Ecommerce ops · Production

Zalando Content Creation Copilot: AI-assisted product attribute extraction from images

The problem

Zalando's product content creation was a largely manual process accounting for approximately 25% of the overall content production timeline, with consistent opportunities to reduce error rates identified through QA analysis and insufficient attribute coverage across the product catalog.

Workflow diagram · grounded in source

Photographer uploads product images

trigger

“Internal content creation tool used by photographers to upload images, which URLs are sent to the Prompt Generator”

Prompt Generator builds attribute prompts

integration

“Generate prompts based on the attributes and attribute sets coming from Article Masterdata. The prompts and image URLs are sent to OpenAI-GPT for further processing.”

OpenAI GPT generates attribute suggestions

ai_action

“Processes the prompts received from the Prompt Generator and provides suggestions based on the prompts. The suggestions or content are sent back to the Content Creation Tool.”

Translation layer converts AI output to attribute codes

integration

“We built a translation layer that converts OpenAI output into information directly usable by Zalando and discards the part that is not relevant.”

Attributes pre-filled in Content Creation Tool

output

“Attributes are now marked with purple indicator (dot) and pre-selected for suggestions coming from the prompt generator in Content Creation Tool”

Copywriter reviews AI-suggested attributes

human_review

“allowing users to concentrate more on QA rather than the time-consuming task of enriching content”

Article published in shop

output

“After QA is completed, the article is published in the shop”

Reported outcome

The Content Creation Copilot achieved approximately 75% attribute suggestion accuracy and enriches around 50,000 attributes per week on average, with improvements in both data quality and coverage of attributes and smooth adoption by content creation teams.

Reported metrics

Manual process share of content production timelineapproximately 25%

Attribute suggestion accuracy rateapproximately 75%

Attributes enriched per weekaround 50,000

Reported stack

Content Creation ToolPrompt GeneratorArticle MasterdataOpenAI GPT-4 TurboGPT-4oaggregator service

Source

https://engineering.zalando.com/posts/2024/09/content-creation-copilot-ai-assited-product-onboarding.html

Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

The Content Creation Copilot achieved approximately 75% attribute suggestion accuracy and enriches around 50,000 attributes per week on average, with improvements in both data quality and coverage of attributes and sm…

What tools did this team use?

Content Creation Tool, Prompt Generator, Article Masterdata, OpenAI GPT-4 Turbo, GPT-4o, aggregator service.

What results were reported?

Manual process share of content production timeline: approximately 25%; Attribute suggestion accuracy rate: approximately 75%; Attributes enriched per week: around 50,000 (source-reported, not independently verified).

How is this ecommerce ops AI workflow structured?

Photographer uploads product images → Prompt Generator builds attribute prompts → OpenAI GPT generates attribute suggestions → Translation layer converts AI output to attribute codes → Attributes pre-filled in Content Creation Tool → Copywriter reviews AI-suggested attributes → Article published in shop.