-
Introduction
Project cost determinants and their impact on project cost at the design, procurement, and construction stages assume increasing importance in project management day by day since competent cost estimation at the initial stage of a project is critical. The author suggests that although traditional models are widely used to estimate the cost of new projects, there is still a need to develop an innovative method to stay competitive in a world with ever-changing technology. This article presents an in-depth exploration of research into the use of generative artificial intelligence to estimate project cost. The author has divided this submission into five sections. Section 1 introduces generative AI and identifies the research gap (e.g. areas for future development). Section 2 reviews various methods and models used to estimate project cost. Section 3 explains the potential use of generative AI to estimate project cost and section 4 presents suggestions for areas of research.
Generative artificial intelligence (AI) is an attractive tool that could be applied to enhance project cost modelling. Project increased complexity introduces variability in features that require cost model prediction. These areas are mainly manual data input, detailed judgment processes of project characteristics by experts, or data mining to the few algorithms. Despite many developments in cost models, diverse methods are needed for the estimation of recurring costs. To deal with this issue, many deep learning architectures have been proposed in the literature. These are expected to embed both explicit and implicit descriptions of a domain required for cost estimation. Since they do not involve explicit requirements, there is also an ever increasing need to study the role of generative AI in project cost modelling.
1.1. Background and Significance
Traditionally, practitioners have relied on statistical methods that identify and categorize the correlations among the variables that define a project in order to forecast costs. Researchers have certainly made use of the latest data modelling and statistical methods as they emerge, with discriminant analysis, neural networks, kernel regression, conditional mean imputation, and many further advanced non-parametric methods in use today. Evidently, the very essence of using rectangular data methodologies, including generalized and structural equation modelling, latent variable analysis, and structural models, is reliant on parametric distributional properties as part of the underlying definitional test assumptions. With notable debate on the traditional lack of clarity in preferences, rigor in testing methods, the importance of correlation in foundational relationships, and the capture of historical data moving forward, the forecasting accuracy has remained incrementally improved using these methods for some time. Although small increments of improvement on existing corrupt data can still be realized, there exists a need for identifying more revolutionary innovations to truly improve accuracy in a meaningful way, and thus alternative data attractions are imperative contemporaneously.
The use of generative AI methodologies is proving increasingly significant, yet this is not reflected in current practice. After all, efficient and accurate computation is part of scientific discovery. The never-ending incremental increases in forecasting accuracy are weighted towards exhaustive analysis through data computation, freeing up time to concentrate on the science of interpreting robust forecasting outcomes, laying the blueprint for visionary project management control methodologies of the future. The cost of global consultancy for worldwide companies and organizations to hone project control capabilities, from identification and decomposition through to neural network models, has gone mainstream and is increasing exponentially. Therefore, decisive future-proofed and gender-agnostic project controls AI methodology-generated innovation has to be ethically and economically available to all. Specialists must be contemporary bar raisers in foundational research that utilizes AI methodologies as envisioned and envisaged methods of controlling mechanisms that dominate thinking. Agile computing practitioners play a key role in translating application-oriented literature and existing requirements to an efficient set of binary variables. The class of one must be developed using novel formative methodology packages as a fine grain in contested content blocks underpinned by the training mechanism paperwork in training workshop documentation for the first time.
The use of generative AI methodologies is proving increasingly significant, yet this is not reflected in current practice. After all, efficient and accurate computation is part of scientific discovery. The never-ending incremental increases in forecasting accuracy are weighted towards exhaustive analysis through data computation, freeing up time to concentrate on the science of interpreting robust forecasting outcomes, laying the blueprint for visionary project management control methodologies of the future. The cost of global consultancy for worldwide companies and organizations to hone project control capabilities, from identification and decomposition through to neural network models, has gone mainstream and is increasing exponentially. Therefore, decisive future-proofed and gender-agnostic project controls AI methodology-generated innovation has to be ethically and economically available to all. Specialists must be contemporary bar raisers in foundational research that utilizes AI methodologies as envisioned and envisaged methods of controlling mechanisms that dominate thinking. Agile computing practitioners play a key role in translating application-oriented literature and existing requirements to an efficient set of binary variables. The class of one must be developed using novel formative methodology packages as a fine grain in contested content blocks underpinned by the training mechanism paperwork in training workshop documentation for the first time.
The use of generative AI methodologies is proving increasingly significant, yet this is not reflected in current practice. After all, efficient and accurate computation is part of scientific discovery. The never-ending incremental increases in forecasting accuracy are weighted towards exhaustive analysis through data computation, freeing up time to concentrate on the science of interpreting robust forecasting outcomes, laying the blueprint for visionary project management control methodologies of the future. The cost of global consultancy for worldwide companies and organizations to hone project control capabilities, from identification and decomposition through to neural network models, has gone mainstream and is increasing exponentially. Therefore, decisive future-proofed and gender-agnostic project controls AI methodology-generated innovation has to be ethically and economically available to all. Specialists must be contemporary bar raisers in foundational research that utilizes AI methodologies as envisioned and envisaged methods of controlling mechanisms that dominate thinking. Agile computing practitioners play a key role in translating application-oriented literature and existing requirements to an efficient set of binary variables. The class of one must be developed using novel formative methodology packages as a fine grain in contested content blocks underpinned by the training mechanism paperwork in training workshop documentation for the first time.
1.2 Defining the focus of this study
The author identifies two main objectives for this study. The first is to develop (or identify) new knowledge on the extent to which artificial intelligence's generative AI technology might be used to improve the practices of project cost modelling. That is to understand its scope, how it can be applied, and to what extent it can support practitioners' tasks to improve efficiencies and reduce project costs. The second is to discover the key factors that might affect practitioners' confidence in a model that is enhanced by generative AI and therefore its integration into project cost models. Establishing these objectives early enables definition of what follows in this contribution.
The practical value of this article aims to contribute to two main areas: academia and industry. The development of a methodology to assess the potential for integrating generative AI into mundane tasks could not only be applied to other fields and other types of AI but could lay the groundwork for further study in this direction. Integrating AI models into physical cost models could have significant impacts on project management efforts to evaluate and anticipate project costs beforehand—from assessing feasibility in the early stages of project initiation to project revenue forecasts during planning and scheduling. This could also reduce costs and inaccuracies if it is integrated during the life cycle of the project, during design, pre-construction, and construction. This can support decision-making, potentially reduce the chances of losing money in construction, which has always been viewed as a high-risk gamble, and can give construction a competitive edge.
2. Fundamentals of Project Cost Modelling
One of the most fundamental aspects of planning for a project is cost estimation, though larger, more complex projects are often subject to budgetary restrictions from the outset. In order for management to make an effective decision, they need to know how much capital they will need and the expected investment return for the project. In operations management practices, budgeting and accounting professionals use project management tools such as work breakdown strategies to allocate resources and develop financial plans. These can include the total amount of an investment or the likely benefits to be produced by the project. The success probability of a project is increased by implementing a better cost forecasting policy, since a large number of projects fail due to financial matters or budget pressures in both the private and public sectors.
In construction and production projects, resource allocation is a critical element in cost estimating, which deals with the assignment of assets and operational materials. A fund, usually reserved for a specified need, is called a budget. In project management applications, the cost plan is also known as the budget plan since it allocates valuable human resources and equipment to various project activities. An efficient project leader helps to control the amount of budget to meet the project plan. The process of generating models to predict the project cost is known as project cost modelling. It is notably useful as it enables firms to avoid the dangers associated with underestimating or overestimating a project’s expenses, many of which can be detrimental or overwhelming. The expense or investment acquired to achieve the intended return from a particular project is a project cost. Science applications, finance applications, software engineering, human resource management, and healthcare are all common in project cost modelling. Investing time and money in project cost modelling increases the likelihood that unexpected issues that may occur during the execution phase would have a good cost forecast, leading to a good project completion period.
2.1 Traditional Methods
Various methods are currently used to forecast construction projects' costs, each with their own strengths and weaknesses. One commonly used method is expert judgment, which involves relying on the judgment of experienced senior managers, veteran foremen, or superintendents. The accumulated knowledge and intuition of these experts may provide an acceptable cost forecast. Another common method is analogous estimating. Analysts may attempt to benchmark some aspects of a new project to previous projects. Labour, material, and equipment costs can then be proportionally scaled based on these benchmarks. Reliable data must be available to conduct these comparisons; otherwise, judgment could be subject to bias. For cost forecasting on a more general project scope, rather than an individual project quantity, experts could perform parametric modelling to calculate cost. Parametric models use established statistical relationships between characteristics associated with the project and its cost.
All of these techniques can provide useful insights into the cost of a new project. Each technique has its relative advantages in certain scenarios. Expert judgment may be very reliable in cases where a project's costs and scope are very familiar to an expert. An analogous estimate may be reasonable for a quantity surveyor specializing in only factory construction. The senior management team may be able to provide quality estimates using their experience on previous warehouse construction projects. Naturally, there are clear drawbacks to these methods, stemming mainly from the reliance on often subjective and labour-intensive processes. In many cases, analogues do not provide the right context in relation to location, supply chain, demand, or other hard-to-quantify parameters. These inputs become more difficult to manage as the project itself becomes more complex. The more complex the project, the more likely current or historical analogues will lead to biased or misleading outcomes.
2.2. Challenges and Limitations
Financial planning of construction projects is critical to ensuring their delivery within budget and resources. While simple construction projects can be evaluated using conventional cost estimating methods, these methods have numerous limitations when applied to large, unique, and complex projects. One of the main limitations in cost modelling is inaccurate project duration and cost estimation due to unforeseen variables during project life. In practice, the inability of a project to conform to its cost model is called a budget discrepancy. A consistent budget discrepancy may lead to budget overruns, a primary cause of original income loss. In essence, a project may become financially or economically unproductive through excessive income reductions due to cost modelling limitations.
Although these limitations have been documented for more than half a century, the effectiveness of current modelling approaches is still a topic of research interest. Another drawback to traditional cost modelling is an inability to adapt to rapid shifts in project requirements. In an attempt to resolve these limitations, researchers and practitioners have addressed the topics of overcoming cost modelling limitations. Given the critical role of cost modelling in successful project execution, new developments in the field are increasingly necessary. The limitations we discuss here already impact many engineering and construction projects today.
3. Generative AI: Concepts and Applications
In recent years, deep neural network based generative artificial intelligence has become much more accessible due to leaps in computing capacity, an increase in research and development, and the release of more advanced algorithms. Among many other applications, generative AI is now being recognised for its potential to enhance the management of the cost modelling process in projects. It will impact areas by creating new digital tools, algorithms, and training techniques that can serve project management teams in the interpretation and recommendation of a wide range of project management-related decisions. Generative AI can be used as a valuable tool to simply provide the decision maker with options. Efficiency savings may be achieved by automating this process. The fact that there is choice makes the decision maker more likely to try the new tool and applies to a wide variety of projects in engineering, procurement, build-out, and supply chains across industries. Further, there are variations in terms of complexity, risk management, stakeholders, and need for flexibility. Risks and mitigation plans need to be articulated and understood. Generative AI is a modern artificial intelligence technique that can be used to create new data examples that are similar to a set of input data. This is often confused with other more common and sometimes more mature AI and machine learning approaches. In AI research, generative refers to the ability to create. The term 'generative' separates an AI technique from typical AI that a layperson might be familiar with, like classification, regression, clustering, or predicting outcomes in future time frames based on current and past state variables. Given this, it is much more useful to illustrate the applications and implications of a generative AI model. In contrast to other techniques, a 'conditionally generative' AI model is provided with seed data that is translated or interpreted in a similar way to other seed data examples.
3.1. Definition and Types of Generative AI
Artificial Intelligence (AI) technology is mainly divided into two categories: generative mode and discriminative mode. The former aims at generating new information given the data, while the latter tries to determine to which class the new information belongs. In recent years, with the thriving development and applications of big data, deep learning, and reinforcement learning, generative AI has become increasingly important and has consequently attracted substantial attention from academia, industries, and businesses globally. Currently, a variety of generative models are available, such as autoencoder, generative adversarial networks, variational autoencoder, autoregressive models, autoregressive integrated moving average, long short-term memory, sum-product networks, sequence generative adversarial networks, etc. All of these generative models essentially construct a mathematical framework to map a distribution of input data from their original domain to a distribution of output data in their replica domain. In other words, such generative models process data with a given distribution, while often involving the modelling of data correlation structure, to generate data with desired correlations.
Different generative models play different roles in scientific research and human life. Note that autoencoder, generative adversarial networks, and variational autoencoder are three state-of-the-art generative models among the list. It is noteworthy to clarify the most distinct features and peculiarities of the generative models in the eyes of this article’s focus. Contrary to traditional AI models, generative AI can model the data and generate new information that is often mastered without human intervention, while the discriminative models perform specific tasks based on the mastering of data. Thus, generative modelling ignores trivial data details during a specific task, which is helpful for addressing uncertainty problems. In the context of project cost management, by fully considering the original data of reasonably divided work packages and the associated literature, generative AI generates the dataset needed for risk assessment and risk distribution, and is in turn beneficial for bestowing the risk attributes of the generated data onto the constructed risk-assessment models and the extracted typical projects. This treatment can optimize the project finance management and cost control aspects encompassed herein. Overall, any type of generative AI can be used for modelling the patterns in cost data, a theory that has already been tested extensively and reported in the subject literature.
3.2. Examples of Generative AI in Various Industries
Current examples of generative AI applications come from various domains and industries, such as healthcare, manufacturing, entertainment, or social networking, where the technology can either enhance existing workflows or generate something completely new. In healthcare, for example, it contributed to advances in disease diagnosis, drug discovery, or medical imagery, dealing with the complexity, uncertainty, and incompleteness of medical data. Applications that were able to create music, art, or text have also been in the spotlight. Experimenting with generative AI in various sectors has unveiled its benefits for diverse fields. In internal machinery, generative AI has the capacity to generate parts with improved designs that are lighter while still offering the same strength levels. Some generative AI applications tackle challenges specific to their industries. For example, generative AI outputs customizable field conditions, predictions, and output analytics to uncover hidden patterns and predict yields on a per-field basis. Although not exclusive, due to the cross-industry popularity of generative AI, these examples show the technology's malleability.
Additionally, some of these enablers address context-specific issues and challenges. An online platform makes use of generative AI to create assets for the users' workspace while customizing elements so that the users' drawings or designs are not lost in similarity. An AI officer for cost planning draws attention to the life cycle and end-of-use implications of various computer chip models. Integrated with other software solutions, it assists in the investigation of sustainability in the supply chain by helping organizations conduct cost simulation scenarios. Even though cost planning applications, these AI enablers require direction to generate outputs that produce novel ideas, solutions, or insights. Considering usage in other domains and the problems they tackle, it becomes clear that generative AI can provide breakthroughs for a variety of project cost modelling issues.
4. Integration of Generative AI in Project Cost Modelling
Project cost modelling practice has been established by academia and industry for many decades. Despite the sophistication and complexity of project cost models, they often require extensive human effort to calibrate and are difficult to configure. Recent advancements in generative artificial intelligence have led to generative adversarial networks, which are capable of learning from data and generating new samples or outputs with similar properties as the training data. As generative AI is currently becoming democratised, some project cost management scholars and practitioners are exploring ways to integrate it into projects' costing practices. Some initial discussions have advocated for a principled and scientifically grounded approach to doing so, based on the specificities of the project cost management problem domain and the respective state of the art of generative AI.
At present, there are three potential approaches to integrating generative AI in project cost modelling practice: Integrated AI Systems; Two-Step Process; and Interpolated Sources. Each approach has great potential to reduce the effort required by personnel to conduct project cost estimates based on accuracy and efficiency criteria. However, potential barriers to generation acceptance and use also need to be addressed, which are thought to include fear of novel technologies and difficulties in AI model training due to inadequate or low-quality training data. Given that models of latent distributions continue to be improved upon with ingenious solutions, the integration approaches are also open to further exploration and fine-tuning. With the international project management and project cost management practice communities discussing potential routes to a more profound and systematic integration of generative AI, actual progress on these issues can be expected.
4.1. Advantages and Benefits
Many benefits and advantages arise from the use of generative AI in project cost modeling as outlined below.
4.1.1 Increased Accuracy in Estimates
One of the main benefits of the proposed approach is the increase in business intelligence, allowing the generation of a more accurate cost model than current methods. This is a result of greater volume and quality of training data, which helps to balance out the noise within the unit cost estimation process. As more noise is removed from the cost estimation phase, outputs are expected to be more representative of the characteristic cost distributions of different asset features, types, and classes, leading to more accurate low-level estimates.
4.1.2 Enhanced Forecasting
The increased number of more accurate low-level estimates consequently enables the generation of calculated cost models. Instead of simply summarizing the central tendency of past costs as in historical reference case cost models, calculated cost models use the increased volume and quality of the training data to, for example, more accurately represent typical projects through greater levels of granularity between different types of asset features, types, or classes. They can also be used to forecast higher-risk deviations from the mean costs to allow for budgets and economic cases that better cater to a range of potential future outcomes.
4.1.3 Decision Exploration in Project Modelling
Calculated cost models may include the ability to generate multiple cost forecast scenarios based on the types of decisions to be made. For example, the cost models for various grids include various scenarios, such as those based on wire cross-sectional area or the type of installation. The environments in which these will be made are characterized by a number of attributes, indicative of the complexity and variability of projects in practice. These include, for example, the scenario complexity – that is, the number of sub-scenarios or attributes the decision is based on, the number of solutions and training data available, and the level of higher risk of uncertainty in the forecast.
4.1.4 Faster Model Generation
The use of generative AI for cost model generation is efficient and reduces the time spent generating fine and calculated cost models in comparison to generating these using more traditional techniques. This time saving translates directly to reduced man-hours spent in cost model development and reduces the fixed and variable costs related to their production for each forecast. The lower time and skill needed to build the developed cost models will further lower the economic barriers to entry for new entrants in the area of renewables and other industries.
4.1.5 Enhanced Resource Planning
The benefits of the proposed approach also aid in the facility and asset management process. By using the developed estimated cost models, buyers can use this as one of the initial qualifiers to screen potential bids and tender responses. Furthermore, budgets for proposed work programs can be allocated prior to final decisions and tender awards, benefiting cost estimates in operations and supply chain.
4.2. Challenges and Considerations
Any digital construction platform will rely on AI algorithms to process and generate data. For cost modelling to benefit from generative AI, detailed and accurate data about cost, progress, and the factors that influence them are also required. As with any AI-dependent platform, the success of generative AI in predictive modelling is dependent on many factors. Data quality or a lack of data will undermine the precision of the model output. As such, the model could generate unrealistic outputs. This could put strain on resources preparing and cleaning the necessary datasets. To improve the usability of the model, there is also the challenge of making the output of the model interpretable. AI models, particularly any that base their output on thousands of minutes of generative analysis, can be seen as a "black box," which some end users may be hesitant to rely on. Furthermore, the reliance on generative AI, inevitably in today’s technological climate, indicates the need for technically skilled personnel. The cost and time implications of hiring, training, or paying an outside contractor for these personnel are important aspects to consider. It is noteworthy that many of these "risks" or "costs" will not be immediately apparent.
There is reluctance from a range of stakeholders to demonstrate interest in some of the most advanced uses of AI, which would include predictive modelling. Fundamentally, for big integration projects, organizations have had to change their culture in order to be successful. The risk is that it is also difficult for organizations to manage and monitor AI integration. To do this effectively, the change has to be phased and effectively managed, with investments in data and the generation of business cases to secure further funding. Bringing generative AI into predictive modelling is not just a plug-in solution; it relies on underlying company processes that need to be transformed first. Consideration needs to be given to monitoring the modelling output and adjusting the model, as all projects are unique. A proportion of the budget (or resources) will need to be spent on this part of the process to ensure clients have confidence in the modelling output.
5. Future Trends and Implications
Trends for the future emerging advancements in computer processing speed, hardware developments in central processing units, and very large memory chips are expected to shift how cost estimation has been performed for decades. Technical experts and data analysts alike should raise their awareness of the current status of cloud services, on the one hand, and the democratisation process of artificial intelligence programming paradigms at the leading companies. Equally important is the debate on the ethical implications of using generative AI, as the author suggests that any organization should remain totally transparent about the technology they use and comply with ethical and legal requirements. Also, future developments in the field of AI have been observed in generative artificial intelligence applications, which trade off performance against interpretability. The Future of Generative AI when looking ahead identifies that it is expected that black-box generative AI models will aim to synthetically generate higher cost samples in the presence of low-cost estimation data. Furthermore, the pre-trained generative adversarial networks that can accelerate learning through synthetic higher-cost samples when the original cost data is limited and are transferable across datasets to synthetically generate novel and high-cost estimations will emerge. The computer-generated drawings or text randomly generated in the field of visual or conventional computing, as opposed to AI applications, are usually positive and specific. Everything changes if these are used in business management, medicine, or finance, as in these cases the details make the difference. Finally the author suggests that the use of deep generative paradigms has to be very carefully adopted and must be based on an ethical and correct analytical procedure. It also needs to be highlighted that generative adversarial networks can be misused, posing a high risk of generating realistic-looking content with the potential to spread false information.