Artificial intelligence is no longer a tool for marketplaces, according to Photoroom.
The Parisian startup sees AI becoming the connective tissue that powers how e-commerce companies operate, scale and grow. That’s because the AI-powered photo editor is assuming that future AI front-ends—the likes of Large Language Model (LLM) assistants and conversational commerce tools—will “route into the best-structured marketplaces.”
Co-founded and helmed by Matthieu “Matt” Rouif, the platform uses generative AI to aid sellers and brands—in particular for marketplaces, like Depop—in creating product images. Its aforementioned assumption is worth a cool half-billion following a Series B funding round that valued the firm at $500 million in February 2024.
Related Stories
“The future I’m most excited about is one where marketplaces feel frictionless,” said Olivia Moore, artificial intelligence partner at venture capital firm Andreessen Horowitz. “AI will quietly embed in every layer, making things simpler for everyone involved.”
An agentic marketplace is one where autonomous AI systems act with minimal human input—handling tasks such as creating listings, optimizing pricing, matching supply to specific user intent and completing transactions—in a collective, closed-loop process.
Auto-QA—short for Automating Image Quality Assurance—refers to the process of using generative AI to replace human-led reviews of product photos to ensure they meet a marketplace’s given standards. While anthropological bottlenecks have burdened these manual workflows in the past, genAI is now rejiggering such tedious steps into automated, end-to-end processes.
Why it matters: Roughly three-quarters of consumers surveyed (71-78 percent) believe that marketplaces—not individual sellers—are responsible for ensuring visual accuracy and listing trustworthiness. More than half (51 percent) would jump ship if a competitor’s imagery were more accurate—though the loyalty jumps (drops?) to two-thirds (66 percent) for those aged 18-34.
“Before AI, reviewing thousands of images a day across all our countries was a massive manual lift,” said Jeff Strauss, head of imaging at Photoroom and ex-director of imaging operations at Wolt. “Now auto-QA saves our team around a hundred hours a day—when it breaks, everyone feels it immediately.”
High-quality visuals were cited by 87 percent of shoppers as the most important factor in a purchase decision, with 63 percent reporting that inconsistent images reduce trust in a marketplace and 55 percent saying that poorly executed or heavily edited AI images decrease trust. In turn, automated QA can function as a reputational safeguard.
After Latin American marketplace Rappi teamed with Photoroom—which uses generative AI to clean up photos, often for secondhand commerce—the region’s first on-demand SuperApp projected a plus-20 percent uplift in buyer conversion, exclusively from AI-enhanced, QA-checked photos.
“We’re a food delivery company, not a photo-enhancing company,” said product director Nicoles Morales. “We can’t move at the same speed as AI models; that’s why we partner with companies like Photoroom.”
Moving on to the more fashion-focused, the Tokyo-based Mercari showed how this speed can compound when automation extends end-to-end, per the report. The secondhand selling marketplace’s AI-powered description-generation tool, now à la Photoroom, helps create hundreds of listings per minute; its adoption is allegedly spreading quickly among sellers.
“If you think about the whole seller experience, we’re looking at any step that users take that can be enhanced and made super-fast. Our leadership went from ‘back-to-startup’ to ‘AI-native’ as our mantra now,” said Nick Pittoni, product lead at Mercari. “AI-native marketplaces will emerge as new competitors. AI will make selling effortless, near-instant and accessible to users who currently find marketplaces too much work.”
For global marketplaces, Photoroom said GenAI unlocks the next growth frontier: cross-border liquidity. That involves centralizing cash through techniques (like sweeping) and mechanisms (like cross-border collateral), in line with a global marketplace’s ability to facilitate the fluid movement of supply and demand across different countries and languages. Effectively, it collapses the geographic and linguistic barriers that previously isolated listings.
For Mirakl, the solution provider’s AI-native tools reduced onboarding time by 91 percent, filling an average of 37 additional attributes per product in 5 seconds per SKU.
Translation? Cross-border liquidity means having the right amount of money in the right place at the right time—and in the right currency. The process used to take months, documenting manual transactions, catalog mapping and verification. According to the report, GenAI, specifically LLM-based translations, now allows marketplaces to reach new buyers almost instantly.
Vestiaire Collective provided “one of the clearest examples” in the report.
By switching to LLM-based translations for listing content, the global marketplace for pre-loved luxury fashion improved its 7-day sell-through by about 4 percent in early tests, according to Photoroom’s findings.
Not to mention that these “LLM-based translations actually beat” Vestiaire Collective’s copywriters, according to chief technology and product officer Stacia Carr. Granted, she also said that it’s punishing to be wrong. “The more precise you need the result, the more cautious you need to be—especially when you’re dealing with two physical objects, like a human body and a garment.”
The Paris-born marketplace uses AI for digital quality control—what Vestiaire said augments, rather than replaces, human authentication. In practice, this means historical data flags potential counterfeits, while humans review those flags to make the call; what Photoroom sees as a “human-in-the-loop” model forming the holy trinity of trust in AI marketplaces: speed, precision, and credibility.
“We still have depth of understanding of the assortment, the authentication, and everything around the legitimacy of the product,” Carr continued. “I don’t see that going away; it’s a matter of how we join forces with LLMs and leverage what each other does best.”
This shift from “assistance” to “autonomy” is fundamentally altering the unit economics of commerce, with one in four Americans claiming ChatGPT outperforms Google for product research.
“Users are not looking for a product,” according to Anne-Claire Baschet, Mirakl’s chief data and AI officer of agentic commerce. “They’re looking for something that solves their problem.”
In turn, a common concern arises: whether general-purpose will eventually replace specialized marketplaces. As Camila Bustamante explained in the report, they won’t, because marketplaces possess a “moat” that general models cannot replicate.
“Marketplaces will continue to lead because they own three things that LLMs cannot generate on their own: liquidity, vertical expertise, and historical transaction data,” said Bustamante, an investor at FJ Labs, a stage-agnostic venture fund focused on marketplaces and network-effect businesses.
While Vestiaire is actively exploring LLM-based commerce integrations, risks persist; for rare items and fast-changing inventories, AI summaries can undoubtedly distort details such as stock availability or pricing history.
For Carr, the question then becomes how to get there without gaming the system in a way that’s misleading to customers. “At the end of the day,” she said, “it’s advertising really; we see LLMs as another partner.”
