Case Studies

Improving e-commerce product data with AI

Turning inconsistent catalog data into faster, cleaner and more scalable product operations

Sector

Retail and e-commerce

Client

Major Benelux e-commerce brand

Challenge

Incomplete and inconsistent product information slowed operations, weakened search visibility and created customer friction

Solution

AI-driven product enrichment connected directly to the retailer's existing product systems

/

/

Improving e-commerce product data with AI

/

/

Improving e-commerce product data with AI

01

The challenge

Product data was becoming a growth problem.

For a large e-commerce business, product information is infrastructure.

Customers rely on it to:

  • Find the right product

  • Compare options

  • Understand specifications

  • Use filters

  • Trust what they are buying

Search engines rely on it too.

The retailer was dealing with recurring quality problems across its catalog.

Product titles varied in tone and structure.

Descriptions were sometimes too short, inaccurate or missing important search terms.

Images were limited or ordered inconsistently.

Filtering attributes were incomplete across similar products.

Product updates depended heavily on suppliers and internal PIM processes.

The result was friction across search, conversion, internal operations and product onboarding.

02

What needed to change

The retailer did not need another team manually rewriting product pages.

It needed a repeatable way to improve product data at scale.

The system had to:

  • Understand the retailer's taxonomy

  • Apply consistent rules

  • Work across a large catalog

  • Connect to existing product systems

  • Generate useful content automatically

  • Keep people involved in approval

  • Support future enrichment beyond text

The goal was better product information without growing the manual workload at the same rate.

03

What we built

An AI enrichment layer connected to the existing commerce stack.

WeAreBrain built the first working prototype around an AI-driven catalog enrichment platform.

The solution connects the retailer's product systems to an AI layer capable of improving and extending product information.

The first implementation covered:

  • automated product classification

  • attribute cleaning

  • attribute enrichment

  • product title generation

  • product description generation

  • translation

  • image improvement workflows

  • identification of product variants

  • related-product discovery

The platform was designed to adapt to the retailer's existing taxonomy and business rules rather than forcing the catalog into a new structure.

04

From raw catalog data to approved product content

The workflow

01 Retrieve

Existing product information enters the enrichment workflow through API integration.

02 Understand

The system interprets product type, taxonomy, existing attributes and available source data.

03 Enrich

AI generates or improves the relevant product information.

Depending on the item, this can include:

  • Title

  • Description

  • Attributes

  • Classification

  • Translation

  • Product relationships

04 Review

The enriched information appears inside an internal management interface.

A person can review and approve the result.

05 Publish

Approved data moves back into the retailer's existing product workflow.

The enrichment layer therefore improves the existing process instead of replacing the commerce platform around it.

05

Better product titles

Consistency without writing every title by hand.

Large catalogs often grow through different suppliers, categories and teams.

That creates inconsistency.

One category may follow a clear naming structure while another does not.

The enrichment engine can apply shared rules across the catalog and generate titles that are more consistent, useful and search-friendly.

This reduces the time employees spend rewriting basic catalog content.

06

Better descriptions

Turn available product data into useful customer information.

Product descriptions often contain one of two problems:

Too little information.

Or large amounts of inconsistent supplier copy.

The AI workflow can use available product information and defined content rules to prepare a more consistent description.

People still control what gets approved.

The repetitive first-draft work moves into the system.

07

Better attributes

Filters are only as useful as the data behind them.

Missing attributes affect more than product administration.

They affect:

  • Category navigation

  • Filters

  • Search

  • Comparison

  • Recommendations

  • Product feeds

The enrichment layer can identify gaps and help populate missing structured information.

That makes better product data reusable across the wider commerce experience.

08

Better product relationships

AI can also help identify relationships inside the catalog.

For example:

  • Variants

  • Similar items

  • Alternative products

  • Related products

This creates opportunities beyond content production.

The same enriched catalog can later support:

  • Recommendations

  • Cross-selling

  • Search relevance

  • Conversational product discovery

  • Personalised experiences

09

Human approval stays in the workflow

Automation should speed up catalog teams, not remove control.

Product data is customer-facing.

Incorrect information can damage trust and create operational problems.

The prototype therefore includes an approval layer connecting AI-generated content to the retailer's internal product management process.

The AI prepares the enrichment.

The product team decides what gets published.

10

The result

From 20 minutes to 2 minutes per item.

The prototype delivered measurable improvements in product-content operations.

~90% less manual effort

The retailer reported around a 90% reduction in manual work for creating product titles and descriptions.

20 minutes → 2 minutes

Average time spent enriching an individual product fell from around 20 minutes to 2 minutes.

~40% AI enrichment coverage

Around 40% of product titles and descriptions were being enriched and optimised through AI at the time of the case publication.

These are the kinds of AI results that matter.

Less repetitive work.

Faster product onboarding.

Better information.

A workflow teams can actually use.

11

Why this matters

AI becomes valuable when it sits inside the operation.

Generating one product description with AI is easy.

Improving product data across a large catalog is a systems problem.

You need:

  • Reliable source data

  • Taxonomy

  • Business rules

  • APIs

  • AI models

  • Structured attributes

  • Review

  • Approval

  • Integration

The value comes from connecting those pieces.

That turns an AI feature into operational infrastructure.

12

Better data creates more than better copy

The first impact is efficiency.

The longer-term opportunity is broader.

Cleaner product information can support:

  • Better search

  • Better filters

  • Stronger SEO

  • Better recommendations

  • Fewer customer misunderstandings

  • More useful AI shopping experiences

  • Faster product onboarding

The enrichment workflow therefore becomes part of the wider commerce data foundation.

AI services used

Data Infrastructure

Connecting product systems, taxonomy and AI enrichment into one reliable data workflow.

AI Agents

Turning repeated product-content work into a structured enrichment and approval process.

Software Engineering

Integrating the AI layer directly with the retailer's existing technology and product administration.

Avatar-1
Avatar-2
Avatar-4
Avatar-4
Avatar-3

Still improving product data one item at a time?

AI can turn catalog enrichment from repetitive content work into a scalable data process. We help connect product information, AI and approval directly into existing commerce operations.