Buchly Behind the Scenes

Weve done web design for 20 years. But maybe the best learning experience of our lives was building our own product: Buchly.at. A bookkeeping and invoicing SaaS specifically for Austrian freelancers and small businesses, with OCR-based receipt recognition and AI classification. Honest behind-the-scenes look.

Why even bother?

Like many web agencies, we struggled for years with classical bookkeeping software. BMD, RZL etc. are great tools – but built for accounting firms, not for freelancers sorting receipts once a month. We wanted:

• Photograph receipts instead of scanning
• Auto-extracted and pre-classified data
• Smart matching of bank movements with receipts
• One-click monthly package for tax advisors
• Full GDPR, AT hosting, no US transfer

It didnt exist in the desired form. So we built it.

The tech stack

• Backend: Node.js with Fastify, PostgreSQL, Redis for cache and job queue. On Hetzner Cloud (CPX21 web, CPX31 DB) in Nuremberg.
• Frontend: Classic HTML/CSS/Vanilla-JS – no React, no Vue. Deliberately "boring tech".
• OCR: Custom pipeline with AWS Textract as fallback and fine-tuned Mistral model for German receipts.
• AI classification: Claude API with categorised few-shot prompt for chart-of-accounts classification (EKR07).
• Storage: Cloudflare R2 for receipt images (much cheaper than S3, EU region).
• Bank integration: via Klarna Kosma / Tink for PSD2-compliant interfaces to Austrian banks.
• Mailing: Brevo for transactional emails and newsletters.
• Monitoring: Sentry for error tracking, Plausible Analytics.

What we got right

Boring tech. We chose against all modern frameworks – no React/Next.js, no fancy DI system, no microservices. Instead, a Node.js monolith thats still maintainable and performant today. We dont regret these decisions.

Early market focus. AT, freelancers and small businesses with flat-rate taxation. No "also DE", no "also Switzerland". Tight focus let us tailor features to Austrian tax law specifically – including 13a-EStG flat rate, OSS reporting, AT bank integration.

Privacy by design. All data in EU data centres. No US third-country transfer. Sentry self-hosted, analytics without cookies. Brought us more customers than expected – "finally bookkeeping software without US cloud" was the most common positive feedback.

What we got wrong

Scaled too early. With first 200 customers we built multi-region setups, HA databases and read replicas. Could have skipped that for a year. Performance was always sufficient.

Featuritis. In beta we built too many features because individual customers wanted them. Result: a product doing everything a bit, nothing properly. Had to switch off four of them after 9 months.

Overestimated OCR accuracy. Thought AWS Textract would suffice. Reality with crumpled till receipts, faded thermal paper and scanned PDFs was sobering. 6 months own development with fine-tuned models needed to reach 92 % auto-recognition.

Onboarding too complex. First version required 14 clicks to first captured receipt. Free-trial drop-off: 65 %. Today: 3 clicks – drop-off 22 %.

Lessons every SaaS team should know

1. Build a painful solution for yourself first. We wouldnt have built so precisely if we werent the target audience.
2. Say "no" to 80 % of feature requests. Only when 3 independent customers want the same feature do we consider.
3. Customer support is product development. Each ticket is a UX gap.
4. Dont price too early. First version € 9.90/month – way too cheap. Today: € 19 starter, € 39 premium. Conversion didnt worsen.
5. GDPR as sales argument. In AT, DE, CH theres a growing audience explicitly seeking EU hosting.
6. Boring tech wins. 2.5 years, fewer than 10 tech migrations. Hipper-stack competitors had many more.

Whats next

2026 roadmap: native iOS/Android apps for receipt photo capture (currently PWA), deeper integration with AT tax software (BMD, RZL), and a "tax advisor workspace" managing multiple clients.

Bottom line

Buchly was and is our toughest lesson. Web design client projects are simple in many ways – brief, build, launch, done. Building your own product means deciding daily what matters and what doesnt. Whats roadmap and whats a bug. What customers really want versus what they say in the moment. This experience flows back into every client project.