Croatia's public procurement market exceeds €14.3 billion annually, but complex administration holds back smaller companies. Two FER engineers have developed the AI agent system TOP to level the playing field.
Although Croatia's public procurement market is nominally accessible to all, in reality it is marked by high administrative barriers that favor large players. This sector generates more than €14.3 billion annually, or over 20 percent of domestic GDP. While small and medium-sized enterprises win 61 percent of contracts by number, they receive less than half of the total value. At the same time, thousands of bids are rejected, most often due to formal errors rather than a lack of expertise.
TOP, short for Tender Operating Platform, aims to solve this problem. It is a Croatian AI agent system developed by young engineers from the Faculty of Electrical Engineering and Computing (FER), Jakov Sikirić and Sven Barac. "When I saw how many companies miss out on tenders not because they aren't qualified, but because they didn't even know the tender existed. The situation is such that a very large percentage of tenders attract very few bidders, and often only one," Sikirić told Financije.hr.
Although the regulations are the same for everyone, the actual ability to participate varies drastically. While corporations have departments dedicated solely to tenders, in smaller companies this burden typically falls on an individual who handles it alongside their primary duties. "It's important to first define what it means that public procurement is hard to access. Formally, public procurement is available to anyone who meets the requirements. However, in practice, there's a big 'but'," Barac emphasizes.
The key barrier lies in operational capacity. Someone must continuously scan announcements, study extensive documentation, verify required conditions, gather references, and keep track of numerous deadlines. Barac explains that for small business owners, this entire process often comes down to one person who, alongside their regular work, struggles with bureaucracy, information overload, and a lack of digital tools that would genuinely ease their workload.
The statistics showing that smaller companies win many contracts but of low value do not indicate their lack of competitiveness. They point to difficulties in entering more demanding procedures. Barac explains that higher market value is concentrated in more complex procurements that require more references, financial stability, guarantees, legal advisors, experience, and greater operational resources. Sikirić adds that a company with an excellent product and references can be eliminated before quality is even evaluated if it overlooks a change in documentation, submits the wrong proof, misses a guarantee deadline, or fails to align its bid with the prescribed criteria.
TOP's ambition is not to be just another tender search engine. The platform is designed to guide a company through all phases: from identifying opportunities to analyzing requirements, creating checklists, tracking deadlines, organizing the team, and assessing readiness to apply. Sikirić highlights the difference between a tool that merely delivers information and one that changes the way you work. When the system automatically notifies you of a tender, enters the deadline in your calendar, and allows document preparation to start weeks in advance, it ceases to be occasional help and becomes a business backbone.
Particularly significant is the institutional memory function. The platform stores application history, certificates, references, and CVs. Sikirić considers this perhaps the most important feature for small business owners. While large companies have specialized teams that share knowledge, in small companies, the departure of one knowledgeable person means a permanent loss of that knowledge. A system that remembers all previous activities becomes independent of any individual.
In public procurement, where even the smallest mistake can mean losing a contract, responsibility cannot be delegated to an algorithm. That's why TOP is designed with the "human-in-the-loop" principle. Sikirić emphasizes that this is not a compromise but a conscious decision. Artificial intelligence organizes, alerts, and suggests, but every significant decision remains in the user's hands. Legal and financial consequences cannot be shifted to software, but all the groundwork leading up to a decision can be.
One of AI's key advantages is its tirelessness. Sikirić explains that a machine doesn't care whether it's analyzing the first or fiftieth document; its concentration doesn't wane, nor does it skip items in a hurry. Moreover, if a user asks for the reasoning behind a conclusion 15 times, it will show each time which data it was based on. While the debate about susceptibility to errors is legitimate, Sikirić believes the real question is whether a company has the capacity to devote full attention to every document every time, and the answer is usually no.
Barac views public procurement as a complex data problem encompassing search, natural language processing, information extraction, condition classification, risk assessment, and historical data analysis. That's why TOP is not designed as a simple chatbot but as an agent system. While a generic tool can answer a question about a document, TOP aims to understand what that document means for a specific company in a specific tender and at a specific point in the process. As Barac describes, it's a shift from software that displays information to software that helps run the business.
An important function of the system is helping companies decide early on to skip tenders that aren't right for them. Sikirić notes that companies don't only spend time on applications they submit, but also on those they abandon after hours or days of analysis when they realize it's not worth it. If the system knows a company's references and capabilities, it can flag mismatches with requirements much earlier.
Additionally, the platform offers market analytics. Users can see who dominates with a particular buyer, at what prices contracts are awarded, who regularly appears as competition, and where the same players repeat. Sikirić points out that it's not the same whether a company bids on a tender from a buyer who consistently works with the same supplier or on a procedure with frequent changes in bidders. Analytics thus helps answer questions that previously depended on individual experience: Is there a real chance? Who is the main rival? What prices have been accepted? How open is the buyer to new offers? As Sikirić concludes, data that exists but no one has time to find, connect, and interpret is not equally useful to everyone in practice.
The question arises: can AI truly open up the market, or will it only increase the advantage of large players? Barac admits this is a real danger. If new tools are used only by those who already have organized data and refined processes, technology becomes an amplifier of existing advantages, making large companies even faster and better informed. But if AI is directed at helping companies start structuring references, understanding conditions, and recognizing opportunities, it can reduce the gap. Barac emphasizes that technology itself is neither fair nor unfair; it amplifies the logic it was built on.
TOP is currently in early access and available free of charge. Sikirić explains that this is not a marketing gimmick but a need for feedback that is more valuable to them than any internal thinking. They hold numerous meetings with users to hear what they truly need in their daily operations. Barac concludes that customers don't pay for architecture, models, or code, but for solving a specific problem. And the problem, as Sikirić summarizes, is clear: the public procurement system is chaotic and confusing, and their goal is to make it more transparent and accessible for everyone.
