AI in public procurement: between experimentation and impact

Governments around the world are starting to use AI in procurement, but not all at the same pace. Some are still exploring. Others are running pilots. A smaller group already has tools in production with real impact.

The opportunity is clear. AI has the potential to help governments manage procurement better, make processes more efficient, automate repetitive work, and reduce costs. That can lead to better decisions, better services, and more transparent markets.

But going beyond the hype is harder. In this post we’ve rounded up a selection of examples of how AI is currently being applied in practice to improve public procurement, sourced from our team and global community of open contracting innovators. They can be grouped into four use cases (roughly in order from most to least mature): improving data processing, automating administrative processes, improving oversight, and market intelligence.

1. Better and more accessible data

One of the biggest challenges in procurement is not the lack of data, but how usable it is. A lot of important information is still stuck in PDFs, unstructured formats, or disconnected systems. This is where AI is starting to help. It can extract information from documents, classify data, and make it searchable using natural language. It does not replace existing systems, but it makes them much more useful. As OCP’s James McKinney notes, AI doesn’t remove the need for structured data, it simply makes structuring data faster and cheaper.

In Chile the public procurement agency ChileCompra uses AI to process PDF documents and extract key information for monitoring procurement processes, reducing a large amount of manual work. Beyond extraction, AI is also being applied to classify procurement data at scale. 

In Ukraine, where OCP has collaborated across multiple procurement reform initiatives, the e-procurement system Prozorro uses machine learning to classify procurement items using CPV codes (a common classification system for products) based on written information, such as the tender and item descriptions, streamlining item classification. This increases classification accuracy, enables analysis of sector-specific market dynamics, and helps more vendors identify relevant tenders, ultimately enhancing competition and value for money. 

In New South Wales (Australia), the Department of Procurement uses similar tools to automatically classify procurement spending across large datasets, enabling deeper analysis and better insight into government spending. 

AI is also improving how data is accessed and interpreted. In the Dominican Republic, the General Directorate of Public Procurement has developed a chatbot that enables suppliers, buyers, and citizens to access real-time information about the procurement system. Users can query procurement processes, review the history of awarded contracts by purchasing units, download tender documents, and consult the catalog of goods and services, reducing the need for technical expertise to navigate the data. In North Carolina (US), another chatbot helps public officials from the Department of Information Technology navigate IT procurement processes, answering questions and providing quick access to forms and guidelines.

2. Automation and operational efficiency

Procurement involves a lot of administrative work, especially when it comes to preparing tendering documents and reviewing contracting documents for compliance purposes. AI is starting to take on some of that work. It can help draft documents, generate requests for proposals (RFPs), and review contracts. This saves time and allows teams to focus on more strategic tasks. 

Cities like Tempe and Murray City in the US are using language models to draft RFPs. In San Antonio, a pilot tool analyzes contracts, extracts key data like renewal dates and payment terms, and checks alignment with internal policies.

In Finland, Palkeet (Financial and Human Resources Service Centre for the Finnish Government) has deployed bots combining Robotic Process Automation (RPA) and machine learning to automate procurement-related administrative processes. One bot validates incoming invoices for correct formatting and data completeness, generating an error report with the percentage of valid invoices. A second bot maintains the supplier register, adding new suppliers upon contract signing and updating existing records when relevant changes occur.

The South Korea Fair Trade Commission (KFTC) offers one of the most advanced cases, having developed the Bid-Rigging Indicator Analysis System (BRIAS). Using data from over 720 public entities, approximately 60,000 procurement cases are evaluated per year. To detect collusion, BRIAS applies sophisticated AI models that identify bidding patterns indicative of collusion risk, and this information is then used to prioritize and guide deeper investigations. The system has proven highly cost-effective, generating fines nearly 40 times higher than its maintenance costs over seven years.

3. Monitoring, analysis, and risk detection

Oversight is one of the hardest parts of procurement. There is a lot of data, and it is not always clear where the risks are. AI can help by identifying patterns and anomalies that are hard to detect manually. This makes monitoring more proactive and targeted. In Brazil, the system ALICE analyzes procurement processes daily and generates alerts when it detects risks. In 2023, the system reviewed nearly 191,000 processes and triggered audits linked to contracts worth over US$ 4.5 billion. In Portugal, the Court of Auditors uses data-driven tools to detect issues like overpriced contracts or bid rigging. In the UK, the SNAP platform connects data across government agencies to detect fraud using network analysis and AI models. In the US, the General Services Administration uses AI to create dynamic risk profiles for suppliers, allowing continuous monitoring by procurement officers instead of static assessments.

4. Market intelligence and decision support

This is one of the most interesting areas, but also one of the least developed. Governments often struggle to understand their markets: who the suppliers are, what prices should look like, and how demand behaves. AI can help fill those gaps. 

In South Korea, procuring entities use AI to predict demand, recommend opportunities to suppliers, and even anticipate congestion in bidding processes – one of the most sophisticated examples of AI being used to understand market dynamics rather than simply automate tasks.

AI is also being used to lower barriers to participation. In the US, the Department of Defense uses AI to match small businesses with procurement opportunities, making it easier for them to participate. In Paraguay, a similar tool developed with OCP support identifies tenders that are likely to be suitable for SMEs based on past data.

In South Africa, supplier evaluation AI models include socioeconomic criteria, helping governments balance inclusion goals with technical requirements.

What is working

The cases that work best have a few things in common. They focus on clear problems, use reasonably good data, and do not try to do everything at once. AI works best as a support tool, embedded in day-to-day tasks, not as a full replacement of systems. In short: outcomes depend just as much on how the problem is defined as on the technology used to solve it.

Risks and limitations

There are also clear risks. Data quality is still a major issue. If the data is poor, AI will not fix it. In many cases, it can make things worse. Training a model on low-quality or biased data leads to unintended outcomes. For example, a contract opportunity recommendation algorithm may disproportionately under-recommend opportunities to women-led businesses if the training data lacked adequate representation, effectively perpetuating existing inequalities rather than addressing them. 

There are also risks around automating sensitive decisions, like awarding contracts. These raise questions about transparency, bias, and accountability. And many projects never go beyond the pilot stage, which limits their real impact. This is often due to institutional and regulatory constraints, limited capacity to integrate these tools into everyday operations, and persistent uncertainty about how to embed them in decision-making. In addition, weak development outcomes and insufficient data infrastructure can prevent these initiatives from delivering value.

An incremental transformation

As in other domains, AI is entering public procurement gradually and there are still a lot of unknowns. Early experiments, pilot projects, and a smaller number of cases with real impact are all happening at the same time. The cases that work best rely on good data and keep humans in the loop. AI supports decision-making, but does not replace it.

As data improves and systems mature, these use cases are expected to become more accurate and more useful. The change is gradual, but that doesn’t make it insignificant. Over time, AI is likely to fundamentally reshape how governments plan, run, and oversee procurement.