Every construction and manufacturing owner asks this. ERP knows outstanding invoices and due dates, but not that Customer X has 30-day terms yet averages 23 days late, except in July when they pay early before holidays. It may not know Project Y retains 10% until handover: not yet invoiced, but already consuming cash.
We decided to solve this.
We added an AI agent connecting directly to the ERP database and understanding your business context: your data, customers, projects and other dimensions you track or want to track.
You ask a question in natural language. The agent turns it into a database query, retrieves data and answers.
The key is the underlying data.
AI cannot work reliably when information is scattered across spreadsheets, emails and employees’ heads. Worse, it may hallucinate a convincing but wrong answer. We started with the foundation: ERP as the system of record, the company’s single source of truth. Only structured, clean and complete data enables reliable answers.
We provide the infrastructure that lets AI work safely.
What does this look like in practice?
The owner of an 80-person construction company asks: “Will I have enough for next month’s payroll if customers pay with their usual delays?”
The agent reviews receivables, compares each customer’s payment history, accounts for upcoming payroll, contributions, rent and lease payments, and answers based on actual payment behaviour rather than invoice due dates alone.
Today, the question takes two days, three spreadsheets and an experienced finance director holding everything in their head. The agent does it in ten seconds.
Under the hood, it connects to the ERP PostgreSQL database, uses natural language processing to understand the question, builds an SQL query, retrieves data, applies business logic including payment habits, and generates an answer.
This is an agent reading your database and understanding your context, rather than ChatGPT reading the internet.
We are developing it with FERIT, the Faculty of Electrical Engineering, Computer Science and Information Technology in Osijek, within an EU-co-funded research project.
