Use case: raising catalog completeness from 30% to 90% in one month

By Manuel Mattar4 min read

In short

This is an example scenario, and the 30% to 90% goal belongs to the scenario, not a promise. A team measures what each product is missing, fills in brand, category and attributes, generates descriptions in batches, reviews quality and measures again every week with the same reports.

This case is an example, not a customer: an imaginary catalog, with scenario numbers, to show how completeness is measured and raised with Bruxel. The goal of 30% to 90% in one month is the example's, not a promise: what you reach depends on your data and your team.

The situation

A tools brand has 2,000 products and sells in Spanish and English. It imported its catalog from several spreadsheets, and the metrics screen says its average completeness is 30%. Management asks for 90% in one month, before opening a new channel.

The first thing is to understand what that number means. In Bruxel each product adds up to 100 points:

PartPointsHow it is filled
Brand and category25Half for each one
Attributes25Required attributes of its family with a value
Content35Languages with a published description
Main image15An image ready on the product page

A 30% average can be the whole catalog half done or one part perfect and another empty. That is what the three buckets are for: complete, partial and not started.

Week 1: measure before moving

The catalog lead opens Account → Analytics, checks the catalog health and uses "Download incomplete (CSV)": one row per product that does not reach 100, from least to most complete, with what it is missing in fixed words such as marca, categoria, atributos:3/10, ficha:en or imagen-principal. In the spreadsheet, a filter on each word tells her how many products have each gap.

In the scenario, the diagnosis is clear: almost no product has a published description, many lack a family and a few have no photo. Descriptions are worth 35 points: that is where most of the way is.

She also saves in Reports a favorite from the Catalog and quality family, to download the same cut every Friday, with the same filters.

Week 2: structure and attributes

Without a family, Bruxel does not know which attributes a product is missing. In Catalog → Families the team builds one family per tool type, marks which attributes are required and, in each category, sets a "Default family"; with "Apply", the products already there receive it.

To fill attributes, instead of going one by one, they select products and use fill with AI: it completes them from what is already known about them and only charges for what it manages to fill. Value rules, such as converting inches to millimeters, are set first, so whatever comes in is normalized.

Week 3: generate and review

In Catalog → Products, the team filters by family and uses the "Generate the descriptions of the N products in this filter" panel. "See how many and how much" gives the price without charging: in a batch, 7 credits per product and per language, so in the scenario each product costs 14. Batches go out 500 at a time and descriptions arrive within a few hours; they are tracked in Content → Batches. Only what succeeds is charged.

Every description goes through free checks when generated —empty fields, filler text, another language, broken HTML— and anything serious ends up "Rejected for quality", without charge. Then, in Content → Detected issues, the team runs "Check quality", which looks for descriptions that talk about another product type, for 2 credits each. Each finding is resolved with "Fix category", "Regenerate" or "Dismiss".

Week 4: close gaps and measure again

What is left are the products without a main image. Those 15 points are not generated by AI out of nothing: the team uploads the missing photo or, where there is already a real photo, uses the Studio for its studio version.

At the end of the month, the catalog lead goes back to Account → Analytics, uses "Recalculate now" if there are products with stale data and downloads the PDF report for management: the summary, what is missing, coverage and completeness by brand, by category and by family. In the scenario, the catalog closes at 90%; the products that do not get there are mostly the ones still without a photo.

What gets solved and what does not

Solved: the team knows what each product is missing, works first on what scores most and always measures with the same cut.

Not everything is solved:

  • Completeness does not measure text quality. That is what the quality check and human review are for.
  • Each channel's requirements are another measure. A complete product may not have every column a marketplace asks for; "Measure" in Channels tells you.
  • Photos depend on you: without a real image, that 15% does not come.

The metrics and quality report, attributes and families and generate descriptions guides have the details. The reports, AI content and review and quality pages show the rest.

Frequently asked questions

How does Bruxel calculate completeness?

Each product adds up to 100 points: 25 for brand and category, 25 for attributes, 35 for descriptions published in the languages you configured and 15 for having a main image. The catalog's figure is the average.

Does Bruxel guarantee reaching 90%?

No. The 90% is this example's goal. What you reach depends on your data: for instance, the main image has to exist and generation needs something to start from.

Where do I see what each product is missing?

In the incomplete list, a CSV with one row per product that does not reach 100, from least to most complete, with what it is missing: brand, category, attributes, languages without a description or main image.

Sources

  1. Data quality (Wikipedia)
  2. Product information management (Wikipedia)

About the author

Manuel Mattar

Founder of Bruxel.AI

In 2014 he founded Wolf Sellers, an e-commerce agency in Mexico City, and later Bruxel.AI, the AI-powered PIM.

See all posts by Manuel Mattar

Try it with your catalog

Create your free account and try Bruxel with your own products using the welcome credits.

Keep reading

Use case: raising catalog completeness from 30% to 90% in one month