October 26, 2026
Predictive Cost Management in Projects using Monte Carlo Simulation
Alexandre Lovato Maidana; Adriano Machado
DOI: 10.22167/2675-6528-202600490
Article derived from a Course Conclusion Work (TCC), with content based on the student’s original work and adapted to the editorial format of the E&S Magazine with the support of the ResumeAI tool, an artificial intelligence solution developed by Instituto Pecege for textual synthesis and organization.
Summary
The inaccuracy in cost and effort estimates in engineering projects compromises financial predictability and stakeholder confidence. This study analyzed the deviations between planned and actual hours in multidisciplinary projects of an engineering company, seeking to identify patterns associated with disciplinary scopes and project size, and to propose a quantitative approach to improve cost estimates. A quantitative and applied methodology was adopted, based on a case study with documentary analysis of 224 completed projects, totaling 504 disciplinary scope executions. Hour consumption, defined as the ratio between actual and planned hours, was statistically analyzed and served as the basis for probabilistic modeling via Monte Carlo Simulation, with 1,000 iterations. The results indicated that the deviations did not occur randomly, exhibiting systematic behaviors. Disciplines such as Automation showed a consistent tendency to underestimate planned hours (average close to 1.40), while larger scopes exhibited greater variability and uncertainty. The simulation generated probabilistic scenarios, with values of 213 hours (P50), 588 hours (P80), and 910 hours (P90), and correction factors for budget adjustment according to risk levels. It was concluded that the integration of historical data with Monte Carlo Simulation replaces deterministic approaches with a risk-oriented perspective, contributing to more robust estimates and better decision-making in project cost management.
Keywords: Statistical analysis; Hours consumption; Cost estimation; Risk; Monte Carlo simulation.
1. Introduction
Project Cost Management is a fundamental knowledge area for the success of ventures, encompassing the processes necessary to ensure that the project is completed within the approved budget (PMI, 2017). The accuracy of cost and effort estimates represents one of the pillars of good project management, directly impacting the financial viability and strategic alignment of initiatives. In today’s competitive landscape, the ability to accurately forecast and control project expenditures is crucial for organizational sustainability and stakeholder confidence.
Despite its critical importance, both academic literature and organizational practice frequently highlight a recurring challenge: significant deviations between planned costs and actual expenditures. These discrepancies can lead to budget overruns, severely compromising project profitability and undermining stakeholder confidence (VARGAS, 2010). Such deviations are not mere random occurrences; they often result from systemic problems in the estimation process.
Traditionally, cost estimates are developed using techniques such as analogous, parametric, or bottom-up estimating. Although widely accepted and disseminated, these approaches tend to treat uncertain variables as deterministic values (PMI, 2017). This deterministic perspective often fails to adequately capture the uncertainties and risks inherent in project activities. Consequently, a pattern of systematic planning errors is observed, particularly the underestimation of the effort required for certain technical scopes, which leads to an excessively optimistic bias in budgets. The fragility in initial estimates demands a more comprehensive approach to risk assessment.
To overcome these limitations, Monte Carlo Simulation emerges as a robust technique for uncertainty analysis in complex systems. This method is based on generating multiple scenarios from probabilistic distributions associated with the variables of interest (HILLIER; LIEBERMAN, 2013). In the context of project management, Monte Carlo Simulation allows for the replacement of deterministic estimates with probabilistic models, incorporating the variability observed in the historical performance of activities. This transition from a single-point estimate to a spectrum of possible outcomes offers a more realistic view of potential project costs and durations.
The application of Monte Carlo Simulation in projects enables a comprehensive evaluation of cost and schedule behavior under different scenarios. It facilitates the estimation of the probability of deviations occurring and provides confidence intervals for estimates (VARGAS, 2010). This capability is particularly relevant in environments with a high degree of uncertainty, where small variations in assumptions can generate significant impacts on project outcomes. By explicitly modeling risk and variability, Monte Carlo Simulation offers a substantial advancement over traditional estimation techniques. It integrates historical data with probabilistic distributions, contributing to the development of more robust estimates that better reflect the actual behavior of projects and support risk-based decision-making.
A central problem that motivated this study is the absence of systematic mechanisms in organizations to adjust cost and hour estimates based on historical project performance. The repetition of underestimation patterns, especially in planning and execution, does not occur randomly; instead, it is related to the specific characteristics of scopes and project complexity. Although project management is often data-driven, a systematic method is needed to mitigate these risks. Identifying recurring historical patterns is fundamental to improving estimation processes and providing subsidies for predictive management.
The continuous refinement of cost control processes, which involves performance monitoring, comparison between planned and actual values, identification of deviations, and implementation of corrective actions (PMI, 2017), highlights the need for more reliable estimation tools. The analysis of historical patterns and quantitative data contributes significantly to reducing inherent uncertainty and increasing the reliability of forecasts. Thus, this study is justified by the need to overcome the limitations of deterministic estimation methods and to provide project managers with a more robust instrument for decision-making, especially in the initial planning phases, where uncertainty levels are higher. This research aims to analyze the deviations between planned hours and actual hours in multidisciplinary projects, using historical organizational data, seeking to identify patterns associated with disciplinary scopes and project size, and to propose a quantitative approach for improving cost estimates.
2. Material and Methods
The research was designed with a quantitative and applied approach, based on the analysis of numerical data to identify performance patterns and variability in the consumption of hours in projects (MARCONI; LAKATOS, 2003; GIL, 2008). The adopted design was the case study, allowing for an in-depth investigation of the phenomenon in its specific organizational context (YIN, 2015).
Data collection occurred through documentary research, using historical records extracted from the internal project control system of an engineering company, with corporate information confidentiality assured. The central unit of analysis was the disciplinary scope, allowing for granularity in the evaluation of hour deviations. The research method was structured in three stages: literature review, collection and treatment of historical data, and descriptive statistical analysis.
The literature review conceptually grounded the study, addressing cost estimation methods, risk analysis, and the application of Monte Carlo Simulation in projects (GIL, 2008), consulting scientific information databases with a focus on planning, project control, and simulation.
In the collection and treatment stage, an internal documentary base with 742 multidisciplinary projects was accessed. Only finalized projects were selected for a complete comparison between planning and execution. Records classified as subtotals and the “General” discipline were excluded. Observations with planned hours equal to or greater than one hundred hours were considered, to reduce distortions from reduced effort scopes.
Following these criteria, the analyzed database was composed of 224 finalized projects, totaling 504 executions of disciplinary scopes. Multiple disciplinary scopes from the same project were treated as independent observations. The central variable was the consumption of hours, defined as the ratio between hours performed and hours planned for each disciplinary scope.
Descriptive statistical analysis of hours consumed by disciplinary scope was performed, calculating measures of central tendency and dispersion. The scopes were segmented by size ranges, using the volume of planned hours as a proxy for complexity, to assess the influence of size on deviations. For stability of estimates, scope-size combinations with a minimum of five observations (N ≥ 5) were considered.
To increase statistical robustness, outlier values in the hours consumption variable were identified and treated using the interquartile range (IQR) method. The quartiles (Q1, Q2, and Q3) of the distribution were calculated, and the IQR was determined as the difference between Q3 and Q1. The lower (Q1 – 1.5 × IQR) and upper (Q3 + 1.5 × IQR) limits were established for outlier identification.
Twenty-seven records were identified as discrepant in a universe of 504 observations. Although kept in the original database for traceability, these outliers were disregarded in the calculation of the statistical input measures for the simulation, such as mean and standard deviation, to avoid distortions in the probabilistic results.
The prescriptive stage consisted of applying the Monte Carlo Simulation to model the risk associated with the estimates of hours consumed in future projects. The average values and standard deviations of hours consumed, calculated from the treated historical base and segmented by discipline and size range, were used as input parameters (VARGAS, 2010).
The simulation was implemented in a Microsoft Excel environment, using structured databases. The modeling considered the generation of random variables for the selection of the discipline and the size range, based on their frequency distributions, and the generation of random predicted hours within the defined intervals for each size range.
The consumption of hours was simulated by generating random values from a normal distribution, parameterized by the mean and standard deviation specific to each discipline-size combination (DE SOUZA RUBATINO et al., 2018). To avoid unrealistic values, a truncation was applied to the hours consumption variable, limiting the simulated values to an interval between 0.4 and 2.0, defined based on the empirical analysis of historical data.
1,000 independent iterations were performed, each representing a possible execution scenario for the scopes. This procedure allowed the generation of an empirical distribution of simulated hours performed (DA PAIXÃO et al., 2021). The distribution of simulated hours performed was analyzed using descriptive statistics and percentiles.
3. Results and Discussion
The analysis of deviations between planned and actual hours in multidisciplinary projects revealed systematic patterns that transcend randomness, offering an empirical basis for improving cost estimates. The study focused on hours consumption, defined as the ratio between actual hours and planned hours, a crucial indicator for evaluating estimate performance. Average values above one signal an underestimation of planned effort, while values close to one indicate greater adherence between planning and execution. This approach allowed for the identification of areas where traditional estimates prove most fragile, as per literature pointing to the recurrence of significant deviations in projects (VARGAS, 2010).
To ensure the statistical robustness of the analyses, only observations with planned hours equal to or greater than 100 hours were considered. This criterion aimed to mitigate distortions caused by low-effort scopes, which often present consumption ratios less representative of typical data behavior. The final database, composed of 224 projects and 504 disciplinary scope executions, was carefully processed, including the identification of 27 outliers by the interquartile range method, which, although retained for traceability, were excluded from the simulation parameters to avoid biases in the probabilistic results.
Analysis of Deviations by Disciplinary Scope
The investigation of hours consumed per disciplinary scope demonstrated significant variations in the performance of estimates. It was observed that disciplines such as Automation, Electrical, and Process presented average consumption values consistently above one. The Automation discipline, in particular, stood out with an average consumption of approximately 1.40, indicating a pronounced and recurring tendency to underestimate the planned hours for these scopes. This finding suggests that the estimation processes for Automation may be overly optimistic or do not adequately capture the inherent complexity of these activities.
The Electrical and Piping disciplines also showed average consumption above one, although with less magnitude than Automation, which still points to an underestimation, but to a less pronounced degree. In contrast, scopes such as Architecture and Mechanical presented average consumption values close to or slightly below one. This behavior indicates greater precision and adherence between planned and actual hours in these disciplines, suggesting that their estimation processes are more effective or that the nature of their activities is intrinsically more predictable.
Beyond consumption averages, the analysis of data dispersion, reflected by standard deviations, revealed varying levels of uncertainty across disciplines. Scopes with greater dispersion indicate higher variability in hour consumption, which translates into greater uncertainty associated with estimates. This result is consistent with project management literature, which recognizes variability as a critical factor in risk analysis and estimation accuracy (PMI, 2017). The presence of systematic deviations and significant variability reinforces the limitation of deterministic approaches and the need to explicitly incorporate uncertainty into the estimation process.
Analysis of Deviations by Size Range of the Scope
The segmentation of disciplinary scopes by size range, using the volume of planned hours as a proxy for project complexity, revealed that the behavior of hour consumption varies not only between disciplines but also as a function of scope dimension. The literature already indicates that variability and uncertainty in costs and efforts tend to increase with project complexity, especially in larger projects, due to the greater number of interfaces, technical dependencies, and operational uncertainties (PMI, 2017; VARGAS, 2010).
The results obtained confirmed this trend, showing that scopes classified in larger size ranges generally presented average consumption values greater than one, indicating greater underestimation. Additionally, these larger scopes exhibited higher dispersion levels, reflected by higher standard deviations. This behavior suggests that as the complexity and hours of a scope increase, the uncertainty in the estimates also rises, making them more susceptible to significant deviations.
In contrast, smaller scopes, in several cases, presented average consumption closer to one and lower dispersion. This greater relative predictability in smaller scopes can be attributed to their lower complexity and the smaller number of variables involved, which facilitates a more precise estimate. However, the analysis also revealed that this pattern is not homogeneous across all disciplines; some showed a consistent increase in average consumption with increasing scope, while others exhibited more stable behaviors or distinct trends between scope ranges.
This finding reinforces the importance of simultaneously considering the type of disciplinary scope and its size when evaluating the performance of hour estimates. The variability of performance, which increases with the complexity of the scopes, underscores the inadequacy of estimation approaches based on single deterministic values. The preliminary findings, therefore, provide a solid empirical basis for probabilistic modeling of hour consumption, which will be explored in the subsequent stage of the research, aiming to explicitly incorporate uncertainty into the estimation process.
Synthesis of Preliminary Findings
The descriptive stage of the research, based on the analysis of historical data, allowed for the identification of consistent patterns in the deviations between planned and actual hours. The characterization of the database, with 224 projects and 504 disciplinary scope executions, and the rigor in data treatment, including the exclusion of observations with less than 100 planned hours and the identification of outliers, ensured the reliability of the statistical analyses. These preliminary findings are crucial for understanding the nature of the deviations and supporting the application of more sophisticated methods.
The results demonstrated that the consumption of hours varies significantly between disciplines, with some scopes presenting a recurring pattern of underestimation of planned hours, such as Automation, with an average close to 1.40. Other scopes, such as Architecture and Mechanics, showed greater adherence between planning and execution. This variability is not random, but is associated with specific characteristics of each scope, indicating that the technical nature and complexity of the activities directly influence the accuracy of the initial estimates.
The segmentation of scopes by size range reinforced this evidence, showing that larger scopes not only present higher average consumption but also greater variability, reflected by higher standard deviations. This behavior suggests that scope complexity, measured by the volume of hours predicted, exerts a significant influence on the uncertainty of estimates, increasing the risk of deviations as the project becomes more complex. Together, these results indicate that deviations are influenced by multiple structural factors of the project, highlighting the limitation of deterministic approaches and the need for methods that incorporate the variability observed in historical data.
Probabilistic Modeling by Monte Carlo Simulation
Based on the variability patterns identified in the descriptive stage, the research moved on to modeling the risk associated with the estimates through Monte Carlo Simulation. This approach allows for the incorporation of uncertainty observed in historical data, generating multiple possible execution scenarios and offering a probabilistic view of the behavior of realized hours, as discussed in studies on stochastic modeling (ARENALES, 2011). The simulation was built from the treated historical base, considering the frequency distribution of disciplines and the scope size ranges.
For each of the 1,000 independent iterations, random variables were drawn representing the discipline, the scope size, the estimated hours, and the hour consumption. Hour consumption was modeled from a normal distribution, parameterized by the mean and standard deviation specific to each discipline-scope combination, a practice widely adopted in simulation models (DE SOUZA RUBATINO et al., 2018). To ensure the representativeness of the simulated values, truncation was applied to the hour consumption variable, limiting the values to an interval between 0.4 and 2.0, based on the empirical analysis of the historical data distribution.
The execution of the simulation generated an empirical distribution of the hours performed, which was analyzed through descriptive statistics and percentiles, following the classic probabilistic simulation approach (DA PAIXÃO et al., 2021). The results indicated an average of 376.19 hours for the simulated hours performed. The distribution showed high dispersion, with a minimum value of 16.40 hours and a maximum of 2834.00 hours, reflecting the intrinsic variability of the historical data and the range of possible scenarios.
The analysis of percentiles allowed for a more robust interpretation of the estimates from a risk perspective, enabling the definition of confidence levels for decision-making (BILINSKI, 2016). The 50th percentile (P50) corresponded to 213 hours, representing the median scenario of the distribution. The 80th percentile (P80), with 588 hours, indicated a more conservative confidence level, with an 80% probability that the actual value would not be exceeded. The 90th percentile (P90), with 910 hours, represented a scenario of high reliability, associated with situations of greater risk and uncertainty.
The comparison of these values with the average of 376.19 hours revealed an asymmetry in the distribution, characterized by a long tail to the right. This behavior indicates that, although most scenarios are concentrated in values close to the median, there is a significant occurrence of cases with substantially higher hour consumption, which directly impacts project planning and budgeting. This asymmetry reinforces the idea that deterministic approaches, based exclusively on average values, tend to underestimate the effort required for scope execution, a behavior widely discussed in the operations research literature (DA PAIXÃO et al., 2021).
From this analysis, relative mean correction factors were determined for each percentile: approximately 0.57 for P50, 1.56 for P80, and 2.42 for P90. These factors allow the simulation results to be translated into practical parameters for decision support, especially in the context of project planning and budgeting. Probabilistic modeling through Monte Carlo Simulation, a technique widely used for analyzing complex systems under uncertainty, enables the construction of scenarios and the assessment of risks associated with input variables (DE SOUZA RUBATINO et al., 2018).
In this way, the application of simulation directly contributes to the objective of this study by providing a structured approach to modeling the risk associated with future estimates. This methodology allows for the adoption of correction factors based on confidence levels, rather than being limited to average values, offering project managers a more robust tool for decision-making. The integration of historical data with Monte Carlo Simulation represents a significant advancement, replacing deterministic approaches with a risk-oriented perspective, which improves cost predictability and the management of project-inherent uncertainties.
4. Conclusion
The present study analyzed the deviations between planned and actual hours in multidisciplinary projects, using historical organizational data, with the aim of identifying patterns associated with disciplinary scopes and project size, and proposing a quantitative approach for improving cost estimates. It was found that deviations in hour consumption did not occur randomly but exhibited systematic behaviors. It was identified that disciplines such as Automation showed a consistent tendency for underestimation of planned hours, with an average close to 1.40. It was also observed that larger scopes showed greater variability and uncertainty in estimates, while smaller scopes demonstrated greater predictability. Probabilistic modeling, performed through Monte Carlo Simulation with 1,000 iterations, generated an empirical distribution of actual hours, with a mean of 376.19 hours and percentiles indicating different confidence levels, such as 213 hours for P50, 588 hours for P80, and 910 hours for P90.
The main contribution of this study lies in the proposition of a structured estimation model that integrates historical data with Monte Carlo Simulation, replacing deterministic approaches with a risk-oriented perspective. This methodology allows for greater cost predictability by explicitly incorporating the variability observed in the actual performance of projects. The application of the results enables project managers to adopt different confidence levels in estimates, according to the project’s criticality level, contributing to more robust uncertainty management and the reduction of budgetary deviations. The determination of mean-relative correction factors, such as 0.57 for P50, 1.56 for P80, and 2.42 for P90, offers an objective instrument for decision-making, especially in the initial planning phases, where the level of uncertainty is higher.
Bibliographic References
ARENALES, M. Pesquisa operacional para cursos de engenharia Rio de Janeiro: ELSIEVER, ABEPRO, 2011.
BILINSKI, P. A. Aplicação da pesquisa operacional na otimização da lucratividade de uma empresa do segmento de marcenaria João Pessoa: ENEGEP, 2016
DA PAIXÃO, Joelson Lopes et al. Métodos Matemáticos de Modelagem e Otimização: Teoria e Aplicações do Método de Monte Carlo. Salão do Conhecimento, v. 7, n. 7, 2021
DE SOUZA RUBATINO, Frederico; BATISTA, Saymon Henrique; DUARTE, Leandro Neves. Probabilidade de Ruína: Aplicação dos Métodos de Aoki e Monte Carlo, 2018;
GIL, Antonio Carlos. Como elaborar projetos de pesquisa. 4. ed. São Paulo: Atlas, 2008.
Hillier, F. S.; Lieberman, G. J. 2013. Introdução à Pesquisa Operacional. 9ed. AMGH Editora, Porto Alegre, RS, Brasil
MARCONI, Marina de Andrade; LAKATOS, Eva Maria. Fundamentos de metodologia científica. 5. ed. São Paulo: Atlas, 2003.
Project Management Institute [PMI]. 2017. Um Guia do Conhecimento em Gerenciamento de Projetos (Guia PMBOK®). 6ed. Project Management Institute, Newtown Square, PA, USA.
Vargas, R. V. 2010. Análise de Risco em Projetos. 2ed. Brasport, Rio de Janeiro, RJ, Brasil.
YIN, Robert K. Estudo de caso: planejamento e métodos. 5. ed. Porto Alegre: Bookman, 2015.
Article originating from the Final Course Project of the Specialization in Project Management of the MBA USP/Esalq
To learn more about the course, click here and access the MBX Academy platform