a jeweler
When the client brought this problem to Cogent Labs, the brief was clear: do not build a generic image search. Build one that understands jewelry, specifically the design attributes that matter to buyers and retailers, not the ones that dominate the image.
The result is a custom computer vision model trained to extract the features that define a ring's character: shank geometry, prong configuration, basket construction. Center stones are filtered out as noise. What remains is a high-fidelity representation of the design, one that matches the way trained jewelers actually assess similarity.
- Sana Commerce Integration: Embedded directly as a page inside the B2B portal, accessible via a header camera icon. No new logins or external tools.
- Jewelry-Specific AI Modeling: The model focuses on shank, prongs, and basket, not center stones. Built for the domain, not repurposed from generic search.
- Smart Tiered Result System: High matches (≥80%) surface first. A "Request Custom Design" CTA appears after top results. Medium matches follow. Users always see relevant options.
- Zero Dead-End Experience: If no strong match exists, the system shows closest visual neighbours and trending products. 0% zero-result rate, guaranteed.
- Full Conversion Tracking: Every search-to-PDP-to-order journey tracked via GA4. Custom design clicks captured as leads.
The system does not just return matches. It makes a decision about what the retailer actually needs, and then makes sure they always have somewhere to go next.
Results
The platform turned inspiration into inventory in seconds
Retailers no longer hit dead ends. When a customer brings in an inspiration photo, the answer is instant: a ranked set of matches, a path to custom design, and a direct link to the product page to close the sale.
- Higher conversion rates: users find relevant products instantly from any image
- Zero lost opportunities: dead-end search scenarios eliminated completely
- Increased custom orders: strategic CTA captures demand when inventory does not match
- Faster sales cycles: retailers identify products in seconds, not hours
- Full funnel visibility: search to results to PDP to order, tracked in real time via GA4
The system transformed the buying journey at the point of sale. What previously required rep consultation, manual catalog navigation, and follow-up calls now takes seconds. And because every interaction is tracked, the client now has real visibility into how visual search drives revenue at scale, across every retail partner.
Conclusion
From image to purchase in seconds.
The client's retailers built something valuable over years of experience: the ability to understand what a customer actually wants from a photo and translate that into a product recommendation. What they did not have was a way to make that capability available at every counter, on demand, without a trained expert standing behind it.
Cogent Labs built the system that does exactly that: a computer vision model that thinks like a jeweller, a results engine that never hits a dead end, and a buying flow that goes from image to product page to order without friction.
The same pattern applies to any domain where visual expertise is the constraint and scale is the goal. When knowledge is built into a system, it becomes available everywhere, without the expert having to be there.
0%
zero-result rate, users always see matches
Instant
image to tiered product matches
3-Tier
smart ranking engine with custom design capture
Have an idea worth building with AI?
Tell us about your workflow. We'll show you what a custom AI system can do.
Book a discovery call