Combining administrative databases to improve public procurement performance

Measuring and preventing waste in procurement spending

Public procurement is one of the core functions of the state, providing key inputs into the delivery of critical public services. From medicines to stationary, governments in developing countries spend a sizable share of their tax revenues on purchases that are determinants of the quality and quantity of final services delivered back to taxpayers. Recent advances in information technology allowed many governments to create electronic invoices for transactions across firms, public sectors, and consumers. These new IT systems have also created a wealth of data that was mostly intended to be used for tax purposes as it improves the tax base’s paper trail. However, this data can also unveil bottlenecks in the public purchases. In this project, we study how the government can leverage invoice data to measure and prevent waste in public procurement.

Leveraging electronic invoices to improve public procurement

This project is in partnership with the tax authority of the state of São Paulo, Brazil. Due to the adoption of electronic invoicing across the country, the government has administrative datasets containing transaction-level data on both business-to-business (B2B) and business-to-government (B2G) sales for a number of products, as well as datasets on the universe of its public procurement processes. This data creates a unique opportunity to ask the following questions:

• How do governments fare compared to private buyers when purchasing similar goods?
• If there are relevant differences between B2B and B2G, what are their key determinants?

The main difficulty in using these electronic receipts data is ensuring that when we compare B2B to B2G purchases, we are able to hold constant the precise nature of the item being purchased. The electronic receipts contain a text field in which sellers describe the item being sold, which we rely on to ensure the comparability of the purchases. However, these free text fields cannot be used as is to make these comparisons and so we will develop a range of new Artificial Intelligence methods building on Natural Language Processing models to use this text to compare across purchases of the same item.

With these tools, we will be able to estimate the mark-up in procurement prices across a wide range of goods and throughout the government. These will allow us to diagnose the key sources of government mark-ups and provide recommendations for potential policy measures that can improve performance. This collaboration can result in considerable savings for São Paulo: the state spends USD 3 billion (6.6% of its revenues) annually procuring 300 thousand off-the-shelf items through its online procurement platform alone, and it has not yet systematically leveraged all the data that they currently have to improve public spending.