Residence Supply understood this problem firsthand. As a retailer known for quality interior lighting, from handcrafted wall sconces to modern ceiling lights that trade professionals and homeowners return to consistently, the catalog was strong. Conversion for shoppers who found the right product was high. The issue was that too many shoppers weren't finding it at all.

What the Data Showed Before Implementation

Before deploying an AI-powered search solution, Residence Supply was experiencing site search exit rates above 40%. Internal search contributed significant traffic but was punching well below its weight in revenue terms, accounting for under 20% of total transactions despite representing a far larger share of high-intent visits.

The catalog presented a real structural challenge. Home lighting is a visually led, preference-driven category. Two shoppers searching for the same type of fixture might want completely different aesthetics. Standard keyword-based search treated both queries as identical. Neither shopper felt like the site understood what they were looking for.

A 2026 Baymard Institute study on e-commerce usability found that 68% of product searches on retail sites fail to match actual shopper intent. The failure rate is highest in categories where style and use case vary significantly. Home lighting fits that description precisely.

The Solution: AI Search Built Around Intent, Not Just Keywords

Softcircles deployed a custom AI search layer that interpreted queries semantically rather than matching them to product labels literally. When a shopper searched "dining room ceiling light," the system didn't just scan title fields. It analyzed browsing behavior, session signals, and catalog attributes to surface fixtures that matched that shopper's style pattern.

The personalization layer worked in parallel with search. Returning visitors saw results shaped by prior interactions. New visitors were served results weighted toward bestsellers within their detected aesthetic cluster. The distinction mattered because most personalization systems only activate for logged-in returning users. Session-level personalization delivers value from the very first visit.

According to a 2025 McKinsey Digital report on retail personalization, AI-assisted discovery reduces search abandonment by up to 35% and increases add-to-cart rates by 22% when personalization operates at the session level. That was the exact use case Residence Supply needed solved.

How the Catalog Was Prepared for the AI Model

The implementation took six weeks. Softcircles integrated with Residence Supply's existing product feed without requiring a catalog rebuild. An automated attribute enrichment process analyzed existing product descriptions, imagery metadata, and category hierarchies to build a richer attribute map the model could train on.

That attribute depth was what made semantic search work. Products could surface based on mood descriptors, material preference, room type, and visual style, not just category labels. A shopper searching "warm ambient bedroom lighting" could be matched with a table lamp tagged for "low output," "warm tone," and "organic form," even if none of those words appeared in the product title.

The phased rollout followed three stages:

  • Weeks 1 to 2: Catalog attribute enrichment and semantic model training
  • Weeks 3 to 4: A/B testing on a subset of search traffic
  • Weeks 5 to 6: Full deployment with live personalization active

A/B results in weeks 3 and 4 showed search-to-product-page conversion increasing 28% for the test group. Full deployment was approved ahead of schedule.

The 90-Day Results

Three months after launch, the numbers were consistent and clear.

Search-driven revenue increased 41% year-over-year, compared to 12% growth in the same period the prior year. Site search went from contributing 18% of total revenue to 29%. Exit rates from search result pages dropped from 43% to 24%.

Average order value shifted as well. Personalized recommendations on product pages and cart summaries contributed to a 17% lift in AOV, as shoppers discovered complementary pieces they hadn't initially searched for. A customer choosing a ceiling fixture saw compatible sconces. A shopper comparing pendants saw coordinated table options alongside them. The cart became a curated set rather than a single item.

Multi-product transactions, sessions where shoppers purchased more than one item, increased 33% over the same 90-day window.

According to Gartner's 2026 Digital Commerce report, retailers deploying AI-assisted discovery tools report customer satisfaction scores 31% higher than those using standard site search, with the gap most pronounced in visually-led product categories. Residence Supply's results tracked closely with that benchmark.

Three Lessons for Home Decor Retailers

Treat Site Search as a Revenue Channel

Shoppers using site search convert at three to five times the rate of passive browsers, according to 2026 Forrester Research data. Most retail teams treat search as background infrastructure rather than a primary conversion surface. That framing consistently undervalues the investment needed to get it right.

The Best Personalization Goes Unnoticed

The goal isn't to show shoppers that results are personalized. It's to make sure results feel accurate. When the system works well, shoppers don't notice it. They just find what they're looking for faster. Visible or intrusive personalization adds friction rather than removing it.

Attribute Quality Determines What the Model Can Do

AI search is only as useful as the data it trains on. Residence Supply had rich product descriptions, but attributes weren't structured for machine interpretation. The enrichment layer translated that human-readable content into signals the model could use. Any retailer considering AI search should audit their attribute depth before anything else.

Frequently Asked Questions

How quickly can retailers expect results from AI search?

Most see measurable improvement within 30 days of full deployment. Personalization signals strengthen over 60 to 90 days as the system builds behavioral data. Initial A/B tests typically show directional improvement within two weeks, which is enough to validate the deployment decision before committing to a full rollout.

Does this type of solution require a large product catalog?

No. The benefit shows up across catalog sizes, but it's particularly strong for mid-size catalogs of 500 to 5,000 SKUs where keyword-based search tends to break down. Smaller catalogs gain the most from intent matching and reduced zero-result experiences.

What separates AI search from standard autocomplete?

Autocomplete predicts the query. AI search interprets the intent behind it. One helps shoppers type faster. The other helps them find the right product even when the query is vague, visual, or doesn't match any product label exactly. They're not the same problem, and solving one doesn't solve the other.

The Bigger Picture

Residence Supply's revenue lift didn't come from changing the catalog or rethinking the brand. It came from closing the gap between what shoppers searched for and what the site actually returned.

That gap exists in almost every home decor catalog. It widens as the product range grows and as shoppers bring increasingly specific, visual intent to queries that keyword systems were never built to handle.

AI search and personalization don't replace strong products or good merchandising. They make sure those products get found by the shoppers most likely to buy them. For any retailer whose data shows high search exit rates and low conversion, the tools to close that gap are available now. The cost of waiting is measurable, and it compounds every month it goes unaddressed.