E-commerce AI automation for manufacturers.
I build automation for manufacturers with their own online stores: preparing product data, supporting internal teams, and preparing B2B enquiries. I connect existing systems such as Shopware, PIM, and ERP. Our first conversation takes about 30 minutes and is free, with no obligation.
POSSIBLE APPLICATIONS
Three possible projects for your online store
I focus on manufacturers selling their own products through B2B or B2C stores. Suitable retailer projects are welcome too. These examples show what a possible implementation could look like. Other workflows can also be a useful starting point.
Prepare and connect product data
Product information is scattered across data sheets, spreadsheets, and existing records. Your team copies details manually and has to resolve conflicting values.
- Workflow: Read source material, map attributes, and check them against agreed rules, then create a draft for the PIM or store.
- Result: Product drafts with traceable sources, flagged gaps, and validation results. Sample data lets your team check whether the mapping is correct.
- Responsibility: Product data owners resolve uncertain information and approve publication. Missing attributes are not filled in simply because they seem plausible.
Support your internal team
Recurring questions about store operations reach the same technical colleagues. Answers depend on documentation, Shopware data, and previous decisions.
- Workflow: An internal assistant searches approved knowledge and retrieves permitted system data. It provides answers with sources and system context.
- Result: Verifiable answers or a prepared handover with findings and unresolved questions. Representative support cases show where the assistant helps and where it does not.
- Responsibility: Unresolved cases go to technical staff. The assistant supports your internal team; each permission and possible action is defined explicitly.
Prepare B2B enquiries
Enquiries arrive by email with attachments and inconsistent product references. Sales staff collect information from documents, product records, and, where relevant, ERP customer data.
- Workflow: Capture the request and line items, match available information, and prepare follow-up questions and quotation documents.
- Result: A structured enquiry with sources, missing details, and a reviewable draft. Real enquiry examples are used to check whether line items and follow-up questions are complete.
- Responsibility: Sales staff check prices, terms, and recipients and approve the quotation. The workflow makes no independent pricing commitments.
MY IMPLEMENTATION WORK
Experience with established commerce systems
I bring 15 years of software engineering and around ten years of experience with complex e-commerce systems. I have worked with agentic AI since 2023. These anonymised examples come from my own implementation work.
- Connected PIM and Shopware
- I built custom middleware connecting PIM and Shopware and implemented several brand storefronts with multilingual versions.
- Implemented B2B workflows
- I synchronised B2B customer groups with the ERP and built a quotation download that lets buyers obtain internal approval for their baskets before ordering.
- Built an internal Shopware assistant
- I built an internal support assistant with access to Shopware code, the database, and PIM middleware. It can investigate questions and escalate cases to technical staff with the conversation context.
What we need to understand about your process
Where does manual work occur today? What data arrives, which systems are involved, and how does your team recognise a correct result? We also establish available interfaces and access, subject-matter owners, and the decisions that should remain with people.
Shopware, PIM, and ERP are examples of existing infrastructure; you do not need all three. A rough description is enough for our first conversation. Before implementation, representative cases, common exceptions, and clear responsibilities help define the work. Access is limited to the data and actions actually needed.
How a working process takes shape
Input → retrieve data → rules and AI step → validation → approved action. I connect the necessary interfaces and build the processing logic, including error handling. Fixed rules handle unambiguous tasks; AI can help interpret inconsistently structured documents. An agent is useful only when it genuinely needs to choose between tools or next steps.
A hypothetical failure case: A data sheet says aluminium, while a spreadsheet says stainless steel. The workflow flags the conflict, holds back the product draft, and passes both sources to the responsible specialist. The change is passed on only after the conflict is resolved. Logs make decisions and failed steps traceable.
How we can work together
We start with a free, non-binding conversation of about 30 minutes. You describe your process; we discuss your goal, existing systems, and open questions. We then decide together whether working together makes sense and what form it should take. Sometimes the right next step is no project for now.
A clearly defined problem may be ready for direct implementation. If causes or options remain unclear, a commissioned analysis may help. A prototype can test a specific technical question. An audit is one possible engagement, not a mandatory first stage. We agree the scope, fee, and deliverables before you commission any work.
When a pilot makes sense
A pilot can test a bounded workflow under agreed conditions. Possible deliverables include a connected test workflow, representative test cases, documented error handling, and an evaluation: What works, what needs human review, and which prerequisites are still missing?
Handover, documentation, and a recommendation for ongoing operations can be part of the engagement. We agree in advance which data, systems, and case volumes are included. A successful pilot does not automatically promise a complete production system. Production access, monitoring, maintenance, and responsibility need a separate agreement.
Questions before working together
What determines the cost?
Process scope, data quality, available interfaces, testing effort, and approval and operating requirements. Ongoing model, hosting, and maintenance costs also matter. The introductory conversation is free; commissioned work receives an individual proposal.
Do you work with other e-commerce platforms?
Yes, when the task and technical access are a good fit. Shopware is an area of experience, not a restriction. Available data, interfaces, and responsibilities are what matter. Switching platforms is not a general prerequisite.
Can our IT team or existing agency stay involved?
Yes. We agree interfaces, responsibilities, and approvals with the people involved. I can implement a clearly scoped workflow while working alongside your IT team or agency.
Who handles ongoing operations?
We define this in the project scope: hosting, credentials, monitoring, incident handling, maintenance, and handover need named owners. Ongoing support is not automatically included, and a pilot does not necessarily cover every production requirement.
What are the limits of automation?
Prices, stock levels, and permissions come from the responsible systems. Missing facts are not invented. Sensitive data, permitted actions, and human approvals are agreed in advance. If data or permissions are missing, the workflow must stop or hand the case over.
Which workflow would you like to simplify?
Briefly describe where manual work occurs today and which systems are involved. In our free introductory conversation, we will explore a sensible next step together. You do not need a finished project brief.