# How Yeppon got a 150,000 SKU catalog marketplace-ready, from its own store outward > Yeppon, one of Italy’s leading electronics ecommerce players, runs 150,000+ SKUs across more than twelve European marketplaces. With CommerceClarity it processed around 80,000 SKUs end to end at roughly eight times its previous speed and activated around 24,000 products across channels, about 19,000 on Leroy Merlin and 5,000 on Amazon. Listing quality went from an average score near 42, below the auto-publish bar, to a target above 70. Every generated listing still passes human review. ![Inside a Yeppon warehouse.](https://cdn.sanity.io/images/xd4hrbt2/production/f1de727b7b2eba60c4e90ff637283725139561b7-800x600.jpg) **Hero figure value:** 8x **Hero figure label:** Faster to market than the manual process **Industry:** Consumer electronics ## Key figures - 80k · SKUs processed end to end - 8x · faster go to market than the manual process - 24k · products activated across marketplaces - 150k+ · SKU catalog in scope ## Agents - Data Enrichment - Taxonomy Mapping - Matching - Leroy Merlin Listing Optimizer - Amazon Listing Optimizer ## Overview Yeppon is one of Italy’s leading electronics ecommerce players: 150,000+ SKUs across electronics, appliances, DIY and home furniture, sold on its own store and on more than twelve European marketplaces. Every product is not one listing but a set of variants, different by channel and by language. The target was both surfaces at once: get the product data right at catalog scale on its own site, and get the catalog onto the marketplaces, starting with Leroy Merlin and Amazon. ## The challenge Before us the catalog was largely a manual job. A part-time resource added new SKUs, enrichment leaned on a third-party feed and on whichever suppliers sent good data, and filters like screen size, RAM and disk were pieced together by hand. Around seventy per cent of listings got created or updated, many with no photo, on an irregular cadence. For fifteen years the economics concentrated the real effort on the twenty per cent of the catalog around Apple, appliances and TVs. The rest went to the marketplaces matched on price and availability alone, never enriched. ### A partner already ahead on AI Yeppon is one of the most AI-forward teams we work with. They had already built their own ChatGPT agents to write titles, descriptions and bullet points. That instinct took them a long way, and then hit the limits of the tool: output that would not stay consistent run to run, gaps filled from the model’s training rather than from the product, images slow and expensive at catalog scale, and no proof a listing would clear a marketplace’s quality bar before it went up. A catalog is a harder problem than a front end. Tens of thousands of products, a quality bar that keeps moving, and results that have to stay provable run after run. That is a system to engineer, not a tool to run. ## How we solved it, agents and a Retail Engineer ### First, their own store Yeppon runs its store on Shopify, built by the team itself. The agents fill in each product, its details, specs and images, and file it under the right category, through an API integration into Yeppon’s own flows. It went live on the Apple range first, proved out on a batch, then rolled across the catalog, and today it runs on its own. ### Then, the marketplaces Every marketplace is a walled garden with its own rules. Three things made this hard. Leroy Merlin scores listings and only publishes above the bar, and Yeppon sat well under it, so close to half the catalog could not be listed cleanly. Nobody hands you the formula, not even the marketplace’s own team. And the numbers are big: around 37,000 brands to match to their codes, and 63,000 product details to fill. 1. **Match** Yeppon’s brands and barcodes to the marketplace’s own product codes. 1. **Create** listings for products not on the marketplace yet. 1. **Improve** existing listings on what actually lifts the score: filled-in details, images, specs, required fields. 1. **Check against the bar** before anything goes up. The same agents work on any category. They were proven on four to start, drills, planters, garden hoses and hose reels, then pointed at the rest. This phase is still running: around 24,000 products are live across channels so far, roughly 19,000 on Leroy Merlin and 5,000 on Amazon. ### The Retail Engineer The agents do not work alone. A CommerceClarity Retail Engineer works as an extension of Yeppon’s team, the same person scoping, building and shipping, inside the way Yeppon already operates. It also gave the catalog something it had gone without for fifteen years: a dedicated owner. Yeppon’s marketing team stays in command, and every generated listing passes human review before it goes online. ## Before and after The do-it-yourself approach against the system that replaced it. - Cost · Unpredictable, hundreds of thousands of tokens for one operation · A controlled pipeline with a predictable cost - Marketplace readiness · Average score around 42, half the catalog under the auto-publish bar · Target around 70, auto-publish on Leroy Merlin and Amazon - Scale and consistency · Unthinkable by hand across 150k+ SKUs · Any category, output scored against a fixed quality bar - Operating model · Could not realistically be shipped in house · Agents and an embedded engineer, people deciding what goes live ## A team that already knew where AI stops *In their words* ### Danilo Longo We had built our own agents on ChatGPT before this, so we knew where AI helps and where it stops. The agents handle the volume across the full catalog, but every listing still passes our team’s review before it goes live. We got the scale without giving up the final call on quality. General Manager, Yeppon ### Valerio Di Dio It connects straight into our flows through the API, so we shape content around what each product actually needs instead of running one template across the board. The same product can go out with a different angle on Amazon than on Leroy Merlin, and the speed holds even when we switch language. Reporting Analyst, Yeppon ## Proves area [Channels & Markets](/solutions/channels-markets) ## More customer stories - [STIGA](/customers/stiga) · How STIGA launched on seven marketplaces in four months - [Next Hardware & Software](/customers/next-hardware-software) · How enriched product pages lifted search and conversion for Next ## See what the agents do with your catalog Bring one messy feed, or the backlog of products you never got to. We will show you the rest. [Book a demo](/book-a-demo) ## Questions ### What did Yeppon get out of it? Around 80,000 SKUs processed end to end at roughly eight times the previous speed, the full catalog in shape on its own store, and around 24,000 products activated across marketplaces, about 19,000 on Leroy Merlin and 5,000 on Amazon. ### Yeppon had already built agents on ChatGPT. Why was that not enough? It worked until catalog scale. The output was hard to keep consistent product to product and run to run, the model filled gaps from its training rather than from the product, images were slow and expensive across 150,000+ SKUs, and nothing proved a listing would clear a marketplace’s quality bar before it went up. A catalog is a system to engineer, not a tool to run. ### What is a marketplace listing quality score? Marketplaces grade the listings you send them and publish automatically only above a threshold. On Leroy Merlin, Yeppon’s catalog averaged around 42 and close to half of it could not be listed cleanly. The target is above 70. Nobody publishes the formula, not even the marketplace’s own team, so getting there is testing and improving rather than a one-shot upload. ### How much matching does a catalog this size need? Around 37,000 brands to line up against Leroy Merlin’s own codes and around 63,000 product details to fill. The agents were proven on four categories first, drills, planters, garden hoses and hose reels, then pointed at the rest. ### Does anything publish without a person seeing it? Not at Yeppon. The agents bring the volume, and Yeppon’s marketing team reviews every generated listing before it goes online. That was their condition, and it is how the work runs. ### Is the work finished? No. The marketplace phase is still running, and by Yeppon’s own reckoning the programme is around two thirds of the way there. Next up are organic growth, new European markets, and using the same system for both organic and agentic demand. ## Entities - [Yeppon](https://www.yeppon.it) · Organization, about - [Channels & Markets](/solutions/channels-markets) · Thing, about - Danilo Longo · Person, mentions - Valerio Di Dio · Person, mentions - [Leroy Merlin](https://www.leroymerlin.it) · Organization, mentions - [Amazon](https://www.amazon.com) · Organization, mentions - [Shopify](https://www.shopify.com) · Organization, mentions - [ChatGPT](https://en.wikipedia.org/wiki/ChatGPT) · Product, mentions - [Cdiscount](https://en.wikipedia.org/wiki/Cdiscount) · Organization, mentions ## Sources - [Yeppon](https://www.yeppon.it) · Yeppon - [Listing quality and product data requirements](https://sellercentral.amazon.com/help/hub/reference/GT4CDCFTFPWKMYNW) · Amazon Seller Central - [Leroy Merlin](https://en.wikipedia.org/wiki/Leroy_Merlin) · Wikipedia - [Mirakl, the marketplace platform behind European retail marketplaces](https://www.mirakl.com) · Mirakl - [Shopify](https://www.shopify.com) · Shopify --- Canonical: https://commerceclarity.com/customers/yeppon Every page of this site is available as markdown: append `.md` to its path. Index: /llms.txt