Skip to main content

Guide Catalog operations

Marketplace catalog management: a practical guide.

In short

Marketplace catalog management is the work of keeping one accurate product record while many sellers list the same items in different ways. It covers onboarding, categorization, attribute completion, variant and offer matching, duplicate control and policy review. This Youzu guide defines each step and where automation fits.

Published by Youzu · Editorial policy

Published
Reading time
9 min read

Key points

  • Keep one record per product, with each seller's price, stock and condition on a separate offer.
  • Map every listing to a leaf of your own category tree before it reaches search.
  • Match new listings to existing products, and send uncertain pairs to review instead of merging them.
  • Review policy findings with a reason code on every decision, shown to the seller.
  • Check queue sizes and reversals per queue, not only listing totals.
In this guide

What marketplace catalog management covers

On a marketplace, many sellers describe the same products in their own words. One seller writes 'jkt olive', another 'Jacket Olive M', and a third uploads 9 separate rows for one jacket in 3 colors and 3 sizes. Marketplace catalog management is the work of turning that stream into one accurate record per product, with each seller's terms attached to it.

The work splits into 5 jobs that never stop: bringing sellers and their data in, placing each listing in the right node of the product taxonomy (the hierarchy of categories, and the attributes attached to each, that organizes a catalog for browsing, filtering and reporting), completing the attributes that node needs, deciding which listings are the same product, and checking listings against policy. Each job has its own queue, its own owners and its own ways of failing.

The goal is not a perfect catalog on launch day. It is a catalog in which every product has one record, every record can be filtered and compared, and every decision can be explained to the seller it affects. Those 3 properties are what shoppers notice, even when they never see the catalog itself.

Roles and queues

Even a small marketplace team ends up with distinct roles. Naming them early makes it clear who decides what, and which queue a stuck listing belongs in.

Typical roles, what each one owns and the queue it works from
RoleOwnsQueue it works from
Seller successOnboarding, feed setup and seller questionsFeeds and files that fail validation
Catalog operationsProduct records, categories and attribute valuesListings without a leaf category or with missing required attributes
Matching reviewersDuplicate and offer decisionsPairs of listings that could not be shown to be the same product
Trust and safetyPolicy rules and moderation decisionsFlagged listings, appeals and reversals
Taxonomy ownerThe category tree and attribute definitionsRequests for new categories, attributes or allowed values

In a smaller team one person holds several roles. Keep the queues separate anyway: a listing waiting for a category decision and a listing waiting for a policy decision need different evidence, and mixing them hides which part of the catalog is falling behind.

Seller onboarding and intake

Clean data is cheapest at the point where it enters. Most marketplaces accept listings in 3 ways: a listing form for individual sellers, a file upload for sellers with large assortments, and an API or feed integration for the largest sellers. Every route needs the same minimum: a title, a category or enough information to infer one, at least one photo, and the attributes the category requires.

Publish a product data specification for sellers, the way sales channels publish theirs for merchants.[1] It should list required and optional attributes per category, allowed values, image rules and how identifiers are expected. A specification gives sellers something to check against before they submit, and gives your team a written basis for every rejection.

  • Validate files on upload and report errors by row and field, not as one failed file.
  • Ask for an identifier such as a GTIN (a globally unique product identifier of 8, 12, 13 or 14 digits issued under GS1 standards and printed as a barcode) where the product has one, and accept listings without one where it does not.[2]
  • Keep the seller's original values beside any correction, so every change can be explained.
  • Give sellers the same reason codes your reviewers use, so a rejected row tells them what to fix.

For individual sellers, the listing form is the intake. The seller listing flow guide covers how to collect the minimum, suggest the rest and validate before publishing.

Categorization and attributes

Every listing needs a leaf in your category tree. Sellers' own categories rarely match yours, so this is a taxonomy mapping (linking one category structure to another, such as a seller's categories to a marketplace taxonomy or an internal tree to a channel's) problem: some seller categories map cleanly, while catch-alls such as 'Home / Misc' need the product itself to decide. The product taxonomy mapping guide covers mapping patterns and how to measure them.

Once the category is known, its attributes are known. The listing should then carry each required attribute as a typed field rather than inside the title. Attribute extraction (pulling structured values such as color, material or size out of text or images and storing them as fields on a product record) from titles, descriptions and photos can propose values; fields the evidence does not support should stay empty and visible as gaps, so a person or the seller can fill them.

Variants, offers and duplicates

This is where a marketplace catalog differs most from a single retailer's catalog. 3 questions run on every new listing:

  1. Is it a version of a product we already have? Then it is a product variant (one purchasable version of a product that differs along defined dimensions, such as size, color or pack size) and belongs under that product, with its own color, size or pack values.
  2. Is it the same product another seller already lists? Then it becomes a marketplace offer (one seller's listing of a product with its own price, stock and condition, attached to a shared product record) on that product, with its own price, stock and condition.
  3. Is it already in the catalog under a different title? Then it is a duplicate, and the 2 records are merged after review.

All 3 are product matching (deciding whether 2 records describe the same product by comparing identifiers, titles, attributes and images) decisions. They compare identifiers where they exist, then model, key attributes and photos. In the worked example below, 6 sofa listings from three sellers resolve into 2 products with their offers, and one listing stays separate as a variant of the same family.

Worked example

6 listings, 2 products and a variant

6 sofa listings as received
ListingSellerTitleModelColorSizeCondition
Listing 1Seller AVelvet 3-seater sofa, navyNV velvetNavy3-seatNew
Listing 2Seller BNV velvet sofa, 3 seats, navyNV velvetNavy3-seatNew
Listing 3Seller CNavy velvet sofa NV, 3-seatNV velvetNavy3-seatUsed
Listing 4Seller AGrey 2-seat sofa, brass legsGS greyGrey2-seatNew
Listing 5Seller BGS sofa, grey, 2 seaterGS greyGrey2-seatNew
Listing 6Seller AVelvet 2-seater sofa, navyNV velvetNavy2-seatNew

How the listings resolve

  1. Listings 1, 2 and 3: One product: velvet 3-seater sofa, navy, with 3 offers.

    • Same model
    • Same color
    • Same size
    • Same photo
    • Different sellers: all offers kept
  2. Listings 4 and 5: One product: grey 2-seat sofa, with 2 offers.

    • Same model
    • Same color
    • Same size
    • Different sellers: all offers kept
  3. Listing 6: A different size is a variant of the same family. It is not a duplicate.

    • Same model
    • Same color
    • Different size: kept separate
When 2 listings cannot be shown to be the same product they stay separate, and uncertain cases can go to review.

2 rules keep these decisions safe. When 2 listings cannot be shown to be the same product, they stay separate until a person decides. And a match never moves one seller's terms onto the shared record: condition, price and delivery stay on the offer. The variants, families and offers guide covers the data model behind both rules.

Policy review

Every marketplace has rules about what may be sold and how it may be shown: prohibited items, photos that do not show the product, contact details in a description, titles that name a different product than the photo. Write each rule so it can be checked against a listing, give every decision a reason code and send uncertain cases to people rather than guessing.

Test a new rule on a reference set of listings before it applies to live traffic, so you see which listings it would flag before sellers do. The classifieds listing moderation guide shows how to move from written policy to rules, reason codes, queues and appeals.

What to track

Measurement keeps a catalog honest, but targets copied from another business mislead. Track these measures on your own catalog, set your own baselines and compare each one with itself over time:

  • Listings mapped to a leaf category, and the size and age of the unmapped queue.
  • Required-attribute coverage per category, counting empty fields as well as filled ones.
  • Matching decisions sent to review, and how often reviewers reversed a proposed match.
  • Duplicate reports from shoppers and sellers, per category.
  • Policy flags per rule, with reversals and appeals per rule.
  • Time from submission to publication, per intake route.

Report reversals next to volumes. A queue that shrinks because reviewers accept everything is not healthier than one that grows because they check carefully.

Common failure modes

  • Category drift. Sellers learn that a catch-all category avoids attribute requirements. Retire catch-alls, or require a leaf at submit.
  • Facts in titles. Color, size and material sit in the title, so filters never see the product. Extract them into fields and ask sellers for them separately.
  • Over-eager merging. 2 similar products are merged and a shopper receives the wrong size. Keep listings apart unless they can be shown to be the same product.
  • Offers written onto products. One seller's used condition or delivery terms appear on the shared record for everyone. Keep seller terms on the offer.
  • Rules without reasons. Sellers see a rejection but not why, and resubmit the same listing. Show the reason code with every decision.
  • One queue for everything. Category, matching and policy work wait in one list, so nobody can see which part is behind. Give each its own queue and owner.

Where automation fits

Automation helps most where a decision is repetitive and the evidence is in the data: proposing categories, extracting attributes, finding candidate matches and flagging likely policy issues. People stay responsible for the rules, for uncertain cases and for anything a seller appeals.

A practical split is to let software propose and route, and to let people decide what the software could not show. Whatever tool you consider, test it on a sample of your own listings, including the messy ones, and judge the output against labels your team trusts.

Catalog intelligence is the name for that proposing layer. Youzu helps marketplaces, classifieds and retailers turn raw seller and supplier product data into structured, reviewable catalog records, and works alongside your existing PIM, ERP, search and storefront.

Boundaries

What Youzu does not do

  • Youzu does not replace your ERP, search or storefront; those systems stay authoritative for their own data. It can replace your PIM, or work alongside it.
  • Youzu publishes no accuracy figures. Measure accuracy on your own sample against your own labels, with critical errors reported separately.
  • Youzu does not promise an onboarding time in advance. We size the timeline with you after seeing a sample.
How an evaluation runs

Checklist

Marketplace catalog management checklist

12 checks across intake, categories, attributes, matching, offers, policy and measurement, with an owner column and an empty status column for your team.

Download the checklist (CSV)

A CSV file of 12 rows, named youzu-marketplace-catalog-management-checklist.csv. No form to fill in first.

Marketplace catalog management checklist, scrollable
AreaCheckOwnerStatus
IntakeA product data specification is published for sellersSeller success
IntakeFile uploads report errors by row and fieldSeller success
IntakeIdentifiers are requested where products have themCatalog operations
CategoriesEvery listing is placed in a leaf categoryCatalog operations
CategoriesCatch-all categories are retired or require reviewTaxonomy owner
AttributesRequired attributes are defined per leaf categoryTaxonomy owner
AttributesUnsupported fields stay empty and are flaggedCatalog operations
MatchingUncertain matches go to review instead of mergingMatching reviewers
OffersPrice, stock, condition and delivery live on the offerCatalog operations
PolicyEvery rule has a reason code shown to sellersTrust and safety
PolicyNew rules are tested on a reference set before activationTrust and safety
MeasurementQueue sizes and reversals are reviewed per queueCatalog operations

FAQ

Questions buyers ask

Short answers to the questions teams ask about this topic.

A retailer catalog describes products the retailer buys and sells itself, so one team controls the data. A marketplace catalog combines listings from many sellers who describe the same products differently. That adds matching, offer management and seller-facing policy decisions on top of the usual category and attribute work.

Give one team, usually catalog operations, ownership of the product record, the category tree and the attribute definitions. Seller success, matching reviewers and trust and safety each own their own queue. What matters most is that every queue has a named owner and every rule has a person who can change it.

Usually yes, but only after matching. A seller's listing should first be checked against existing products; if it matches, it becomes an offer on that product. Only when no match can be shown should it create a new product, and new products in sensitive categories can wait for review before they go live.

The repetitive, evidence-based parts: proposing categories, extracting attributes, finding candidate matches and flagging likely policy issues. Rules, uncertain cases and appeals stay with people. Measure accuracy on your own sample against your own labels, with critical errors reported separately, before deciding how much of each job to automate.

Youzu helps marketplaces, classifieds and retailers turn raw seller and supplier product data into structured, reviewable catalog records. Products, variants and seller offers are separate records: the same product sold by several sellers is one product with several offers. Evaluations start with a sample of your own catalog and expand to more workflows once the results are reviewed.

Last reviewed

Sources

  1. Product data specification, Google Merchant Center Help. support.google.com/merchants/answer/7052112. Accessed . Back

  2. Global Trade Item Number (GTIN), GS1. gs1.org/standards/id-keys/gtin. Accessed . Back

About this guide

Published by Youzu · Editorial policy

Published

Every guide defines its terms, works through one example and cites the public sources it relies on, with the date each source was checked.

Guides explain methods. They do not rank Youzu above other approaches, and an example is never presented as a customer result.

More in all guides and the catalog glossary.

Next step

Automate Commerce. Make shopping personal.

Bring your catalog, listing flow or shopping idea. Let's find the right place to start.

Book a demo

Tell us about your catalog and we will get back to you.