Lily AI Closes the $300 Billion Search Gap

01 /The Concept

A $300 billion gap sits between what shoppers type into search bars and what online stores think they're looking for. Lily AI exists to close it.

The mismatch is structural. Shoppers describe what they want in natural, human language – "colorful patterned shirt," "mid-thigh jacket for cold fall weather," "loose-fit tank top for summer." Retailers catalog the same items using manufacturer terminology. These two vocabularies rarely overlap, and every failed search is a lost sale.

Lily AI's approach is to annotate every product in a retailer's catalog with additional tags – the words and phrases real shoppers actually use when they go looking. The platform analyzes product photos using computer vision, identifies what's in each image, and generates descriptive tags far richer than anything in the original product description. To source the right vocabulary, the system draws from search query logs, fashion publications, blogs, and social conversations – anywhere people naturally describe clothing in their own terms.

Pure automation isn't enough. A fully machine-generated taxonomy risks missing subtle distinctions or introducing noise. So Lily AI layers in a team of human experts who build training datasets, oversee model quality, and review outputs before updates go live. The training corpus has grown to over one million manually labeled products, with more than 20,000 attributes and three billion individual data points used to evaluate each item.

Critically, the platform doesn't require retailers to replace their existing software. Search, filters, recommendation engines, and SEO copy generation all continue working as before – Lily AI simply expands their vocabulary. For one customer processing 275,000 new product photos per week, the platform has increased the number of products actually sold by 15%, driven by a 2–3 percentage point lift in purchase conversion. Bloomingdale's, GAP, and J.Crew are among the brands already using it.

Lily AI recently raised a $12.4 million round, bringing total funding to $51.9 million.

02 /Why It Matters

Lily AI isn't alone in tackling this problem. Curated For You, which has raised $5.9 million, is working a similar angle with a stronger emphasis on lifestyle search formulations – "date night dress in Miami," "outfit for the first day at a new job," "this season's trending silhouette." Its engine incorporates real-time context like local weather and trend analysis from fashion media to interpret those queries intelligently.

The underlying thesis is the same: online retailers are leaving $300 billion a year on the table because their product search doesn't speak the buyer's language. That shared claim is what makes both companies credible to investors.

Kive, [covered here](/review/fig-chego-najdjosh), applies comparable AI tagging technology to creative asset management. Its platform lets users find photos described as "noir-style images of middle-aged men with a slight upward tilt to the head" – query language no traditional metadata system would handle. Kive has raised $8.8 million.

The "write in your buyer's language" principle extends well beyond retail. Fluint, [reviewed previously](/review/prezentacija-prodazham-ne-pomoshhnik), applies it to B2B sales: its AI analyzes transcripts of sales conversations and rewrites proposal decks using the specific words and phrases the buyer used during the call. With $1.6 million raised in its first round, the idea is early but the underlying insight – that vocabulary alignment drives conversion across every sales context – is the thread connecting all of these companies.

03 /Opportunities

The most direct path is building an alternative to Lily AI or Curated For You. The problem is real, measurable, and almost universally present across e-commerce. Retailers don't need to grow their traffic or expand their catalog to capture the upside – they just need their search to actually work. That's a compelling ROI story.

The market is large enough that regional specialists can succeed without displacing incumbents. Fashion vocabulary is culturally specific: what shoppers call things in one market differs from another, which means regionally trained models carry genuine competitive protection that a US-built platform can't easily replicate.

The broader opportunity is a vocabulary-translation layer that works beyond retail – landing pages for SaaS products, app store listings, service marketplaces. Most conversion optimization tools focus on layout and design rather than copy. The data suggests copy is where the bigger gains are. A platform that automatically rewrites product or service descriptions using actual search query language, calibrated to user behavior in a specific category, would address a gap that most existing tools don't touch.

The entry point with the clearest feedback loop: pick a single vertical with dense product catalogs and measurable search-to-purchase conversion data, build the taxonomy for that category, and use the performance delta as the proof case to expand.