Ecommerce · Deep dive

Fleek

A London-run, Karachi-and-Delhi-powered B2B marketplace that puts the world's secondhand-clothing bale trade online — 2,000+ verified wholesalers selling graded vintage to 50,000+ retailers in 100+ countries — now betting a $25M Series B that its Fleek Sort vision model becomes the grading and pricing standard for the 24 billion used garments sorted by hand every year.

emerging

The question that decides it: Fleek exists because COVID-era online brokers charged vintage retailers ruinous commissions — and every completed shipment teaches a Karachi wholesaler and a London buyer exactly who each other are, inviting them to trade direct on WhatsApp next time. Does Fleek Sort — the vision-language model trained on four years of Fleek's own transaction data that grades and prices a garment from one smartphone photo — get adopted as the pricing standard inside sorting hubs moving 600,000 lbs a day, including facilities that never transact on Fleek's marketplace, before counterparty familiarity and cheap off-the-shelf vision models let both sides route around the commission?

My take ★★ Very interesting ★ Interesting − Maybe ✗ Not interesting

Snapshot

Fleek is a London-based B2B marketplace that moved the world’s used-clothing bale trade online: roughly 2,000 verified wholesale suppliers and graders — concentrated in Karachi, Delhi and Dubai, where much of the planet’s donated clothing is actually sorted — selling graded vintage and secondhand stock to more than 50,000 retailers, resellers and boutiques in over 100 countries (company figures, July 2026). In July 2026 it raised a $25M Series B led by Burda Principal Investments, Vinted’s early backer, with eBay joining the cap table, bringing total funding to $45M. The pitch has shifted from marketplace to infrastructure: Fleek Sort, a vision-language model trained on four years of the company’s own transactions, grades and prices a garment from a single smartphone photo. It matters because supply — not demand — is the binding constraint on a $200B+ resale market, and whoever standardizes grading controls the choke point.

Founding story

Abhi Arora, fresh from a Cambridge MBA and living near Brick Lane, watched his favorite vintage shop announce its closure in 2021. The owner explained the mechanics: she used to fly to Asia and the Middle East to pick stock directly from wholesalers; COVID killed the trips; online brokers filled the gap, charging heavy commissions; one scammed her out of more than $15,000, and with no recourse and shrinking margins she shut the store (Fortune, July 2026). Separately, Sanket Agarwal — a software engineer with about a decade in Silicon Valley including Google — was hearing the mirror-image complaint from his mother-in-law, who sold clothes on Poshmark and could not source inventory through pandemic supply chains (TechCrunch, November 2024). The two had met in an online founders’ community, put the pieces together, and founded Fleek in late 2021, joining Y Combinator’s W22 batch. The unfair advantage was biographical: both grew up in India, and the world’s used-clothing sorting industry lives on the subcontinent. They speak the languages, visit the warehouses in person, and hired ops staff in Karachi as early as 2022 — a founder-market fit most Silicon Valley teams could not replicate.

How it works

The physical chain Fleek sits on is invisible to most consumers. A garment donated in London or New York is baled, exported, and lands in a sorting warehouse in Karachi, Delhi or Dubai, where workers hand-sort staggering volumes — a single large wholesaler can take in, sort, mend, clean and ship up to 400,000 kg of clothing a day (TechCrunch, November 2024); Fleek’s largest partner moves some 600,000 lbs daily (Fortune, July 2026). Output is graded and packed three ways: bales sold by weight, bundles sorted by brand, style, size or material, and handpicks — individually selected premium pieces. Pre-COVID, Western retailers flew in and picked racks in person; after, a scrappy layer of Instagram brokers emerged, negotiating over video calls with no accountability.

Fleek formalizes exactly that behavior. Wholesalers list inventory with photos, video and stated grades; buyers browse by brand (Carhartt, Ralph Lauren, Stone Island, Nike), category or weight, join live video handpick sessions on the app where a grader walks the racks on camera, and haggle through a “Make an Offer” feature. Fleek stands behind the transaction: payment runs through the platform, a Buyer Protection policy covers misdescribed goods, and quality-control hubs in Pakistan and India inspect shipments for authenticity, grade and cleanliness, with dedicated counterfeit-check centers in Karachi and Delhi for frequently faked brands. Fleek then handles shipping and customs end-to-end, with typical delivery of 10–14 working days (company site, accessed July 2026). The newest layer is Fleek Sort: graders photograph a garment with an ordinary smartphone and the model identifies brand, style and category, flags rips and faulty seams, and predicts both clearing price and time-to-sell. It is live with graders in Pakistan, India and Dubai, with pilots starting in the UK, Europe and the US (Fortune, July 2026).

Product and business overview

Five named components. The marketplace is the core: bales, bundles and handpicks from verified suppliers across 10 sourcing countries. Live video handpicks replicate the in-person rack-picking ritual remotely and justify premium pricing. Fleeky, an AI matching assistant, turns a buyer’s wishlist into automated supplier introductions. The logistics and trust stack — QC inspection, buyer protection, customs clearance, shipping, and a BNPL option letting buyers defer payment up to 30 days after receipt — is what separates Fleek from the Instagram brokers it replaced. And Fleek Sort is the strategic bet: built in-house because running third-party models at bale-trade volumes would be prohibitively expensive, it is being extended from photos to video, with conveyor-belt integration studied and robotic sorting explicitly described by Agarwal as still far off (Fortune, July 2026). If Sort gets adopted by facilities that never transact on the marketplace, Fleek becomes an infrastructure company; today it is a feature of the marketplace.

Business model and pricing

Fleek books revenue as a commission on payments buyers make to wholesalers, and — unusually — has never published the rate. Agarwal has said only that the cut varies with the volume and quality of goods sold (TechCrunch, November 2024). Buyers pay no subscription; suppliers price their own stock, negotiable via offers. Shipping, customs and buyer protection are bundled into the transaction; a BNPL program (up to 30 days post-receipt) adds a financing margin. Latka estimates $15.3M revenue in 2024 on a ~105-person team. The unpublished take rate is not a trivial omission: the company’s own founding story villainizes brokers who charged high commissions, which means Fleek’s rake must stay below the pain threshold that pushed the Brick Lane shop under — while still covering QC inspectors, refunds and international freight coordination. Fleek Sort pilots hint at a second revenue line — grading software licensed to sorting facilities — but no pricing has been disclosed.

Traction over time

The shape is real: buyers up 5x and suppliers 2x in twenty months, with deliberate asymmetry — in a supply-constrained market, more demand per supplier raises supplier lock-in. But the disclosure gap is equally real: no GMV, no take rate, no repeat-purchase rate has ever been published, and the only revenue number is a third-party estimate. For a five-year-old marketplace raising a Series B, that opacity is itself a data point.

Market analysis

BCG pegs the secondhand apparel market at over $200B today (2025). ThredUp’s GlobalData-conducted 2025 Resale Report (March 2025) projects $367B globally by 2029 at a 10% CAGR — 2.7x faster growth than apparel overall — with the US market alone reaching $74B by 2029 and online resale growing 23% in 2024. Beneath the consumer numbers sits Fleek’s actual market: up to 24 billion garments a year moving through the sorting supply chain (MDPI figure cited by the company, 2026), almost entirely hand-graded. Three structural forces run in Fleek’s favor. First, supply scarcity: Vinted scrapped seller fees to attract inventory and Depop and eBay followed — demand platforms are literally paying for what Fleek aggregates (Fortune, July 2026). Second, regulation: EU rules requiring separate textile collection took effect in 2025, and rules barring retailers from destroying unsold stock are coming into force in 2026, both pushing volume into exactly the channels Fleek serves. Third, tariffs and Gen Z preference keep pulling Western retail toward resale. The countervailing force: the trade’s economics are thin, informal and relationship-driven, and most of its participants have never paid for software.

Competitive intel

Fleek claims no direct competitor in wholesale secondhand grading (Tech Funding News, July 2026), and narrowly that is true — but the flanks are crowded. Vinted (€8B, April 2026) and eBay own the downstream demand and could verticalize; eBay hedged by investing. Whatnot ($11.5B, October 2025) owns the livestream resale format Fleek’s own buyers sell into. ThredUp is the vertically integrated counter-model — own warehouses, own grading, Resale-as-a-Service for brands — which wins if infrastructure needs to be operated rather than marketplaced. And the deepest threat is the incumbent layer itself: the Karachi, Panipat and Dubai sorting houses and Western brokers like Bank & Vogue that constitute Fleek’s supply base. They have the goods, the labor and the relationships; Fleek has the buyer list and the data. Full breakdown in the competitive set above.

History and evolution

• Late 2021 — Founded in London by Arora and Agarwal after the Brick Lane shop closure and the Poshmark sourcing complaint.
• Winter 2022 — Y Combinator W22; a16z leads a $5.6M seed (kept quiet until 2024). First ops hires in Karachi to build the supply side.
• 2022–2024 — Grinds out the unglamorous marketplace layer: supplier verification, QC hubs, live video handpicks, buyer protection, customs handling. 2.5M items moved by late 2024.
• Nov 2024 — $14.8M Series A led by HV Capital, with Finkelstein, Raga and Plaice as angels; first press cycle. Counterfeiting openly admitted as an industry-wide problem.
• Mid-2025 — Begins building Fleek Sort in-house, roughly “a year ago” from July 2026, after concluding third-party models were too expensive at bale volumes.
• Jul 2026 — $25M Series B led by Burda with eBay, FJ Labs and H14; narrative repositions from “vintage wholesale marketplace” to “AI infrastructure for the circular economy.” Sort pilots begin in the UK, Europe and US.

What people say

The case for. Trustpilot reviewers (several hundred reviews, accessed July 2026) skew positive with consistent themes: bales arrive quickly and match their descriptions, the app is easy to use, and — tellingly — when first orders went wrong, several buyers report full refunds and responsive service, which is precisely the trust gap Fleek was built to close. Resellers credit the platform for making sourcing viable without travel or minimum orders beyond reach of a side-hustle budget. The angel list (Shopify’s president, Depop’s ex-CEO) and eBay’s strategic check function as professional character references for the model.

The complaints. The negative reviews cluster around the seams between Fleek’s promises and its suppliers’ behavior. One buyer reported that 80% of their orders ended in partial refunds or cancellations, with argumentative suppliers — one allegedly threatening to fly from Pakistan to the UK over a dispute. Multiple reviewers say the vaunted QC misses fakes, including counterfeits with visibly misspelled brand names, and that buyers must pay out-of-pocket for third-party authentication to claim reimbursement. Missing items and slow support responses recur. The company’s own admission that counterfeiting is a huge industry problem (TechCrunch, November 2024) confirms the reviews are signal, not noise. Employee-review data is genuinely thin: Fleek shares its name with several unrelated companies on Glassdoor, and no reliable, attributable review base exists for the joinfleek entity — a gap worth noting rather than papering over. The recurring reviewer advice — “vet the individual seller” — is itself a criticism: it means the platform’s verification does not yet fully substitute for buyer diligence.

Outlook: the open question

For the bull case to hold, Fleek Sort must become the industry’s grading and pricing oracle — adopted inside sorting hubs as a work tool, including facilities that never sell on Fleek’s marketplace — before the marketplace’s own success teaches its best counterparties to trade without it. The ingredients are genuinely scarce: four years of proprietary transaction data linking photos to actual clearing prices and sell-through times, founder access to subcontinent wholesalers no Western competitor can match, regulatory tailwinds force-feeding volume into the channel, and a cap table (Burda, eBay) that doubles as distribution. If graders in Karachi and pilot facilities in Europe come to price inventory in Fleek’s grades the way diamonds are priced in GIA’s, the commission stops being a broker fee and becomes a standards royalty — and the $200B trade gets its Bloomberg terminal.

The bear case is that Fleek is a broker that has not yet discovered it. B2B marketplaces die by disintermediation: every completed order reveals the counterparty, and wholesale relationships are repeat by nature. Fleek’s defenses — buyer protection, customs handling, BNPL — are real but replicable, and its refusal to disclose GMV or take rate makes it impossible to verify that transactions are compounding rather than graduating off-platform. Meanwhile the cost of capable vision models is collapsing; if grading-by-photo becomes a commodity capability by 2028, a 600,000-lbs-a-day sorting house can run its own model and sell direct to the buyer list Fleek introduced. Watch three tells: whether Sort signs facilities with no marketplace relationship (infrastructure) or only existing suppliers (feature); whether Fleek ever publishes GMV and repeat rates; and whether counterfeit complaints fall as Sort scales — because an AI grading standard that still ships misspelled fakes is not a standard.

How a challenger would attack it

Arm the sorting houses before Fleek does. Fleek’s whole moat is the buyer list plus four years of transaction data — but the physical assets, goods, labor, and grading expertise all belong to its suppliers, and vision-model costs are collapsing. The direct attack is a supplier-side toolkit: sell the Karachi, Panipat, and Dubai hubs their own commodity-model grading, storefronts, and payment rails, positioning as “keep 100% of your margin, own your buyer relationships” against a marketplace whose founding myth villainizes broker commissions while refusing to publish its own take rate. That undisclosed, volume-dependent rake is a soft target: every supplier already knows what they’re paying, and every completed shipment already introduces the counterparties to each other. The second vector is trust: Trustpilot complaints of counterfeits with misspelled brand names slipping through QC, one buyer reporting 80% of orders ending in partial refunds, and buyers paying out-of-pocket for third-party authentication mean the challenger leads with hard authentication guarantees — free third-party verification, penalty-backed grading accuracy — and turns Fleek’s “vet the individual seller” reviewer advice into its pitch deck. Third, cherry-pick the premium layer: handpicks and brand-sorted bundles (Carhartt, Stone Island) carry the margin; a curated, authenticated-only marketplace for the top grade leaves Fleek the by-weight bale commodity where thin economics can’t fund QC hubs.

Same playbook, new buyer

Grade different goods, or sell to the people who make the clothes. Fleek’s transferable machinery — verified cross-border wholesale of hand-graded used goods, with live video inspection, buyer protection, QC hubs, and BNPL — maps directly onto adjacent secondhand categories that flow through the same ports and warehouses: used shoes and handbags (higher counterfeiting stakes, higher margins), refurbished electronics out of Dubai, furniture and homeware. None inherits Fleek’s clothing-trained model, so the data moat resets to zero for everyone. The more defensible shift is the buyer: brands and retailers, not resellers. EU rules barring destruction of unsold stock from 2026 force every European retailer to find resale channels — a compliance-driven, contract-revenue version of Fleek’s marketplace (ThredUp’s Resale-as-a-Service points the way, but from expensive owned warehouses rather than an asset-light network of subcontinent sorting capacity). Fleek can’t easily pivot there: its supply base, QC infrastructure, and Fleek Sort training data are all tuned to donated-garment bales, and its 50,000 small-reseller buyers are the opposite motion from enterprise compliance sales. A third shift is geographic: intra-Africa and Latin American bale trade runs on the same broker informality Fleek fixed for Western buyers, with no incumbent platform at all.

Sources and further reading

• Fleek, a marketplace for wholesale secondhand clothes, sews up $20.4M (TechCrunch, Nov 2024)
• Exclusive: Fleek, an online marketplace connecting vintage clothing wholesalers and retailers, raises $25 million (Fortune, Jul 2026)
• Fleek raises $25M from eBay and Vinted’s early backer to rebuild secondhand fashion’s supply chain (Tech Funding News, Jul 2026)
• Fleek raises $25M in Series B funding (FinSMEs, Jul 2026)
• Secondhand wholesale platform Fleek secures $20M in seed, Series A rounds (Sourcing Journal, Nov 2024)
• ThredUp 2025 Resale Report (ThredUp / GlobalData, Mar 2025)
• Fleek: a B2B marketplace for wholesale second hand fashion (Y Combinator, accessed Jul 2026)
• How Fleek hit $15.3M revenue with a 105 person team in 2024 (Latka, accessed Jul 2026)
• Fleek reviews on Trustpilot (Trustpilot, accessed Jul 2026)
• How it works (joinfleek.com, accessed Jul 2026)