AI implementation and process automation

How I help

Four situations where I can help

A
You are launching sales in Germany or through a GmbH
- Starting in Germany from the operations side: Amazon and eBay accounts, Billbee, the order of steps and what to set up along the way.
- Setting up invoicing, document flow and system configuration for German VAT, intra-EU supplies and OSS, as agreed with your accountant or Steuerberater.
- Shipping from a warehouse in Poland and a warehouse in Germany, with consistent stock and orderly paperwork.
- German registration duties from practice: VerpackG (LUCID), WEEE (ElektroG) and related, registered and verified in my companies. I know what needs to be in place before sales start.
This is operational experience, not legal or tax advice. Settle legal form and taxes with a tax adviser (in Germany: Steuerberater) and a lawyer.
B
You sell on Amazon, eBay and Kaufland and want to do it better
- Listings and variations across the whole catalogue and A+ content, on Allegro too.
- GPSR data: I automate how it is compiled and published. The manufacturer or importer remains responsible for product compliance.
- Listing analytics on historical data (including Keepa): keywords, ranking, listing quality and compliance, with recommended changes.
- Margin-based repricing on eBay and Amazon: the selling price only drops to the break-even point calculated from purchase costs, fees and shipping. Below it the system does not cut, it flags.
- Amazon, eBay and Google advertising judged by profit, with ROAS calculated per SKU instead of guesswork.
C
Gambio, Billbee, Apilo: you want one source of truth for stock and orders
- One stock level for every channel and both warehouses: a sale in one place removes the item everywhere.
- Billbee: sale and return webhooks, stock sync, documents for printing.
- Gambio, PrestaShop and Apilo: variations, bulk changes, automatic order import, custom website modules.
D
You want AI that does the work, not the slide deck
- AI implementation beyond e-commerce too: office work, B2B trade, customer service, documents, reports. Every company has its own processes, so we start by reviewing what repeats and what needs a person.
- Agents prepare listings, copy, replies and reports; a person assesses the result.
- A helpdesk with suggested replies, automatic printing of invoices and labels.
- The automation can run in your infrastructure or partly in mine. We agree on that during implementation.
- Your own integration saves subscriptions but brings its own maintenance. Compare both in the tool cost calculator.

Full competence map and boundaries

In production runs in my business every day, with results
Developing first deployments and the basics
Not my field a different specialisation

Competency map: area, level and evidence from practice
Area	Level	Evidence from practice
Stock sync across channels and warehouses	In production	One source of truth for every channel and two warehouses (Poland, Germany), with a complete change log.
Amazon (SP-API, A+, GPSR)	In production	Several accounts, A+ content, GPSR data; listing analytics on Keepa data (keywords, ranking, compliance); margin-based repricing.
eBay, Kaufland, Allegro	In production	Bulk listing, margin-based repricing on eBay with a break-even floor, an auto-lister.
Billbee and Apilo	In production	Webhooks, stock sync, order import, several accounts on one back end.
Gambio: modules, templates, cross-selling	In production	Custom modules (including a live “amount left to free shipping” bar in the basket), listing templates per product, cross- and upselling from catalogue data.
PrestaShop 9	In production	A website synchronised with the system via API: stock, prices, orders.
Advertising: Amazon, eBay, Google	In production	Campaigns via API, a daily bid review, judged by profit.
Invoicing and document flow PL⁠–⁠DE	In production	Automatic invoices and system configuration for VAT, intra-EU supplies and OSS, as agreed with the accountant.
Product copy and video with AI	In production	Titles, descriptions and A+ written for the product and the platform; short clips from a single photo.
Facebook and Instagram posts	In production	Posts created from the catalogue and published automatically.
Helpdesk and document printing	In production	Several mailboxes in one view, reply drafts, a phone line with a voice bot (sipgate), automatic printing of invoices and labels.
Agents and custom AI tools	In production	More than 60 tools in production, listed in the tool catalogue.
Meta Ads	Developing	Campaign basics; Messenger runs in the helpdesk.

Evidence

Results from my own businesses

1.2 h
median time to the first human reply in the helpdesk; in June 2026 it was 9.7 h. Over the same period the number of conversations grew fivefold.

June → September 2026 (to 25 September). Elapsed hours, nights and weekends included, excluding automatic acknowledgements.
Method
−53%
eBay ad spend for the same ad-attributed sales, after rebuilding the campaigns around margin.

June → August 2026. Ad-attributed sales according to eBay reports.
Method
9,000+
new listings published automatically on Amazon, eBay, Allegro and Kaufland.

Since February 2026 (to 25 September). Distinct new listings, excluding imports and updates of existing ones.
Method
c. 230
supplier orders prepared by the system and sent via API. A person approves the quantities.

4 weeks: 31 August to 25 September 2026.
Method

As of 25 September 2026. Sources, definitions and method: in the case studies.

Case studies and all figures Tool catalogue

How it works for me

From single scripts to one system

My first catalogue upload was 19 separate scripts with instructions on which order to run them in, and I fixed every line by hand. That is how I still worked in 2024. Today an agent loads the same catalogue and I approve the result.

I have run my own e‑commerce business since 2011, and worked in sales before that. I have tested most of these solutions in my own companies, including the mistakes that cost time and money.

I am not a senior developer with fifteen years of programming behind me. My strengths are business know-how and understanding problems where e-commerce, sales, the warehouse and systems meet. I have been building programming experience for a long time in daily work with AI (Claude Code, Codex) and in running my own systems in production. The Anthropic certificate (Foundations level) confirms the fundamentals of this work.

2011: my own e‑commerce for Austria and Germany

First websites on PrestaShop for customers in Austria and Germany, shipped from Poland. I have been growing sales on the German market ever since. This is where I learned that margin disappears in fees, shipping and returns.
A Polish company, then a German Einzelunternehmen

I first sold to Germany as a Polish company. Then I registered a sole proprietorship (Einzelunternehmen) in Germany and sold from within Germany as well.
A GmbH where I am managing director

Today I also sell through a GmbH where I am managing director (Geschäftsführer). I know both forms in Germany, Einzelunternehmen and GmbH, from registration to day-to-day operations.
3 companies, 2 warehouses, 1 system

The biggest challenge turned out to be the flow of goods and information between two countries, two warehouses and three companies. Since January 2026 I have been bringing it into order in my own system.

How it runs today: one system for three companies and two warehouses

Point at or tap a node to see what happens there and what it is connected to.

Node

1 system

The operational core of my companies: one source of truth for stock and orders across both warehouses, where every sale, return, delivery and transfer is an event in the log. A margin controller and margin-based repricing protect profitability, and I calculate ROAS per SKU. The system analyses stock rotation, proposes purchase orders and reorders via API in suppliers’ systems. Stock goes out to every channel automatically, invoices and labels to the print queue.

All nodes described in text

1 system: The operational core of my companies: one source of truth for stock and orders across both warehouses, where every sale, return, delivery and transfer is an event in the log. A margin controller and margin-based repricing protect profitability, and I calculate ROAS per SKU. The system analyses stock rotation, proposes purchase orders and reorders via API in suppliers’ systems. Stock goes out to every channel automatically, invoices and labels to the print queue.
Warehouse in Poland: Each of the three companies can buy and ship goods from here. Stock is synchronised with the warehouse in Germany.
Warehouse in Germany: Daily transport between the warehouses, stock synchronised in the system. Every company can ship from here.
Company 1: A GmbH in Germany: its own marketplace accounts and invoices, goods from both warehouses.
Company 2: A second company selling in Germany: separate accounts and bookkeeping, shared stock across both warehouses.
Company 3: A company in Poland: operations and sales on the Polish market, including my own website domabau.pl.
Billbee · 2 accounts: One account per German company. Billbee receives the sale and return webhooks, and current stock goes back to it in bulk. I keep it deliberately as an intermediary between the marketplaces and my system: it is well established on the German market, meets German requirements and covers the GDPR side, meaning retention, archiving of customer data and documents.
Apilo: Allegro and domabau.pl orders run through Apilo. The system regularly collects paid orders and reports current stock back, and Apilo passes the data on to accounting.
PrestaShop 9: The domabau.pl website is connected to the system directly via API (stock, prices), and orders and stock also run through Apilo.
Amazon DE + EU: Several accounts, listings in Germany and other EU countries. A+ content, GPSR data, listing analytics and margin-based repricing.
eBay DE + EU: Bulk listing with variations in Germany and other EU countries, margin-based repricing, advertising judged by profit and a daily check of images and listing data.
Kaufland DE + EU: Listings in Germany and other EU countries, with stock kept in line by the same system as every other channel.
Amazon, eBay, Kaufland, Gambio: Marketplaces in Germany and the EU plus three own Gambio websites: A+ content, GPSR data, bulk listing with variations and margin-based repricing.
3× Gambio: Three own websites on Gambio with custom modules, listing templates and cross-selling from catalogue data.
Allegro: An auto-lister prepares listings for slow-moving stock, published with a single click.
domabau.pl: My brand on the Polish market, a website on PrestaShop 9 fed from the same stock as every other channel.
AI helpdesk: Email, chat, messenger and phone in one view. On the phone, in cooperation with sipgate GmbH, voice bots take the caller’s details and, if the number is in the system, recognise the customer and the order, so nobody has to dictate an email address or phone number. The call becomes a ticket, which a person then handles with AI support.
Print queue: Invoices and labels print automatically on the right printer, with a retry when a printer briefly drops off the network.
Accounting (iFirma): Orders from Apilo go to accounting: invoices and settlements are created in iFirma without retyping data.
Advertising: Amazon Ads, eBay and Google Ads: a daily review and bid proposals, judged by profit rather than revenue.

The system I built: StockHub

All sales of my companies are overseen in one system. We used to work across several panels and spreadsheets; today there is one view and a short list of decisions.

AI does not replace order: the best model connected to a mess only produces a faster mess, so every implementation starts with clean data.

Each company has its own accounts, permissions and documents; the system shares only warehouse stock and logic. Margins stopped being a gut feeling: purchase costs and every sale are in the system, so I see straight away whether an item makes money.

Lessons base. For more than a year I have recorded every pitfall from integrating Billbee, Gambio, eBay and Amazon. With a client I do not start from zero.

Obsidian as the company’s second brain. I keep the company’s knowledge, meaning processes, lessons from integrations and decisions, in Obsidian. I use it every day, and AI also reads part of that knowledge when working on the system.

Official APIs, not scraping. I hold developer accounts with eBay (Developer Program), Amazon (SP-API and Ads API), Google (Ads API, Cloud) and Meta (Facebook, Instagram). They are what makes it possible to connect every system through official interfaces.

30 integrations with external APIs · more than 80 background workers · more than 100 admin views

StockHub is my own system, built with Claude Code. I do not sell it as a product; for clients I build a solution around their processes.

How I implement

Dry run first. I check changes in a dry run and a test environment before they touch real listings and stock.
From a specific version. Deployment from a specific Git commit, with service health checks and a rollback ready.
Separate permissions and secrets. Access split per system, no keys in the code.
Monitoring and a log. Errors go to Sentry, every stock change to the event log.
Handover. Code and documentation stay in the client's repository.

All system features (15)

Margin controller
An alert when an item’s margin drops below the threshold, calculated after all fees and shipping costs.
Margin-based repricing on eBay and Amazon
The selling price only drops to the break-even point from purchase costs, fees and shipping. Below it the system does not cut, it flags.
ROAS per SKU
My own calculators show when advertising for an item pays off and when it does not, instead of guesswork.
Rotation and supplier orders
Stock rotation analysis, purchase proposals and sending the order to the supplier via API with one click.
B2B portal for suppliers
Suppliers see how their products sell and can plan deliveries accordingly.
Helpdesk instead of a ticketing system
Email, chat, messenger and phone in one view. AI prepares quotes and reply drafts, so the team works noticeably faster. Since August 2026 it has replaced LiveAgent.
Phone line with a voice bot (sipgate)
Voice bots take the caller’s details and recognise the customer by number. The call becomes a ticket in the helpdesk.
Team goals board
A goals board shows the team its daily sales progress.
Advertising review
Amazon Ads, eBay and Google Ads: a daily review and bid proposals.
Facebook and Instagram from the catalogue
Posts created from the catalogue and published automatically.
Auto-listers with content checks
Listings for eBay, Amazon, Allegro and Kaufland prepared automatically and checked before publishing.
Amazon listing analytics
Historical data from Keepa, keywords, ranking, A+, GPSR and compliance, with change recommendations.
Gambio websites
Custom modules (for example a live “amount left to free shipping” bar in the basket), listing templates per product, cross- and upselling from catalogue data.
Print queue
Invoices and labels print automatically on the right printer in the warehouse.
Stock event log
Every stock change recorded with its cause: what happened, when and from which channel.

Automation in practice

What runs automatically and what a person decides

Stock, prices and rules are handled by code. AI prepares copy, replies and proposals that a person assesses.

System log · one morning sample run based on real events

06:40syncMarketplace order → stock deducted in 4 channelsok
06:41printInvoice and label → warehouse printerok
06:52listings12 new listings → content check: 12 of 12 without violationsok
07:05adsBid review: 3 proposed reductions → preparedwaiting for a person
07:18helpdeskDelivery date question → reply draft with order datawaiting for a person
07:30stockSupplier delivery booked in → items available in every channelok
07:31controlStock mismatch on 1 listing → corrected and loggedok

These items are only prepared by the system; a person publishes them.

Example: listing copy generator

This is what the generator produces for me: a product description becomes a German title and bullet points for a specific platform, and the copy goes through checks before publication.

Input · product description

Wall bracket for shelving made of galvanised steel, 3 mm sheet. The load arm is 250 mm long and the wall leg 150 mm high; suitable for shelves up to 250 mm deep. A pair of brackets holds up to 40 kg with the load spread evenly across the shelf. Two Ø 6 mm holes, 100 mm apart, for screws up to 5 mm; screws and wall plugs are chosen to suit the wall material. The zinc coating protects against corrosion indoors. For shelving in garages, cellars, workshops and pantries. Contents: 1 bracket, without screws or wall plugs.

Output · copy in German

Example resultAI-generated

Amazon.de

Wandkonsole für Regale aus verzinktem Stahl, 3 mm Blech, Arm 250 mm, Tragkraft bis 40 kg pro Paar, für Garage, Keller und Werkstatt

Title: 131 / 200 characters

ROBUSTES MATERIAL: Die Konsole ist aus 3 mm starkem, verzinktem Stahlblech gefertigt. Die Zinkschicht schützt das Metall in Innenräumen vor Korrosion.
PASSENDE MASSE: Der tragende Arm ist 250 mm lang, der Wandschenkel 150 mm hoch. Damit eignet sich die Konsole für Regalböden mit einer Tiefe von bis zu 250 mm.
TRAGKRAFT: Ein Konsolenpaar trägt bis zu 40 kg. Der Wert gilt für eine gleichmäßig verteilte Last auf dem Regalboden, der auf beiden Konsolen aufliegt.
EINFACHE MONTAGE: Zwei Bohrungen mit Ø 6 mm im Abstand von 100 mm nehmen Schrauben bis 5 mm Durchmesser auf. Schrauben und Dübel wählen Sie passend zum Material Ihrer Wand.
EINSATZ UND LIEFERUMFANG: Für Regale in Garage, Keller, Werkstatt und Vorratsraum. Geliefert wird 1 Wandkonsole, Schrauben und Dübel sind nicht enthalten.

eBay.de

Wandkonsole für Regale, Stahl verzinkt 3 mm, Arm 250 mm, bis 40 kg pro Paar

Title: 75 / 80 characters

Stahlblech 3 mm, verzinkt, für Innenräume
Arm 250 mm, Wandschenkel 150 mm, für Böden bis 250 mm Tiefe
Tragkraft bis 40 kg pro Paar bei gleichmäßiger Last
2 Bohrungen Ø 6 mm, Abstand 100 mm, Schrauben bis 5 mm
Lieferumfang: 1 Konsole, ohne Schrauben und Dübel

No shipping information in the copy
No return terms in the copy
No third-party trademarks
Only facts from the input
Platform character limit respected

I show the generator live in a call, on your product.

Why my own tools rather than another ready-made one

Marketplace requirements change often. Attributes, policies and formats on eBay and Amazon can change from one week to the next. I apply changes faster because I write and fix integrations in hours, not weeks.
An example from my own business: rattan cut to size. Cutting from full rolls needed its own stock logic: an order goes to an exactly matching offcut, then to the smallest piece that is long enough, and only then to the largest available. It is plain program logic, covered by tests and written with Claude Code in a few hours. No ready-made tool had it.
An example from my websites: Gambio modules built to my own rules. The “amount left to free shipping” bar in the basket, listing templates per product and cross-selling from catalogue data are modules I wrote myself with Claude Code. Each module is built for a specific website and its rules, instead of fitting the business to a ready-made extension.

About me and credibility

A practitioner who builds his own tools

Marcin Dobrzyński with his hand at his chin, portrait against a dark background

Marcin Dobrzyński, AI implementation and process automation specialist based in Zielona Góra, Poland, and a Polish–German e-commerce practitioner. He implements Claude AI in companies using tools he runs in production himself.

Hi. I have been trading online for fifteen years. In Poland I run my own company, and in Germany I am the managing director (Geschäftsführer) of a GmbH in German e‑commerce.

I work in Polish, German and English. E-commerce is what I know best, but the same methods work in office work, B2B trade and customer service.

My team works with Claude every day. Ideas for automation come from the warehouse and the sales office, and some team members design and deploy them themselves.

Code, documentation and access stay with the client, in their repository and on their accounts. After implementation you do not depend on me: you can carry on yourself, with someone else or with me.

My own businesses show me where AI actually helps.

My LinkedIn profile

Certificate and affiliations

Claude Certified Developer – Foundations

Issuer: Anthropic
Issued: 24 September 2026
Valid until: 24 September 2027
Verification: Badge profile on Credly
Certificate: Certificate (PDF) PDF · 164 KB

The exam is run by Pearson VUE on behalf of Anthropic, entirely in English. It covers the Claude API, Claude Code, designing tools and MCP servers, and building and deploying agents and production applications.

Sellon is a member of the Claude Partner Network (Registered tier).

Claude and Claude Partner Network are trademarks of Anthropic, used here only to state this fact. The Registered tier means participation in the programme, not a recommendation or authorisation by Anthropic.

Team

I am looking for people for my team

I am building a team of people who work with Claude seriously. Ten people is the goal.

I want my marketplace agency Sellon to reach the next level of Anthropic's partner programme, and that requires a larger team of certified people. Together we want to take on projects for larger companies and corporations.

I am looking for nine people with the Claude Certified Developer or Claude Certified Architect certificate, or on the way to it. What the exam looks like.

The form of collaboration is open: B2B, project-based, remote.

Building something with Claude, growing in it, or already certified? Let’s meet for a coffee, online or in Zielona Góra. We will see what we can do together.

Let’s have a coffee

FAQ

Frequently asked questions

Where do we start?

With a short message: what your company does, which systems you use and what you still do by hand. From that I will tell you whether automation makes sense and point out the first step. If you are thinking of replacing subscriptions with your own solution, compare the costs first in the tool cost calculator.

Do you work with companies outside Poland and Germany?

Yes. The conversation can be in English, German or Polish. My hands-on knowledge is strongest in the German and Polish markets: Amazon, eBay and Kaufland in Germany, Allegro in Poland, and systems such as Gambio and Billbee. For other markets I tell you plainly what I know from practice and what I do not.

Where does your help with a German GmbH end?

At operations. I help with marketplace accounts, Billbee, the order of steps, and invoicing and document flow for VAT, intra-EU supplies and OSS, as agreed with your accountant or Steuerberater. I have sold to Germany as a Polish company, through an Einzelunternehmen and today also through a GmbH. I do not give legal or tax advice.

Where do the figures come from?

From my own system, the helpdesk and the sales and advertising accounts of my companies. Every figure has a measurement period and a definition, and estimates are labelled together with their assumptions. I describe the full method in the case studies. I do not publish revenue or margins.

How do you implement, and what stays with the client?

I test changes in a dry run first, then deploy from a tagged version with a rollback ready. Permissions and secrets are kept separate for each system. Code, documentation and access stay with you, so after implementation you can carry on yourself, with someone else or with me.

Do you work on site or remotely?

I am based in Zielona Góra, in western Poland. I run AI implementation and automation projects remotely for companies across Poland, Germany and Austria, in English, German or Polish. We review the processes together on a video call; a visit on site can be arranged.

Contact

Tell me what you are dealing with.

I will assess whether and where I can help, with automation or advice, and point out the first step. If the topic is outside my field, I will say so plainly.

Email: kontakt@marcin-dobrzynski.plOpen in email app
LinkedIn: Marcin Dobrzyński
Reply: usually within 1–2 working days
Languages: EN · DE · PL

I find the places in a company where automation saves time. And I tell you where it does not pay off.

You are launching sales in Germany or through a GmbH

You sell on Amazon, eBay and Kaufland and want to do it better

Gambio, Billbee, Apilo: you want one source of truth for stock and orders

You want AI that does the work, not the slide deck

2011: my own e‑commerce for Austria and Germany

A Polish company, then a German Einzelunternehmen

A GmbH where I am managing director

3 companies, 2 warehouses, 1 system

How it runs today: one system for three companies and two warehouses

The system I built: StockHub

How I implement

Margin controller

Margin-based repricing on eBay and Amazon

ROAS per SKU

Rotation and supplier orders

B2B portal for suppliers

Helpdesk instead of a ticketing system

Phone line with a voice bot (sipgate)

Team goals board

Advertising review

Facebook and Instagram from the catalogue

Auto-listers with content checks

Amazon listing analytics

Gambio websites

Print queue

Stock event log

Example generator result

Why my own tools rather than another ready-made one

Claude Certified Developer – Foundations

Claude Partner Network