n8n

My main use cases for n8n  involve workflows I built. The first is a genfin mapping workflow, which takes trial balance data and uses GPT-4 with structured output parsing to automatically classify accounts into the correct financial statement line items. A second workflow is the basis of preparation workflow, which ingests PDFs, extracts financial context, and then generates a complete basis of preparation document using an LLM. That one involves webhook triggers and custom JavaScript nodes for output fixing.

The best features n8n  offers that stand out to me are three things. First, the visual workflow builder allows me to see the entire data flow and orchestration logic graphically, making it much easier to debug and iterate, especially when working with complex multi-step processes. Second, the JavaScript code node provides full flexibility for custom data transformations, which most low-code platforms simply do not offer. I can write logic to fix JSON parsing issues, merge outputs, and handle edge cases without breaking out to a separate service. Third, the LLM integration nodes allow me to drop in GPT-4 API calls directly into the workflow and chain them with other operations, which is genuinely powerful.

One thing worth mentioning about my main use cases and the workflows I built with n8n is the real impact it had on our client side. We are automating tasks that chartered accountants did manually, such as trial balance mapping and financial statement preparation. We see roughly a 70% reduction in manual processing time on the document pipeline side, which means our clients can focus on higher-value audit and compliance work rather than data entry. That is where n8n really shines for us. It is not just about automating something technical; it is about delivering tangible business value to financial services.

n8n can be improved, and beyond LLM output validation, there are a few other pain points I notice. First, the debug experience could be stronger. When a workflow fails partway through a complex multi-step process, it is sometimes hard to pinpoint exactly where and why, requiring manual stepping through each node.

A couple of things come to mind regarding needed improvements. Documentation around advanced use cases, especially combining JavaScript nodes with LLM integrations, could be more comprehensive. There are patterns we figured out through trial and error that are not well-documented, which slows down new team members. On integrations, the PDF handling could be smoother. We use third-party nodes for PDF integration and had to write custom JavaScript to accurately extract structured data from PDFs. A native, more robust PDF processing node would have saved significant development time for anyone building document automation workflows.

I have been using n8n for approximately one and a half years at Radian Services, where I work as a full-stack developer in an AI-focused role.

n8n is quite stable in my experience. It has been reliable in production.

n8n's scalability has been decent as we scale up workflows over time with increasing client volume. We can handle hundreds of documents per day through the trial balance and disclosure note workflows without major issues. The limitations we start to see are with very large batch processing. When we try to process thousands of documents in parallel or handle deeply nested loops with complex LLM calls, the system gets sluggish.

Customer support has been more reliant on documentation, community, and direct support. n8n has a fairly active Discord community where I can ask questions and get responses quite quickly from both the team and other users who have built similar workflows.

Before n8n, we were building custom Python and FastAPI microservices to handle document processing workflows. It worked, but it required full backend development cycles. We would write code, containerize it, deploy it, and maintain it; every workflow change meant touching the codebase. With n8n, we can iterate much faster. Business logic changes that would have taken days of deployment and testing can now be done in hours through the visual workflow builder. We do not completely abandon custom code; we still use Python for heavy lifting such as OCR and LLM fine-tuning, but n8n becomes our orchestration and integration layer instead of building everything from scratch.

Since we self-hosted, we avoided n8n's SaaS pricing entirely. Our costs are primarily on the Hostinger  VPS subscription, which is quite reasonable at around fifteen to twenty dollars per month, plus the time investment in setting up Docker , configuring persistent storage, and managing the infrastructure ourselves. The licensing side is straightforward because the self-hosted version is open source. We do not have per-user seat costs or tiered pricing to negotiate. That makes it very cost-effective.

I have seen a return on investment, and the clearest metric was the 70% reduction in manual processing time I mentioned, which translates directly to our clients handling roughly three times the document volume with the same team size. For a chartered accountant firm, processing financial documents at that scale is substantial.

Since we self-hosted, we avoided n8n's SaaS pricing entirely. Our costs are primarily on the Hostinger  VPS subscription, which is quite reasonable at around fifteen to twenty dollars per month, plus the time investment in setting up Docker , configuring persistent storage, and managing the infrastructure ourselves. The licensing side is straightforward because the self-hosted version is open source. We do not have per-user seat costs or tiered pricing to negotiate. That makes it very cost-effective.

Before choosing n8n, we evaluated a few alternatives. We looked at Make , formerly Zapier , because it is popular and has good LLM integrations, but it felt more suited to simple automation rather than complex workflows with custom logic. We briefly looked at Airflow  for orchestration, but that is really designed for data pipeline scheduling rather than real-time workflow automation with webhook triggers and immediate responses. There were other low-code platforms as well, but most of them did not provide the JavaScript flexibility to write custom fixes for LLM output issues.

If you are looking into using n8n, I would recommend this: if you have developers in your team who are comfortable with JavaScript and APIs, n8n is genuinely worth evaluating. It is particularly strong for AI automation workflows where you need to orchestrate LLMs with custom logic. Start with a proof of concept on a specific use case rather than trying to replace your entire automation infrastructure at once, and budget time to learn the JavaScript code node properly, as that is where most of the power comes from. Finally, be realistic about self-hosting, as it gives you cost savings and control. I would rate this product an 8 out of 10.