At the 2026 Vicenzaoro jewelry show, one luxury brand ran a pretty gutsy campaign built around AI in visual search for products. The whole idea was to fix a classic e-commerce problem for high-net-worth shoppers: you see a piece of bespoke jewelry you love, but you can’t find it online because you don’t know the right keywords. They gambled that this tech could finally connect that visual inspiration to an actual sale.
Key Takeaways
- Visual search users converted at a rate 28% higher than people using old-school text search which immediately justified the AI spend.
- They sank $150,000 into AI model training and integration alone. It’s a big upfront number, but it paid off with measurable returns.
- Personalized retargeting ads, sent to people based on what they visually searched for, bumped up the average order value (AOV) by 15% among returning customers.
- The campaign’s real engine was a real-time feedback loop where user uploads immediately helped tune the product catalog, closing the gap between discovery and checkout.
- Initial problems with the image upload tool were a harsh reminder that you have to do rigorous UX testing, especially when you’re asking luxury clients to try something new.
“Forrester found that 94% of B2B buyers used AI during recent purchase processes. Of those, 55% used AI to compare vendors, 54% to research products, and 47% to build internal business cases, all before talking to a single sales rep.”
Campaign Teardown: Vicenzaoro’s AI Visual Search Initiative
Selling luxury jewelry online is tough because aesthetics and tiny details are everything. A customer might see a necklace in a magazine or on Instagram, but they’ll have a hell of a time describing its specific filigree or gem cut in a search bar. One major jeweler, let’s call them “Aura Gems”, saw this gap and launched a targeted campaign right at the 2026 Vicenzaoro show, the global stage for the industry. Their goal was to become the go-to platform for finding jewelry with a picture, using some serious AI to do it.
Strategy and Objectives: Beyond Keywords
Aura Gems’ plan was to completely move past keyword dependency. They wanted customers to just upload a picture of a ring or bracelet, and their system would spit out identical or similar items from their catalog. The main objective was to get more people interacting with their online collection and, obviously, to drive higher conversion rates from these visually-driven searches. As a bonus, they hoped to see bounce rates drop and to collect much better data on what their customers actually find beautiful. This captured purchase intent at its most primitive, visual source, which is exactly what you need for high-ticket sales. The entire campaign had a $450,000 budget to cover tech, marketing, and support over a tight six-month period.
The Technological Backbone: AI Product Search Implementation
The engine of the whole campaign was a custom-built AI product search. Aura Gems brought in a specialized AI firm to build a computer vision model from the ground up, training it on millions of images of jewelry to teach it every style, material, and gemstone cut imaginable. That training alone ate up a huge chunk of the tech budget, around $150,000, for all the data labeling, model tuning, and raw computing power. They built the system to do a few specific things:
- Feature Extraction: It could look at an image and automatically tag visual details like the metal (gold, platinum), the stone’s shape (round, pear), the setting (prong, bezel), and even general design styles.
- Similarity Matching: It converted uploaded images and catalog items into vector embeddings, a type of mathematical signature, to find and rank the closest visual matches.
- Real-time Learning: The model got smarter with every use, learning from which search results users clicked on and what they eventually bought.
They plugged the visual search tool right into the main e-commerce site and pushed it hard with dedicated landing pages. You could upload a photo from your phone or just paste an image URL. While they looked at off-the-shelf platforms like Google Cloud Vision AI for its solid object detection, they in the end decided that the weird, bespoke nature of luxury jewelry demanded a completely tailored model.
Creative Approach: Show, Don’t Tell
The creative was all about demonstrating the tool in action. At their Vicenzaoro booth and all over their digital ads, they used simple “before-and-after” visuals. You’d see a picture of some incredible necklace from a fashion shoot, then a screenshot of their visual search tool instantly pulling up near-identical options from their own stock. They ran short 15- to 30-second video ads on platforms like Pinterest Business and LinkedIn Ads that just showed how easy it was. The copy was dead simple: “See it. Find it. Own it.”
The smartest thing they did was set up an interactive kiosk at their Vicenzaoro booth. Attendees could snap a picture of any piece of jewelry they saw at the show, upload it right there, and see what Aura Gems had that was similar. This bit of experiential marketing created a ton of buzz and gave them a firehose of first-party data on what people were actually looking for.
Targeting: Precision for the Affluent
Aura Gems used a layered targeting strategy to get in front of people with money to spend.
- Geographic Targeting: They hammered major luxury hubs in North America, Europe, and the Middle East, with a special focus around the Vicenzaoro event itself.
- Demographic and Psychographic Targeting: They used anonymized third-party data to find people with high disposable income who followed high-fashion and art content. This also involved building custom audiences from lists of past luxury buyers and subscribers to expensive lifestyle magazines.
- Contextual Targeting: Ads were placed on jewelry blogs, high-end travel sites, and fashion magazines where their target audience was already hanging out.
- Lookalike Audiences: Building lookalikes from their existing high-value customer list was a particularly good move for finding new, relevant shoppers.
- Event-Specific Retargeting: With consent, they captured data from visitors at the Vicenzaoro booth and immediately started hitting them with personalized ads that referenced the visual search tool and the types of products they’d looked at.
What Worked: Data-Driven Success
The results were solid, and a few KPIs really stood out:
| Metric | Pre-Campaign Baseline | Campaign Result | Improvement |
|---|---|---|---|
| Conversion Rate (Visual Search Users) | N/A (New Feature) | 3.5% | Baseline for new feature |
| Conversion Rate (Text Search Users) | 2.7% | 2.8% | 0.1% |
| Click-Through Rate (CTR) on Visual Search Ads | N/A | 1.8% | Strong engagement |
| Cost Per Lead (CPL) for Visual Search Engagements | N/A | $12.50 | Efficient lead generation |
| Return on Ad Spend (ROAS) for Visual Search | N/A | 3.8x | Positive ROI |
| Average Order Value (AOV) for Visual Search Conversions | N/A | $8,200 | High-value transactions |
| Bounce Rate on Product Pages (Visual Search Origin) | N/A | 18% | Low, indicating relevance |
The biggest win, hands down, was the 3.5% conversion rate for people who used the visual search tool. That number crushed the 2.8% conversion rate from traditional text search during the same timeframe, a 28% jump in effectiveness. It’s clear proof that the users who tried the AI were highly motivated and found what they were looking for. The average order value (AOV) for these sales was also way up at $8,200, which suggests the tool was helping people find and buy the more expensive, aspirational pieces. Across their $300,000 in marketing spend (not counting the AI dev cost), they hit a 3.8x ROAS, a healthy return.
That kiosk at Vicenzaoro was a goldmine, pulling in over 1,500 unique visual searches over the four-day event. This gave the dev team immediate, real-world feedback on the model’s performance. The ads promoting the tool also did well, pulling a CTR of 1.8%, which showed that people were genuinely curious about the feature.
What Didn’t Work: Learning from Friction
It wasn’t all perfect. The biggest headache was the user experience (UX) for uploading images. The first version of the interface saw a 15% drop-off rate right at the upload step, which was way higher than they’d expected. People were getting tripped up by image resolution limits or slow uploads, especially on their phones with spotty Wi-Fi in the exhibition hall. Some just gave up before they even saw a result.
The AI model also had its own issues. It was great at identifying classic styles and standard gem cuts, but it got confused by really abstract or avant-garde pieces. For those niche categories, the search results just weren’t very good. That problem made it clear they needed to keep training the model on a much wider, more contemporary set of jewelry designs.
Optimization Steps Taken: Iteration and Refinement
Aura Gems moved fast to fix the problems:
- UX Enhancement for Image Upload: They immediately threw out the old upload interface and built a new one with clearer instructions, a progress bar, and client-side image compression to speed things up. They also bumped up the acceptable file size and added more format support. Those changes cut the drop-off rate at that step by 7% in just two weeks.
- Model Retraining and Expansion: Using the data they got from Vicenzaoro and the website, they started retraining the AI. They specifically fed it more images of complex and abstract designs from independent jewelers and high-jewelry collections to patch its weak spots.
- Personalized Retargeting Sequences: If someone used the visual search but didn’t buy, Aura Gems didn’t let them go. They set up retargeting sequences that showed ads with products visually similar to the original search, sometimes at different price points. That move alone resulted in a 15% higher average order value (AOV) from returning customers who came back to buy.
- A/B Testing of Search Result Layouts: They started A/B testing the results page, playing with grid sizes and how much product info to show. This constant tweaking led to a 5% lift in click-throughs from the results page to the actual product pages.
- Integration with Customer Service: They created a special process for customer service reps to manually review bad visual search results. This “human-in-the-loop” system not only helped save a sale when the AI failed but also provided perfect data for the next round of model training.
The Aura Gems campaign at Vicenzaoro proved that putting real money behind AI in visual search for products isn’t just a gimmick in the luxury space. It actually works, driving serious conversions by respecting how wealthy customers actually discover things they want to buy.
For any luxury brand trying to stand out, this kind of visual search tech is becoming table stakes. It’s a way to reshape the entire customer journey, making online discovery feel as intuitive as the products themselves. The next phase of luxury e-commerce is going to be driven by intelligent systems that find what a customer desires before they can even type it. How AI commerce is evolving will have a huge impact, so it’s smart to also get your E-commerce SEO in order so people can find you in the first place.
What is AI visual search for products?
It lets people use an image as their search query instead of typing words. An AI model looks at the picture you upload and pulls up similar-looking products from an online catalog.
How does visual search benefit luxury e-commerce?
It solves the problem of customers not knowing the right words to describe intricate or unique luxury goods. It closes the gap between seeing something you want and actually finding it online, which is perfect for visually-driven industries like jewelry.
What kind of data is needed to train an AI visual search model for jewelry?
You need a huge and diverse image dataset covering every possible style, material, gem type, cut, and setting. Importantly, every image has to be carefully labeled with its specific attributes so the AI can learn what makes one piece different from another.
Can AI visual search improve conversion rates?
Yes, absolutely. By giving people search results that are incredibly relevant to their visual taste, it removes a lot of friction and gets them to the “buy” button faster. The Vicenzaoro campaign saw a 28% conversion rate increase for this reason.
What are common challenges when implementing visual search?
The big ones are getting the model’s accuracy high enough across a wide range of products, making the image upload process painless for users (especially on mobile), and committing to the continuous cost and effort of retraining the AI. The initial investment in the tech and data labeling is also pretty steep.
