The Use of Artificial Intelligence to Calculate the Estimate of a Public Procurement Act ^†

Abstract

Public procurement refers to the purchasing of goods and services for public entities. Before launching the call for tender, the public body prepares an estimate of the procurement act; this estimate is taken into consideration by the tender commission before awarding the contract. Through technological innovation, buyers can now rely on new solutions as a support to improve the way of calculating the estimate. In this paper, we present research that has been performed in this field, to produce different AI solutions that can be used by buyers to make the estimate more accurate.

1. Introduction

Public procurement is an important business function of government [1]; it is a field of public administration that deals with the government’s procurement of goods, services, and works. These include supplies needed for both the daily running of government services and operations (fuel, stationery, airline tickets, cars, etc.) and investment projects (school buildings, roads, ports, technology, etc.) [2].

Public procurement is an essential element of the economy for the country; it represents between 15% and 17% of national GDP [3], and in Europe, it represents around 14% of GDP, which is about EUR 1.9 trillion [4], and it is also considered as a tool used by governments to establish their policies [5] and to create economic development for the country.

The process for public procurement is managed by regulations, which set the conditions and forms in which work, supply, and service contracts are launched by public bodies and awarded to suppliers [6].

In the management of public procurement, public bodies should take into consideration both economic and legal dimensions and manage public procurement effectively to be able to deal with various external constraints (scarcity of resources, wars, pandemics, budgetary restrictions, and other external factors) [7].

Like in the private sector, procurement in the public sector is carried out according to a process that starts with the reception of the request of purchasing and ends with the execution of the procurement act [8].

In the process of public procurement, there are some repetitive tasks that are still performed manually, such as the redaction of specifications or the evaluation of bid suppliers, which has a detrimental effect on the process’ overall efficiency [9].

All the parties involved in public procurement generate data at every stage of the process; this data may be valuable and utilized as a tool to assist in making decisions and improve the efficiency of public procurement [10].

In the field of public procurement, there are numerous issues with contract execution, such as late execution, supplier failure, and requests for price revision, as a direct result of non-control of prices in the submission phase [11].

In the sphere of public procurement, corruption can arise at any stage of the procedure, which tends to interfere with market mechanisms and negatively impact the advancement of the economy for the country [12].

In the context of tenders, the contracts concluded between the public body and suppliers are subject to information asymmetries (the company knows its costs and the economic environment better than the public party), the difficulty to foresee all the events that may occur during the execution of the contract [13], and fluctuations in prices, which complicate the task for buyers.

Nowadays, artificial intelligence offers powerful solutions that could be used in the field of procurement [14]; in fact, governments worldwide are starting to use artificial intelligence in the public procurement process, as a means of enhancing the performance of the public sector, owing to the benefits that this technology offers [15].

This paper aims to highlight the benefits of artificial intelligence for public procurement in general, with a focus on the AI solutions that could be used to make the estimate of the procurement act more accurate.

In this paper, we present the solutions of artificial intelligence that have been the subject of scientific publications from the year 2000 to 2024. To start, we will give a definition of public procurement and artificial intelligence to be able to look at the intersection between these two fields.

2. Definitions

Public procurement: Public procurement means buying works, services, or goods for the benefit of a public body, and it has the meaning of “purchasing” used in the private sector; in the public sector, the term of “purchasing” is often replaced by “procurement” [16], and it is an essential mechanism for the government for allocating funds for the procurement of goods, services, and works used for projects and programs [17] necessary for public services such health, education, roads and highways, public transport, defense, and waste management [18].

Public body: Public institutions (core government, local authorities, public agencies, and public enterprises) that deliver public programs, services, or goods [19].

Decree n°2-22-431: The decree that came into effect on 1 September 2023; it determines the conditions and methods of awarding contracts for works, supplies, and services on behalf of public institutions in Morocco [19].

Artificial intelligence: Artificial intelligence (AI) is the engineering and science field that aims to develop solutions in which human behaviors and characteristics are imitated, such the capacity to learn from past experiences [20]; it is also considered as an integration of physiology intelligence and computer science in simple language [21].

Artificial intelligence could be applied in different fields; in this paper, we review its application in calculating the estimate of the public procurement act.

In this paper, artificial intelligence will be replaced by AI.

3. Problematic

One of the most important issues facing public administration is the effectiveness of public procurement [22], which is directly linked to the financial aspects of the procurement act.

The public procurement process starts with identifying the inputs required to carry out public programs and projects, classifying these inputs into furniture, works, and services, and then elaborating on the specifications, publishing the call for tenders, elaborating on the estimate, awarding the call for tenders, and finally executing the contracts [23].

The value of the award price depends on the value of the estimate of the call for tenders; therefore, having a more accurate estimate leads directly to having a correct award price.

By improving the way of calculating the estimates, the publics bodies would be able to have enormous savings; in fact, 1% of gain could lead to approximatively EUR 20 billion of savings [24].

4. Methodology of Research

To investigate our subject, research was performed on Scopus and Web of Science, and 135 articles were found between 2000 and 2024 by using artificial intelligence and public procurement as keywords.

At the beginning, we found 135 articles related to these keywords, and by reading the summary, the introduction, and the results for all those articles, we reduced the number to 43 articles dealing directly with the use of artificial intelligence in editing the estimate of a public procurement act; those articles have been read and analyzed to come to the results mentioned at the end of this paper.

The following Figure 1 gives a description of the conceptual framework of research adopted in our paper.

The following

Table 1. Main ideas of articles identified in the research scope.

Document Title/Year of Publication	Authors	Year Main Idea of the Article
Award Price Estimator for Public Procurement Auctions Using Machine Learning Algorithms: Case Study with Tenders from Spain. 2023	Rodriguez, MJG;	2023
	Montequin, VR, Ubierna, AA, Hermida, RS, Araujo, BS, Jauregi, AZ	Through a comparison between random forest and linear regression, with isotonic regression and popular artificial neural network models, the author explains the efficiency of each model and comes up with a model capable of delivering the awarding price with more accuracy.
Conceptual estimation of construction duration and cost of public highway projects. 2022	Mohamed Basma, Moselhi, Osama	This study assesses a typical set of machine learning methods’ applicability to this task. The less studied paradigms, including isotonic regression and well-known artificial neural network models, are contrasted with the conventional regression techniques, like random forest and linear regression. Based on the Spanish public procurement announcement (tender) dataset, numerous tests are carried out using a variety of error measures and WEKA and Tensorflow 2 implementations.
Budget-feasible Procurement Mechanisms in Two-sided Markets. 2018	Wu, WW (Wu, Weiwei); Liu, X (Liu, Xiang); Li, MM (Li, Minming)	This study examines the mechanism design problem. Every seller is permitted to bid the price of their private commodity, if it has public worth. Buyers may present their own budgets, not always the accurate ones. The objective is to find financially feasible solutions that guarantee that sellers receive sufficient payment, and that purchasers’ budgets are not surpassed. The authors principally contribute a random method that ensures several required theoretical guarantees, including budget feasibility, simultaneous veracity on the part of sellers and buyers, and continuous approximation to the best possible overall procured value of purchasers.
Predicting costs of local public bus transport services through machine learning methods	Amicosante, Andrea, Avenali Alessandro, D’Alfonso Tiziana, Giagnorio Mirko, Manno Andrea, Matteucci Giorgio	The study generates a model based on a machine learning system capable of predicting expenses for public bus transportation. To train the algorithms, a built-in dataset from 269 transportation providers offering urban services in the US between 2015 and 2019 was used. The model proposed could give various insights, such identifying the key factors of transportation cost, which lead to an improvement in service contract management.
Model of Predicting Bidding Costs for Construction Projects in Nigeria using Public Procurement Act. 2007	Mohammed Lawal Yahaya; Isma’il Umar; A. J Babalola; Mohammed Sani	The aim objective of this article is to create a model capable of predicting the cost of bids related to construction projects. The study concludes that, on average, contractors’ transaction costs when bidding on construction projects amount to 8.21% of the contract sum. This information will be useful to new companies entering the market as bidders because it will let them know what to expect in terms of entry costs for public projects.
Reliable procurement cost estimation methodologies: A case study in complex environments. 2004	Perluka, B.Y	The process of awarding the call for tenders is based on the price of the bid; therefore, knowing how to estimate the project is important for both suppliers to be competitive, and for public agencies to avoid a budget overrun. Establishing the estimate of a project is a complicated task regarding the multiple factors that are involved, such the appearance of new product development. In this article, the author aims to presents methodologies that could be useful for calculating the estimate and making it more reliable.
Price Estimation Model Using Factor Analysis in Procurement. 2022	Achmad Faizal, Zulkarnain Zulkarnain, Isti Surjandari, Authors Info and Claims	Managing procurement, which is essential for cost-saving, must involve bargaining with suppliers to acquire the best pricing when acquiring goods and services. A buyer’s price estimate is created as part of the negotiation process. Procurement professionals disagree on the aspects that go into pricing estimation. The goal of this study is to identify the key variables that influence price estimation in procurements and to develop a model based on those variables, particularly when it comes to leasing assets.
The Accuracy of Independent Estimates of the Procurement Costs of Major Systems. 2005	David L. McNicol, Project Leader Karen W, Tyson, John R. Hiller, Harely A, Could, Joshua A, Minix	The results of an evaluation of the reliability of the independent cost estimates for procurement that the Department of Defense employs for significant acquisition program milestone reviews are presented in this study. Initially, the IDA investigated sixty-three significant programs that were authorized to start Engineering and Manufacturing Development between 1985 and 1998. For just 25 of these programs could the information required to assess the precision of the independent procurement cost estimates be found. For eighteen of the twenty-five examples, the IDA determined that the independent cost estimate was fairly accurate. In the remaining cases, the independent estimate was sometimes much too high and sometimes substantially too low.
Optimized artificial intelligence models for predicting project award price. 2015	Chou, Jui-Sheng, Lin, Chih-Wei, Anh-Duc Pham, Shao, Ji-Yao.	This paper aims to estimate bid award amounts for bridge construction projects by using artificial intelligence algorithms such multiple regression analysis, artificial neural networks (ANNs), and case-based reasoning (OR). Information was gathered for public bridge building projects from the Taiwanese government’s e-procurement system. The study shows that the mathematical model for artificial neural networks (ANNs) offers more dependable simulations and has a better fit.

gives an overview of the most relevant articles related directly to the use of artificial intelligence in calculating the estimate of the procurement act.

5. Results

By analyzing the articles mentioned in the table, we come up with the following results:

-

The efficiency of a public procurement act is directly related to the estimate of this procurement act, since the bids of competitors depend on the value of the estimate.

-

Artificial intelligence plays a key role in establishing an accurate estimate.

-

There are different artificial intelligence solutions that could be used to enhance efficiency in public procurement; machine learning, is one of these tools used to establish an accurate estimate.

-

There are different types of machine learning, and the ones that are mentioned in the table above are as follows: linear regression, random forest, and artificial neural networks (ANNs).

For further analysis of the results found, we will start first with a definition of Machine Learning; then, we will give more details about all the subfields of machine learning to come up with a conclusion about the most suitable one for an accurate estimate.

5.1. Machine Learning

Machine learning is an field of artificial intelligence, which aims to make algorithms that upgrade their performance by learning from previous experiences [24], therefore progressively increasing its precision.

In procurement, the use of machine learning to establish the estimate requires a substantial quantity of reliable data from previous call for tenders to adjust them and use them to predict the estimate.

5.1.1. Random Forest Algorithm

A random forest algorithm is a set of tree predictors in which each tree is reliant on the values of a randomly selected vector that is distributed uniformly across all the trees in the forest [25].

The ensemble’s prediction is calculated by taking the average of each model’s projections. An ensemble approach that lessens the bias of individual models and yields a more adaptable predictor that is less likely to overfit is exemplified by this example.

5.1.2. Linear Regression

Linear regression is an ML technique that makes a correlation between input and output data through the following equation:

$$Y=\sum _{i=1}^n\left(Bi Xi+\alpha \right)$$

where Bi are the coefficients that measure the value of the input data and α is a constant value.

5.1.3. Neutral Network

Neutral networks or artificial neutral networks (ANNs) are a subfield of machine learning; they are made through node layers, which include an output layer, an input layer, and one or more hidden levels. Every node, or artificial neuron, has a threshold and weight that are related to one another. Any node is activated and sends data to the network’s next layer if its output exceeds the designated threshold value. Otherwise, that node does not transmit any data to the network’s subsequent layer [26].

5.2. Comparison Between Random Forest, Linear Regression, and Neutral Networks Using a Case Study

One study concerns a dataset including 102 087 tenders in Spain between 2014 and 2020 [23]. The dataset was compiled with multiple elements, such as budget of project, location, and the awarding price. According to [23], the use of linear regression, alongside other AI solutions, provides a positive insight in terms of accuracy for the estimate of the tender. This study compares different ML algorithms, including ANNs, linear regression, and random forest. The performance of the algorithm was confirmed by using multiple errors metrics such as Root Mean Squared Error (RMSE) and Mean Absolute Error (MAE). Findings: Despite its simplicity, linear regression leads to better accuracy compared to other models such ANNs and random forest; its application makes the estimate more accurate and helps to predict the right award price, which is important for a better management of public funds.

6. Conclusions

The objective of this paper is to highlight the different artificial intelligence solutions that could be used to enhance efficiency in public procurement, especially in the process of editing the estimate.

To investigate our subject, research was performed on Scopus, Web of Science, and ScienceDirect, and 135 articles were found between 2000 and 2024, by using artificial intelligence and estimate of public procurement as keywords.

By reading the summary, the introduction, and the results for all those articles, we reduced the number to 43 articles dealing directly with the use of artificial intelligence in public procurement.

The public procurement process starts with the elaboration of the specifications, the editing of the estimate, the launching of the call of tenders, and then the execution of the contract.

The financial performance of the public procurement act relies on the value of the estimate; in fact, according to regulations, the award price should not exceed a certain threshold compared to the value of the estimate.

Artificial intelligence could play a key role in making the estimate more accurate, leading directly to the financial performance of the public procurement act.

The findings suggest that the AI solutions that could be used in editing the estimate is machine learning.

Different subfields of machine learning exist, and the one that fits the step of editing the estimate is linear regression.

Future research work should be more interested in the implementation of a linear regression algorithm, which could be used to make the estimate of a public procurement act more accurate.