The $200B Problem with No Digital Plumbing

Fleek, a London-based startup, just closed a $25 million Series B to scale its AI-powered sorting system for secondhand clothing. The round, led by Burda Principal Investments (early backers of Vinted), brings total funding to $45 million. Existing investors include Andreessen Horowitz, HV Capital, and Y Combinator.

Every year, 24 billion secondhand items travel from donation bins in cities like London, Paris, and New York to sorting hubs worldwide. Yet the infrastructure remains stubbornly analog: garments assessed by hand, graded against inconsistent standards, traded through disconnected networks. Secondhand fashion grows three times faster than traditional apparel, but the plumbing can't keep up.

Fleek Sort: A Custom Vision-Language Model

At the core is Fleek Sort, a vision-language model trained on millions of secondhand transactions gathered over four years. It identifies, categorizes, grades, and merchandises garments from photos or video. The model learns from what actually sells, sharpening its accuracy over time.

Currently deployed in sorting hubs in Pakistan, India, and Dubai, with pilots in the UK, Europe, and US. The model turns a manual task into a digital one: a grader snaps a photo, Fleek Sort outputs a standardized grade (e.g., "Like New", "Minor Wear") and a suggested price range. This data flows directly into Fleek's marketplace.

Marketplace and Network Effects

Fleek connects over 2,000 verified wholesale suppliers and graders with 50,000+ retailers, resellers, and boutiques across 100+ countries. Every transaction generates more data, feeding back into Fleek Sort. The company claims it has kept 12 million items in circulation, saving 13 billion liters of water and avoiding 23,000 tonnes of CO₂.

Technical Details: How Fleek Sort Works

While exact architecture isn't public, the model is a custom vision-language model (VLM) fine-tuned on proprietary transaction data. It likely uses a transformer-based encoder for images and a language model for attributes like brand, fabric, and condition. The training data includes millions of examples of garments with final sale prices, enabling regression for pricing.

Key features:

  • Multi-label classification: Identifies brand, size, color, material, and condition simultaneously.
  • Pricing engine: Learns from historical sales to output a dynamic price range.
  • Search and matching: AI-powered search connects stock with buyers based on demand signals.

Why This Matters for Developers

Fleek's approach is a textbook example of applying AI to a fragmented, offline industry. The model's ability to learn from transaction data creates a data moat: each sale improves the model, making it harder for competitors to catch up. For developers building in B2B marketplaces, the lesson is clear: invest in proprietary data pipelines and domain-specific models.

The company frames the opportunity as "captured data." CTO Sanket Agarwal: "There's more data locked inside the global secondhand supply chain than almost any other market." Fleek Sort is the first AI trained specifically to read what used inventory is, what it is worth, and who wants it.

The Funding and Future

The $25M will fund larger engineering teams, expand the supplier and buyer network, and build a more AI-native marketplace. Burda's Julian von Eckartsberg: "We backed Vinted when secondhand fashion was still considered niche. Fleek is the infrastructure the next generation of fashion will run on."

What Developers Should Do Now

If you're working on marketplace or logistics AI, study Fleek's strategy: start with a specific, painful manual process (garment grading), build a custom model with proprietary data, and let the marketplace generate more training data. This flywheel is hard to replicate without the initial data advantage.

Also, note the trend: Vinted's US expansion and live-commerce plays signal that secondhand fashion is a growing sector hungry for tech. If you're looking for B2B AI opportunities, this is a prime candidate.

Direct Quotes

  • "Most people have no idea what happens to a piece of clothing after they part with it," said Abhi Arora, co-founder and CEO. "We started Fleek because that system is broken, the market it serves is exploding, and nobody is building the technology and infrastructure to fix it."
  • "There's more data locked inside the global secondhand supply chain than almost any other market," said Sanket Agarwal, CTO.

Metrics

  • Series B: $25M, total funding $45M
  • 2,000+ suppliers, 50,000+ buyers, 100+ countries
  • 12 million items kept in circulation
  • 13 billion liters of water saved, 23,000 tonnes CO₂ avoided
  • 24 billion secondhand items traded annually

Tags

secondhand-fashion, ai, marketplace, fleek, computer-vision, supply-chain, sustainability

This article was written based on the source article from The Next Web.