# How I Replaced Spreadsheet Chaos With an AI-Powered Operating System for My Business Author: Charles Haworth · Published: 2026-04-09 · URL: https://charleshaworth.com/how-i-replaced-spreadsheet-chaos-with-an-ai-powered-operating-system-for-my-business/ · Category: AI & Systems I replaced company budget spreadsheets with Postgres, NocoDB, Metabase and Claude on a roughly €10-a-month self-hosted server, built in a few hours. Most businesses do not have a software problem. They have an operations problem disguised as a spreadsheet problem. What starts as a few useful Excel files slowly turns into a mess of duplicated data, manual reconciliations, broken reporting, and expensive software layered on top to patch the cracks. I have seen this pattern too many times, and today I found myself fixing it properly for my own business. In a few hours, I built a much better system for managing our company budget, financial operations, and reporting, with Claude helping me turn a self-hosted stack into something far more powerful than the sum of its parts. What struck me was not just that it worked. It was how obvious it suddenly felt that this is where business operations are heading. ## The real problem is not Excel Let me be clear: Excel is brilliant. I still use it all the time, and I am not one of those people who thinks spreadsheets are obsolete. They are incredibly useful. But they are the wrong foundation once your business data becomes interconnected. That was the issue I was dealing with. We needed to manage the full company budget, from HQ costs to project-specific DEVEX, across six projects. We also wanted proper visibility on suppliers, personnel, invoices, accounts payable, actuals reconciliation, and FX updates. Trying to run that through separate Excel files would have meant exactly the kind of chaos most growing businesses know too well: * multiple versions of the truth * manual reconciliation across files * poor visibility across projects and countries * slow board reporting * weak auditability * too much time spent moving data around instead of understanding it If you want to analyse spend by supplier across projects, compare personnel costs by country, or update assumptions cleanly across the whole business, spreadsheets become fragile very quickly. That is the point where you do not need another workbook. You need a proper system. ## What I built instead I self-host on a Hetzner VPS that costs about €10 per month. On that server I run: * **Postgres** as the core database * **NocoDB** as the spreadsheet-like interface on top of it * **Metabase** for dashboards and reporting * **Coolify** to manage the self-hosted software services On my laptop I have **Claude Desktop**, connected through MCP servers, to work across the whole stack This matters because each layer plays a different role. Postgres gives me a structured source of truth. NocoDB makes the database easy to work with in a very human way. Metabase gives me management reporting, dashboards, and board visibility. Claude becomes the intelligent layer that can help design, shape, update, and interrogate the whole system. That is the key point. AI is most powerful when it is not trapped in a chat window, but connected to the systems where the real work happens. ## Why Claude changed the game I already had the infrastructure in place, so I was not starting from zero. The server was running, the database existed, and the tools were connected. But the actual budgeting and reporting system came together in around three hours. That is the part I still find remarkable. Claude helped me: * think through the table structure * design account codes and the underlying logic * take a messy Excel file and translate it into the right schema * populate and update the database directly * build a full Metabase reporting layer from prompts * create dashboards for board reporting and CFO-level oversight * implement drill-down filters so I could move from summary to detail instantly The result was not just faster execution. Claude also surfaced useful reporting ideas I had not explicitly thought of. I could describe the type of visibility I wanted, and it built dashboards that were immediately useful. That is a big shift. A few years ago, doing this properly would usually have meant one of four things: more spreadsheets, more SaaS tools, outside consultants, or internal developers. Now, if you control your own data and give AI access to the right systems, the distance between an idea and a working internal tool becomes dramatically shorter. ## Why owning the data matters The deeper lesson here is not really about dashboards. It is about ownership and structure. Once your data is organised properly in a database you control, you can start asking much better questions. I can now analyse costs by supplier, by person, by category, by project, by country, by currency, and by time period. I can track budgets, actual transactions, invoices, accounts payable, and reconciliation in the same environment. That changes the quality of decision-making. This is no longer just a budgeting tool. It is the beginning of an internal operating system for the business. And because the system is ours, we can shape it around the way we actually work, instead of contorting ourselves around someone else’s SaaS product and pricing model. That is strategically important. ## Why I think this matters for a lot of businesses I am not arguing that every SaaS tool is bad. Many are excellent. But think about what I just described. A full budgeting and reporting system across six projects, multiple currencies, and all business operations, built in a few hours for about €10 a month in infrastructure. That same scope would easily be a five-figure annual SaaS commitment, or weeks of consultant time, or a permanent spreadsheet mess that nobody trusts before board meetings. A lot of businesses have quietly accepted that trade because building something better used to feel out of reach. That assumption is no longer safe. If your business data lives in a structured database you control, AI can now help you build, maintain, analyse, and report on your operations at a fraction of the usual cost. That is not a marginal improvement. That is a different operating model. ## The bottom line What I built today reinforced something I have been feeling for a while. The future of business operations is not just buying more software. It is owning your data, structuring it properly, and using AI to help you work with it. For a very low monthly infrastructure cost, plus my Claude subscription, I now have a system that can handle budgeting, actuals, reconciliation, invoice tracking, accounts payable, FX updates, and reporting across HQ and six projects. Not long ago, that would have meant either a spreadsheet jungle or a very expensive SaaS stack. Now it is something a lean business can build and control for itself. I think companies that understand this early will have a real advantage. The question is no longer whether businesses can build their own internal operating systems. The question is how long you can afford not to. Every month you spend reconciling spreadsheets or paying for tools that do not talk to each other is a month your competitor might be spending building something they actually control. If you are thinking about making this shift, I would genuinely like to hear what is holding you back. Reply or leave a comment. I will answer. ## Resources if you want to build something similar If you want to explore this kind of setup yourself, these are the main tools I used: * [Hetzner VPS](https://hetzner.cloud/?ref=G208eZgLRt6P) * [Coolify](https://coolify.io/) * [Postgres](https://supabase.com/) * [NocoDB](https://nocodb.com/) * [Metabase](https://www.metabase.com/) * [Claude Desktop](https://claude.ai/) Helpful setup videos: [Watch on YouTube: How to Self Host Your Free Automation Stack In 30 Minutes](https://www.youtube.com/watch?v=lN67hrwj9wc) [Watch on YouTube: Self Host 101 - Set up and Secure Your Own Server](https://www.youtube.com/watch?v=Q1Y_g0wMwww)