80% of an ERP for Free — Testing Open Mercato with Claude

TL;DR

Hi, I'm Mateusz Kozłowski and I help companies adopt AI and automation. In this post I'm walking through a case study from a sewing workshop producing eco-friendly bags — after six months in a low-code tool, we rebuilt the whole thing on Open Mercato. Three reasons it paid off:

• 1

  80% out of the box, 20% custom

  Open Mercato shipped sales documents, roles, history, custom attributes and the admin panel for free. We only had to write the pricing logic specific to the workshop.

• 2

  AI-native, not AI bolted-on

  A chat in the bottom-right corner lets you ask about quotes, duplicate them or generate reports in natural language. No mainstream ERP comes close to this today.

• 3

  From 40 minutes to 5–10 (sometimes 1–2) per quote

  Knowledge from 20–30 variables and a dozen spreadsheets now lives in a single, centralized configurator. The end of tribal knowledge.

Video walkthrough

If you'd rather watch than read — I walk through the whole implementation on YouTube. The written version below keeps the key takeaways.

The problem: quoting chaos

The core pain in this workshop — and not just this one — was chaos in quoting. General-purpose spreadsheets, product-specific spreadsheets, quote spreadsheets, helper spreadsheets. Every employee had their own copy, every price change required updating a dozen files, and in practice they were never all up to date at the same time.

The second problem was tribal knowledge. The pricing algorithm relies on 20–30 variables: quantity, print quality, number of colors, print type, number of pattern pieces, add-ons, product margin, regional margin, labor cost. That knowledge lived in the heads of the longest-tenured employees — and was very hard to extract.

Phase one: six months in Retool

We started with serious prototyping and conversations. Over about six months we built a Retool app handling the simplest orders — the ones that ate up around 40% of the quoting effort. We deliberately left the most complex products for later. The goal was a quick reduction in clicking and in mental load on the owner.

Retool was great for phase one — building was fast and the results were tangible. The pain showed up when we hit the complex products: variable count exploded, the code drifted apart, and testing in low-code is dramatically harder. Every fix went through a long "change → click through manually → ship" cycle, and maintenance started eating more time than it saved.

That was the moment I started looking for something with a stronger foundation for the next phases — not to throw out low-code, but to move the core onto a tool that scales properly.

Why not Comarch / Symfonia

Fair question: if low-code wasn't enough, why not reach for an off-the-shelf ERP like Comarch or Symfonia? The common pattern in those systems was a deal-breaker:

• Closed to external integrations. A single system never covers every case we need, and getting custom code in requires a lot of gymnastics.
• Demos are promises, not products. Salespeople say "that's possible." Mid-implementation it turns out it isn't — the feature is missing, it's on the 2027 roadmap, or engineering has other priorities.
• No AI-native mindset. In a world where AI is genuinely accelerating code, a "closed system with a customer portal" is a dead end, not an advantage.

Turning point: Open Mercato

The idea to change direction came from a talk by Tomek Karwatka, where I first heard about Open Mercato — an open framework for building business apps. I decided to check how much of it would actually carry an ugly, real-world case from a sewing workshop.

The biggest surprise was the speed of coding and the AI-first / AI-native approach. The project is built to support a developer working with an LLM — at a level I don't see in any commercial ERP today.

Configuring it for a sewing workshop

I added four tabs tailored to this company — and they're what turns Open Mercato from "yet another tool" into a working pricing configurator:

• 1

  Materials — fabrics

  A fabric dictionary with arbitrary fields: price, color, quality, availability. Every field type in Open Mercato is configurable — you don't have to modify the database schema to add a new attribute.

• 2

  Patterns — pattern pieces

  Pattern pieces split into basic and add-on. Any product can be made of any number of pieces, and each piece has its own fabric, color and print.

• 3

  Addons — extras

  Zippers, straps, patches, labels — anything attachable to a product. Default variants are wired in automatically when a specific product is picked.

• 4

  Pricing Rules

  Quantity ranges (from–to), region, margin, labor cost by product. This is where the knowledge that used to live in spreadsheets and people's heads now lives.

The result: knowledge is centralized in one place. A fabric price change propagates to all new quotes automatically, and a new hire can learn on a working tool rather than on someone else's spreadsheets.

Quoting, step by step

Up to the Pricing tab the quote view looks identical to vanilla Open Mercato — sales document, customer, channel, currency, custom attributes. The fun starts when you add a line item:

1. Pick a product from the category tree — the system suggests dimensions and required pattern pieces (e.g. front + back).
2. Configure each piece separately: fabric, color, optional print with a number of colors.
3. Enter product dimensions, quantity and cut layout (e.g. 6×3) — that last one generates real material savings.
4. Automatic price calculation across all variables: margin, labor, region, add-ons, quantity rules.
5. Save the line item, optionally duplicate for a client variant, view the full breakdown inside the quote.

The strongest part is that history, versioning, comments, corrections and quote-to-order conversion are already built into Open Mercato. Writing that from scratch is months of work in a simple CRM — here you get it from day one.

AI agent — talking to the system

In the bottom-right corner of Open Mercato there's a chat — and in my view it's the biggest game-changer in the ERP segment. Instead of clicking through a dozen screens to find quotes by Mateusz Kozłowski, you just ask: "show me all quotes by Mateusz Kozłowski." The system pulls the data through MCP, shows a list with order-number suffixes, and lets you keep working with it.

During testing I asked it to duplicate a specific quote. The agent fetched the full details, confirmed intent ("create a duplicate?") and ran the operation. This isn't a gimmick — it's a different working model than clicking through menus.

Architecture and core protection

Underneath, Open Mercato has a very mature structure. The repository contains a dedicated .ai folder with instructions, boundaries, schemas and an architecture description. This gives the language model a "map" it sticks to — it doesn't hallucinate files, doesn't overwrite core, doesn't invent packages that don't exist.

What usually breaks attempts to "bolt features onto a packaged ERP" is designed in from the start here: you extend the system in your own modules, and the core stays untouched. In practice that means you can pull upstream Open Mercato updates without losing your changes.

Results in numbers

Wrap-up

Open Mercato gave us 80% of the system in the first week, and the remaining 20% — the workshop-specific logic — we wrote in code with heavy AI assistance. It's now my default recommendation for companies outgrowing spreadsheets and low-code that don't want to fall into the closed enterprise-ERP trap.

If this fits your company — make-to-order production, many variables in pricing, quoting knowledge living in a few heads — let's talk about whether Open Mercato is the right fit.

Mateusz Kozłowski

Założyciel flowbiz · Ekspert automatyzacji procesów

Wdrażam automatyzacje, integracje i AI w średnich firmach na Pomorzu i w Kujawsko-Pomorskiem.

LinkedInWięcej o autorzeUmów konsultację