How AI is transforming product data management in PIM

What this episode covers

The tenth episode covers the two AI features most used in PIM systems today: generating translations and generating product descriptions. Karolina shows how they work, which licences include them, and what specifically to put in a prompt so the output is publishable.

Key takeaways

• Translations are generated from a button next to the text attribute, with no copying into another tool. It takes a few to a dozen seconds.
• A custom prompt solves the terminology problem: you supply an industry glossary, the purpose of the translation, the audience and the brand tone.
• Descriptions are built from selected attributes. The prompt sets structure, length, tone, target channel and whether to skip empty attributes.
• The mechanism ships in paid Akeneo licences and in every Ergonode edition. In Akeneo Community it works through plugins, including ours.
• Always review AI output. An approval workflow in the PIM system should hold a description back until somebody signs it off.

Translations without leaving the system

In Akeneo on the Grow, Advanced and Premium licences, and in every edition of Ergonode, the translation mechanism is built in. A button appears next to a text attribute: you pick the source language version and the target language, and after a few or a dozen seconds you see the result in the right locale field.

What matters is not the speed as much as the absence of context switching. You do not copy text into another tool and carry it back, so there is no place to lose a version.

The prompt is where terminology gets solved

The most common worry is specialist vocabulary, where there is little room for interpretation. The answer is the ability to write custom prompts. What is worth putting in one:

• the purpose of the translation: marketing, technical or formal,
• the audience: experts or consumers, because the style follows from it,
• a glossary, meaning key terms with the preferred translation, if your industry has its own vocabulary,
• formatting to preserve, for example bulleted lists,
• naturalness in the target language: avoiding calques, with permission to paraphrase where needed,
• the brand tone, ideally with an example of an earlier translation,
• a character limit, and whether shortening is allowed or the translation should be one to one.

The last hint is the most important: it is worth stating that the translation should be clear and natural, not merely linguistically correct.

If you run the open source edition

In Akeneo Community the feature is not included by default, but it can be added with plugins. Our team wrote a plugin connecting the PIM system with ChatGPT to support translations, and alongside it an external application for translations and for generating descriptions from other attributes. There is more about it on the Akeneo OpenAI Translator page.

Descriptions generated from attributes

The second feature creates a new description from data already in the system. You indicate which attributes AI should use: product name, dimensions, marketing features. The prompt also lets you set boundaries, for example that the product name must not change and certain device parameters must not be altered.

For description prompts it is worth settling: the structure (heading, feature list, summary), the sections to keep (general description, applications, specification), the list of attributes to include, whether to skip empty and incomplete attributes, the tone, industry vocabulary, the length (for example up to 500 characters), the target channel, because a description for Amazon, for a marketplace and for your own e-commerce are three different texts, and whether to include SEO keywords.

Writing the first prompt takes a while. It pays back on every product page after that.

Data quality, and why a workflow

Despite how advanced the technology is, the recommendation is unambiguous: content generated by AI has to be checked. So it is worth implementing a review workflow in the PIM system, so that somebody responsible verifies and approves content before it reaches the shop.

The benefits are three: speed, since a description or translation takes moments; savings compared with commissioning translations and copy externally; and flexibility, because custom prompts enforce the company’s standards.

What is coming

This is only the beginning. One direction being tested is importing unstructured product data into a system whose structure already exists, and using AI to assign that data to the right attributes. There are more ideas, but each needs repeated testing before it becomes real support.

Want to see these features live in the system? Book a free consultation. More on the PIM systems page.