Intelligent Methodologies for Preparing Reference Prices in Brazilian Public Biddings

Context: This research focuses on building reference prices in Brazilian public tenders, integrating statistical techniques and AI to improve automated decision accuracy and detect issues like overpricing and underpricing. Problem: Traditional methods for calculating reference prices, like arithmetic mean and lowest price, often cause distortions in public tenders. This research evaluates how AI can enhance accuracy and transparency in audits. Solution: The proposed solution is software that combines statistical algorithms with machine learning to calculate reference prices more robustly. The first version uses statistical methods; the second integrates TensorFlow, expanding irregularity detection. IS Theory: Based on Design Science Research, this study develops and evaluates artifacts to solve identified problems, applying scientific techniques to optimize reference price calculations. Method: A quantitative experiment using public tender data compares the software’s performance with traditional methods, focusing on average deviation and machine learning accuracy. Summary of Results: The first software version reduced average deviation to 2.53%, versus 15.58% for traditional methods. The machine learning version achieved 70% accuracy in detecting pricing irregularities, adding an extra verification layer. Contributions and Impacts in the IS area: This research contributes to Information Systems by proposing a system that combines statistics and AI, enhancing accuracy and transparency. The solution improves public bidding efficiency and is adaptable to other government and industry areas.