In the world of public bidding, where every variable can affect the final price, predictive models offer a powerful tool for making more informed, confident, and strategic decisions. For companies, having an estimate of the expected price allows them to submit more competitive bids with less uncertainty. For public administration, these models can indicate whether a tender is significantly above or below a reasonable range, thus promoting efficiency and transparency in the use of public resources. In this article, we explore how a predictive model based on data from the Inter-American Development Bank (IDB) has revealed key insights to improve this process, even with data limitations.
The input dataset
To build the model, we used a dataset of 3,641 rows and 12 variables, each related to IDB contracts. These variables included crucial data such as the supplier's country, the type of service contracted, and the selection process. Before beginning the modeling, we performed a data enrichment process, i.e., we enriched the initial data by extracting additional features from the contract dates and descriptions. This step allowed us to obtain details such as the day, month, and year of the contract, as well as information extracted from textual descriptions, which enriched the initial database and added more context to the variables.
The Modeling Process
For model development, we tested both CatBoost Regressor and Linear Regression, two popular algorithms for predicting numerical values. While Linear Regression is a simpler and more traditional approach, CatBoost is a decision tree-based algorithm that typically handles datasets with many categorical variables better.
During the process, we tuned parameters and optimized validation techniques using cross-validation with five partitions to ensure the reliability of the results. Although both models were tested, the CatBoost Regressor emerged as the best, with a Mean Absolute Error (MAE) of 30,374. Despite a considerable margin of error, the process allowed us to identify the most influential variables in determining contract prices.
Key Insights from the Model
The data analysis, although limited by the number of observations and the distribution of contract values, yielded key insights about the variables that most influence contract prices.
- Consulting, the Predominant Service: More than 60% of IDB contracts are for consulting services, underscoring the high demand for specialized knowledge in international projects. This suggests that technical capacity and specialized advice are essential in most IDB tenders.
- Spain, Leader in Competitive Procurement: Spain stood out as the country with the highest number of competitive procurements. This fact highlights the strong presence of Spanish companies in IDB projects and their ability to offer competitive proposals.
- The highest-value contracts are linked to requests for proposals (RFPs): Contracts requiring a Request for Proposal (RFP) have significantly higher average bid amounts, reaching USD 210,000. This suggests that larger, more complex projects typically employ this selection process, where detailed planning and competition among suppliers are key.
- Focus on sector expertise: The IDB's Vice Presidency for Sectors and Knowledge leads in the number of contracts, reflecting a clear priority to strengthen sector expertise within the organization. This strategic approach to contracting demonstrates an effort to consolidate specialized knowledge in key development areas.
Conclusions
In conclusion, the development of the predictive model, although limited in accuracy, was a valuable tool for better understanding the factors that influence the price of public bidding. The insights obtained not only help improve the design of future tenders but also provide organizations—both public and private—with the ability to make more strategic decisions, reduce uncertainty, and align price expectations with market reality. This study demonstrates the potential of predictive models to optimize procurement processes and promote more efficient and transparent decisions.
However, there is room for improvement: expanding the dataset, increasing geographic diversity, and incorporating external variables such as inflation or exchange rates could reduce model error and further enhance its value for decision-making.
If your company is looking to optimize its bidding processes or improve decision-making through predictive models, at Mottum we can help. Our team of data intelligence experts is ready to create customized solutions that boost the efficiency and transparency of your projects. Contact us now!