Drafts from source data
Product copy or attribute suggestions, fed by PIM or ERP fields. The model never fills in the price or the stock level itself.
Service
AI automation for e-commerce
AI here is not a chatbot that replaces your team. It is a way to shrink repetitive work: draft copy from source data, a summary of orders that get stuck, an internal answer to a question about the catalogue. A person publishes, until the source data is boring and reliable.
At a glance
- Drafts from source data
- Making exceptions readable
- Internal questions about the catalogue
- A boundary around the model
Custom quote after a free intro call
What we do
What is included
Making exceptions readable
Orders or sync errors that now disappear into a log, turned into a short list for the person who has to fix them.
Internal questions about the catalogue
A colleague asks the company's own data something ('which items have no image but do have a B2B price'). That is different from a bot on the front of the shop.
A boundary around the model
Which model, which data it may see, and whether that data leaves the EU. We settle that per project before connecting anything.
Who it is for
Who this fits
- Teams retyping product copy by hand from an ERP or PIM.
- Customer service fishing the same order exceptions out of a log every morning.
- Organisations that already run a Medusa shop with data clean enough for a model to read something useful.
Honest advice
When we advise against it
- An autopilot that publishes product pages without review. We do not build that.
- A project where the source data is messy and AI is supposed to hide that. Data first.
- Reselling AI tools. We do not use affiliate links and we do not install a tool because there is a commission on it.
Price
What determines the price
A custom quote after a free intro call. These factors determine the price, not a menu.
- State of the source data
- Clean fields make a small automation. Free-text fields full of exceptions make a clean-up project.
- Which system is the source of truth
- The model reads. It only writes back to a place where a person or a workflow can still stop it.
- Volume and review
- Ten drafts a week are different from a catalogue of tens of thousands of items with an editor in the loop.
- Model and data flow
- A model that stays in the EU and needs little context is different from a heavy model receiving entire order files. We only do the latter with an explicit agreement.
Process
The process
- 01
One task, not a platform
We pick one repetitive task that hurts. Not an 'AI strategy' with thirty use cases.
- 02
Pilot on real records
A small number of products or orders. We measure how often a person has to step in. We do not make that number up in advance.
- 03
Inside the workflow
The output lands where someone already looks: a draft in the PIM, a list in the admin, an email to customer service.
- 04
Stop if it does not work
If the pilot mostly produces nonsense, we stop. That is an outcome, not a failed sale.
Insights
Further reading
From this service to the articles, and from each article back again.
- ArticleMigrating to Medusa: why, and how to do it rightWhy companies move to Medusa, from which platforms it makes sense, the six steps of a good migration and when you are better off staying.Read more
- ArticleWooCommerce vs Medusa: when WordPress holds the shop backWooCommerce and Medusa compared on extensions, hosting, three-year cost, B2B and multiple shops. With real extension prices and our Shingyo project.Read more
- All articlesPlatform comparisons and the migration guide, with sources.To the articles
Do you put AI on product pages without review?
No. A model never publishes product facts on its own with us. First a draft, then a person, then the page. Once the source data is clean and stable, we can shorten that loop. That is a separate decision.
Which tasks are suitable?
Repetition with a source: copy from existing fields, summaries of errors, internal questions about data already in Medusa or the PIM. Open-ended advice to your end customers ('buy this too') only if someone can vouch for the quality.
Does our data have to go to an external model?
Sometimes. Which model and which processor, and whether data leaves the EU, we write down per project before connecting. Without that agreement, no customer data leaves the building.
Does this replace an ERP integration?
No. Stock, price and order status belong in a regular integration, not in a language model. AI on top can explain what a failed sync means.
What if our source data is messy?
Then the output becomes confident and wrong. We do not start with a model then, but with the field that is wrong. Less spectacular, and the only thing that works.
Free intro call
What is your challenge?
Twelve shops that need to become one, an ERP that sets the price, or a checkout that does not do what you want. Tell us in one call. You will hear honestly whether Medusa fits.
Based in Nijmegen, the Netherlands, working with companies across Europe. How we work with European teams.