Anyone who has run a serious B2B catalogue knows the quiet truth behind every storefront: the product data is never finished. Descriptions are thin or inconsistent. Half the attributes are missing. The same product reads beautifully in French and barely at all in English or Spanish. Multiply that across tens of thousands of references and several languages, and product content stops being a content problem. It becomes an operational one.
For the last two years the industry has reached for the same answer, and it is the right one: artificial intelligence. What has changed at Aleph Front is not whether we use AI for this, but how, and crucially, where it runs.
What everyone is doing, and why
The market has converged quickly. Across enterprise e-commerce, AI is now the default engine for what the industry calls product data enrichment: generating descriptions, completing specifications, classifying products into the right taxonomy, normalising inconsistent values, and filling missing attributes at scale. The reason is simple. Modern shoppers, and increasingly the answer engines they rely on, abandon listings that feel incomplete or confusing, and they reward catalogues that are rich, structured, and machine readable.
Three uses have become standard practice:
- Description generation. AI drafts benefit-driven, SEO-aware product copy from whatever raw material exists, including spec sheets, supplier PDFs, and existing attributes.
- Translation and localisation (i18n). Catalogues that need to live in French, English, Spanish, and beyond can be localised consistently, with terminology that stays coherent across the whole estate rather than drifting product by product.
- Attribute enrichment. AI detects the empty fields, infers values from existing data, standardises colours, materials, and sizing formats, and maps everything to the category rules that marketplaces and search systems demand.
This is genuinely valuable, and we believe in it. But there is a catch that most of the market quietly accepts, and we do not.
The catch nobody likes to mention
Almost all of these capabilities are delivered the same way: your product data, your supplier information, and sometimes your pricing logic are sent to a third-party AI service running on someone else’s infrastructure. Every enrichment call leaves your perimeter. You pay per token, so costs rise in lockstep with your catalogue and never stop. And you inherit a dependency on a vendor whose pricing, availability, and data-handling policies you do not control.
For many businesses that trade-off is acceptable. For the clients we serve, distributors, industrial groups, and purchasing networks handling sensitive commercial data, it is increasingly not. Data residency and sovereignty have moved from a nice-to-have to a hard requirement, and the per-token economics of cloud AI become punishing exactly at the scale where enrichment matters most.
Our approach: local, independent AI at the front
This is where Aleph Front takes a different path. Rather than wiring your catalogue into a third-party AI service, we train and run local, independent AI models that sit inside your own environment, at the front of your commerce platform.
The principle is straightforward. Modern open-weight models have become good enough, and small enough, that a well-chosen and properly tuned model running on your own hardware can handle description generation, translation, and attribute enrichment to a professional standard, without your data ever leaving your control. We select the right model for the task rather than the largest one, tune it to your catalogue, your terminology, and your brand voice, and integrate it directly into your enrichment pipeline.
The benefits compound:
- Data sovereignty by design. Your product data, supplier information, and commercial context stay inside your perimeter. Nothing is shipped to an external service to be processed, logged, or used elsewhere.
- Cost that stops scaling against you. A local model replaces an open-ended per-token bill with a predictable infrastructure cost. Past a certain catalogue size, which most of our clients comfortably exceed, this is simply cheaper, and it does not penalise you for growing.
- Independence. No lock-in to a single AI vendor’s pricing trajectory or roadmap. The model is yours, tuned to your business, and it keeps working on your terms.
- Consistency you can govern. Because the model is tuned to your terminology and rules, translation and enrichment stay coherent across the whole catalogue, with human validation and guardrails where they matter rather than blind automation.
Why this fits the way we build
This is not a bolt-on. It is the same philosophy that runs through everything we deliver on Intershop: capabilities presented as managed services, integrated cleanly into the commerce platform, and built to keep our clients in control of their own data and their own costs. Product content is one of the highest-volume, highest-value data flows in any catalogue. It deserves to be enriched intelligently, and it deserves to stay yours.
AI belongs in your catalogue. It does not have to leave your house to get there.
If you are wrestling with thin descriptions, multilingual catalogues, or incomplete attributes across a large Intershop estate, we would be glad to talk about what a local, sovereign enrichment pipeline would look like for you.
