1. Introduction
Enterprises in mechanical engineering face growing pressure to develop products more cost-effectively (Reference BjörkdahlBjörkdahl, 2020). Consequently, cost estimation has become critical in the early phases of product development, particularly during the concept and design phases. In these phases, cost estimates have far-reaching consequences (Reference Hennebold, Klöpfer, Lettenbauer and HuberHennebold et al., 2022; Reference Niazi, Dai, Balabani and SeneviratneNiazi et al., 2005): if they are inaccurate or delayed, uneconomical concepts continue to be pursued and design changes shift to later phases, where they are much more expensive. By contrast, accurate and timely cost estimates support economically viable design decisions, while reducing costly late modifications and shortening product development cycles and time-to-market (Reference Röhrenbacher, Moerth-Teo, Schwarz, Schnöll and RamsauerRöhrenbacher et al., 2021). However, generating such cost estimates is challenging, as data are limited, uncertainty is high, and time pressure is significant in the early phases.
In recent years, various non-AI-assisted methods, so-called traditional methods, have been developed to address these challenges (Reference Wouters, Morales, Grollmuss, Scheer, Epstein and MalinaWouters et al., 2016; Reference Niazi, Dai, Balabani and SeneviratneNiazi et al., 2005; Reference Hicks, Culley and MullineuxHicks et al., 2002). While these traditional methods support decision-making in the early phases, many of them are highly time-consuming and costly, as they typically require manual execution by cost engineers. Beyond execution time, cost engineers also need in-depth knowledge of relevant products and processes (Reference Niazi, Dai, Balabani and SeneviratneNiazi et al., 2005). Without such expertise, accurate cost estimation is nearly impossible because detailed product design and manufacturing processes are not yet defined in the early phases.
In this context, artificial intelligence (AI) offers opportunities for faster and more accurate cost estimation methods in the early phases (Reference Loyer, Henriques, Fontul and WiseallLoyer et al., 2016). For example, AI-assisted methods for cost estimation have shown promise in reducing the time and expertise required by traditional methods while improving accuracy and scalability, even when only limited data are available (Reference Hennebold, Klöpfer, Lettenbauer and HuberHennebold et al., 2022). Although the technical potential of AI in cost estimation has been widely acknowledged (Reference Bodendorf, Merkl and FrankeBodendorf et al., 2021; Reference ElmousalamiElmousalami, 2020; Reference Shamim, Hamid, Nyamasvisva and RafiShamim et al., 2025), empirical knowledge about practitioners’ actual needs, challenges, and expectations regarding its application has not yet been systematically investigated. To address this gap, the goal of this paper is to empirically examine current cost estimation practice and practitioners’ needs and expectations toward AI-assisted methods in the early phases. The study provides three contributions: (1) empirical evidence on how cost estimation in the early phases of product development is performed and used to support design decisions in mechanical engineering; (2) an empirically grounded overview of the dominant trade-offs, challenges, and limitations that shape cost estimation in the early phases; and (3) practical requirements for future AI-assisted methods for cost estimation in the early phases, derived from practitioners’ expectations. The study combines a qualitative interview study with cost engineers and a quantitative online survey across mechanical engineering enterprises of different sizes and product domains. Overall, the study is guided by three research questions (RQs):
-
• RQ1: What is the role and current practice of cost estimation in the early phases of product development in mechanical engineering?
-
• RQ2: Which trade-offs, challenges, and limitations shape cost estimation in the early phases of product development in mechanical engineering?
-
• RQ3: What do practitioners expect from future AI-assisted methods for cost estimation in the early phases of product development in mechanical engineering?
This paper is structured as follows: Section 2 reviews related work, with a focus on existing studies on cost estimation in mechanical engineering and related domains. Section 3 describes the methodology of the study. Section 4 presents the results. Section 5 discusses the practical implications of the results and highlights their relevance for both research and practice. Finally, Section 6 provides a summary.
2. Related work
In mechanical engineering, cost estimation during the early phases of product development has been widely examined in the scientific literature, highlighting its importance in preventing unexpected costs and ensuring the economic feasibility of products (Reference SavorettiSavoretti, 2018; Reference Ma’ruf, Nasution and LeuveanoMa’ruf et al., 2024). In addition, prior work has proposed structured classification and framework-based approaches to organize product cost estimation methods (Reference Niazi, Dai, Balabani and SeneviratneNiazi et al., 2005; Reference Altavilla, Montagna and CantamessaAltavilla et al., 2017; Reference Wouters, Morales, Grollmuss, Scheer, Epstein and MalinaWouters et al., 2016). Traditional methods, such as case-based reasoning (Reference Takai and BangaTakai & Banga, 2014) or parametric models (Reference Kasie and BrightKasie & Bright, 2021), are widely used in the field. While these methods can be effective in well-understood domains and when applied by experienced cost engineers, they are often time-consuming and cost-intensive (Reference Bodendorf, Merkl and FrankeBodendorf et al., 2021). Moreover, they tend to lack adaptability and scalability in the face of increasing product complexity and shorter product development cycles (Reference Röhrenbacher, Moerth-Teo, Schwarz, Schnöll and RamsauerRöhrenbacher et al., 2021).
In response to these limitations, several authors (Reference Manuguerra, Mandolini, Germani and SartiniManuguerra et al., 2023; Reference Rapaccini, Cadonna, Leoni and De CarloRapaccini et al., 2022) have proposed the use of AI and, in particular, machine learning (ML) to support cost estimation. Many of these studies demonstrate that ML-based methods can deliver accurate cost estimates while reducing the time and expertise required. However, most of these contributions adopt a technical perspective and tend to overlook practical challenges (Reference Shamim, Hamid, Nyamasvisva and RafiShamim et al., 2025), such as how ML-based methods can be applied in enterprises and integrated into established product development processes, particularly in the early phases. In addition, many ML-based methods still act as black boxes (Reference MolnarMolnar, 2022), making it difficult for cost engineers to understand how the cost estimates are generated (Reference Ribeiro, Singh and GuestrinRibeiro et al., 2016), which in turn limits trust (Reference Yoo and KangYoo & Kang, 2021; Reference Moawad, Islam, Kim, Vijayagopal, Rousseau and WuMoawad et al., 2021).
Empirical research that examines cost estimation from practitioners’ perspectives is still limited. However, a few empirical studies provide insights into real-world challenges and experiences. Reference Roy, Colmer and GriggsRoy et al. (2005) conducted two case studies in the automotive domain and emphasized the importance of structured and reliable data for accurate cost estimation. Reference Bendul and ApostuBendul and Apostu (2017) reviewed various cost estimation methods and empirically evaluated 190 cost estimation cases in the automotive domain, identifying differences in accuracy and application-specific limitations. Reference Lutz, Bodendorf, Stepanek and FrankeLutz et al. (2021) addressed the question of whether software for cost estimation actually led to cost and time savings by empirically analyzing its use in aerospace, automotive, and mechanical engineering. Reference Bodendorf and FrankeBodendorf and Franke (2022) examined the acceptance of ML-based methods at an automotive manufacturer. Their findings, based on user surveys and observations, highlight the need for improvements in perceived usefulness and user-friendliness to support broader adoption. A case study by Reference HammannHammann (2024) demonstrated that ML can outperform cost estimates by cost engineers in the early phases, particularly for highly complex automotive systems, while also illustrating the practical challenges of interpreting cost estimates.
Despite these contributions, there is no empirical study that systematically investigates how practitioners in mechanical engineering currently approach cost estimation, what specific challenges they face in the early phases, and how they assess the potential of AI-assisted methods in this context. Existing studies primarily focus on evaluating the performance of specific methods, rather than examining the underlying problems, needs, and expectations of practitioners. This shows the need to shift attention from method-centered evaluations toward practitioner-centered insights. This need has become even more relevant, as many practitioners have recently gained initial experience with AI (Reference Chatterjee, Rana, Dwivedi and BaabdullahChatterjee et al., 2021), and AI is increasingly being adopted in mechanical engineering enterprises (Reference Kinkel, Baumgartner and CherubiniKinkel et al., 2022).
3. Methodology
To address the RQs, an exploratory sequential mixed-methods design was applied, following Reference CreswellCreswell (2014). This design combines qualitative and quantitative methods to capture in-depth practitioner insights and their broader prevalence across the field. It is suitable for exploring real-world practices in contexts where prior empirical knowledge is limited, as it allows building on qualitative findings to inform a subsequent quantitative phase. The study was conducted in two sequential steps: a qualitative interview study (April–July 2025), followed by a quantitative online survey based on these results (September 2025). Figure 1 illustrates the study design.
Methodology of the empirical study
In Step 1, a qualitative interview study was conducted with cost engineers (sample size n = 22) from 14 mechanical engineering enterprises. Participants were selected based on their direct professional responsibility for cost estimation. Initial participants were recruited from the researchers’ professional environment, with additional participants identified through contacts in other enterprises. The semi-structured interviews were conducted on-site or via Microsoft Teams and were held in German. Each interview lasted approximately 60 minutes and was audio-recorded, transcribed, and anonymized. Before each interview, participants were informed about the study purpose and data usage and provided verbal informed consent. Participation was voluntary and all data were processed in accordance with applicable data protection regulations. The interview guide comprised a total of 38 open-ended questions covering all three research questions (RQ1–RQ3). The complete guide is available as supplementary material (Reference Michelberger, Klöpfer and HuberMichelberger et al., 2026). The guide was structured by the RQs and used a consistent set of questions. The analysis followed established qualitative research principles (Reference Bortz and DöringBortz & Döring, 2006) and a research-in-the-typical perspective (Reference Kitchenham, Pickard and PfleegerKitchenham et al., 1995). Reflexive thematic analysis was applied according to Reference Braun and ClarkeBraun and Clarke (2006, Reference Braun and Clarke2019). In total, 1,196 initial codes were generated through detailed line-by-line coding of the transcripts. These codes were refined, merged, and structured in several iterative cycles, ultimately resulting in 23 themes and 50 subthemes. Throughout this process, constant comparison and reflection ensured methodological rigor. The final themes were reviewed against the criteria of internal homogeneity and external heterogeneity as proposed by Reference PattonPatton (1990).
In Step 2, a quantitative online survey was conducted to test and generalize the qualitative findings. The survey assessed agreement and prevalence across enterprises of different sizes and domains. It was available in German and distributed via professional channels and enterprise contacts, including direct invitations, LinkedIn outreach, and forwarded invitations. The questionnaire was systematically developed from the qualitative results of Step 1. The complete questionnaire is available as supplementary material (Reference Michelberger, Klöpfer and HuberMichelberger et al., 2026). The questionnaire comprised 22 statements and seven questions. The statements and one of the seven questions were derived from one or more themes and subthemes identified in the thematic analysis in Step 1. For this purpose, the themes were translated into clear, practice-oriented statements and questions. Theme statements were first formulated, then converted into survey items, and iteratively refined to ensure clarity, unambiguous wording, and practical relevance. The remaining six questions captured demographic and enterprise-related characteristics. For each item, respondents rated their level of agreement on a 4-point Likert scale. After pretesting with three target users, wording adjustments were made. No incentives were provided for participation and incomplete survey responses were excluded from the analysis. The final sample (sample size n = 102) included 32 design engineers, 56 cost engineers, and 14 managers. Data were analyzed using descriptive statistics and cross-tabulations to triangulate and validate the qualitative insights from a research-in-the-large perspective (Reference Kitchenham, Pickard and PfleegerKitchenham et al., 1995).
4. Results
This section presents the results for RQ1–RQ3, outlining the role, current practice, trade-offs, challenges, and limitations of cost estimation, followed by expectations for AI-assisted methods.
4.1. Role and current practice of cost estimation (RQ1)
In practice, cost estimation has a central role in the early phases of product development and is characterized by limited data and a shift toward more function-oriented cost estimation.
Finding 1. Cost estimation as a design decision enabler
The interview study revealed that all 22 cost engineers viewed cost estimation not only as a calculation task but as a key enabler of early design decisions. Eight emphasized its role in legitimizing design proposals to management and stakeholders. However, cost engineers from two of the 14 enterprises reported that cost estimation is still regarded as a formal task rather than as an integral part of design decisions, indicating that the potential of systematic cost estimation for savings is not yet recognized.
The online survey confirmed this finding: 92% of respondents rated cost estimation as indispensable in the early phases. Respondents from small and medium-sized enterprises (SMEs) agreed more strongly than those from large enterprises (LEs) (Figure 2 A–A2). Only a minority (8%) of respondents, particularly design engineers, regarded cost estimation as a formal obligation.
Importance of cost estimation (online survey, n = 102)
Finding 2. Experience and implicit knowledge as enablers of early cost estimation
The interview study revealed that experience and implicit knowledge are key enablers of cost estimation in the early phases of product development. Ten of the 22 cost engineers described it as an iterative process: cost estimates in early phases are mostly approximate and intuitive, while more data in later phases allow for greater analytical accuracy. All 22 cost engineers emphasized that, when design data are limited and uncertainty is high, experience and implicit knowledge form the basis for cost estimates.
The online survey supported this finding, with 99% of respondents agreeing that the reliability of early cost estimation depends primarily on experience and implicit knowledge rather than on data alone.
Finding 3. Early cost estimation relies on traditional methods and tools
The interview study showed that cost estimation still depends on traditional methods and tools. Cost engineers combine different methods, often referencing similar existing products. When no comparable product is available, cost estimates rely on intuitive assumptions about weight, material, manufacturing processes, or target costs. The choice of method depends on data availability, time, and required accuracy. Five of the 22 cost engineers noted a shift toward function-oriented cost estimation, meaning that costs are estimated based on product functions or performance requirements rather than defined components. However, despite this shift, 20 of the 22 cost engineers stated that spreadsheets remain the primary tool in the early phases. These 20 also reported using enterprise-specific templates or in-house tools. In addition, six of the 14 enterprises used specialized software such as Product Cost Calculator (PCC) or Teamcenter Product Cost Management (TcPCM). Cost estimation processes are diverse, often enterprise-specific, and rarely AI-assisted. Only one cost engineer reported using an AI-assisted tool developed in-house that identifies similar products to derive reference costs for new products.
The online survey supported these findings: 48% confirmed a shift toward function-oriented cost estimation, mainly in SMEs, and 84% stated that spreadsheets remain the primary tool (Figure 3 B–C2).
Practice of cost estimation (online survey, n = 102)
In terms of AI use, the online survey showed that 69% reported not using AI in cost estimation, 28% were unsure because they only use cost data, and 3% indicated that AI is already applied. One mentioned product cost optimization, another an in-house tool, and a third the use of ChatGPT and Gemini.
4.2. Trade-offs, challenges, and limitations in cost estimation (RQ2)
In the early phases of product development, cost estimation is shaped by trade-offs as well as challenges and limitations that affect its accuracy, speed, and practical use, often driven by data availability.
Finding 4. Speed vs. accuracy: fast decisions over precision
The interview study showed that cost engineers face a trade-off between speed and accuracy. In the early phases, they work under time pressure and with limited or unreliable data, forcing them to prioritize fast, intuitive cost estimates. 14 of the 22 cost engineers emphasized that accurate cost estimation requires high-quality data and sufficient time, both typically unavailable in early phases. Consequently, early cost estimates are fast but less accurate, while later ones are more accurate but time-consuming.
The online survey supported this trade-off: 74% agreed that in early phases, speed outweighs accuracy. Respondents from development departments agreed more strongly than others (Figure 4 D–D2).
Finding 5. Over- vs. underestimation: practitioners’ tendency to overestimate
The interview study showed that cost engineers face a trade-off between over- and underestimation. Overestimation can stop product developments too early, while underestimation may continue unprofitable ones. Six cost engineers noted that overestimation is easier to justify, as underestimation carries greater risks when unforeseen costs arise later, so cost estimates include contingencies.
The online survey confirmed this: 74% of respondents indicated that overestimation is more tolerated, with those from development departments agreeing more strongly (Figure 4 E–E2).
Finding 6. Cost focus vs. design freedom: cost pressure limits innovation
The interview study showed that cost engineers face a trade-off between cost focus and design freedom in the early phases of product development. When cost focus dominates, design engineers rely on established solutions, as new solutions are harder to estimate and often seen as more expensive. Four of the 22 cost engineers emphasized that cost focus can limit innovation and reduce creative freedom, since product development typically continues only when sufficient profitability is expected.
The online survey supported this finding: 83% of respondents agreed that a strong cost focus restricts design freedom. Those from development departments agreed more strongly (Figure 4 F–F2).
Trade-offs in cost estimation (online survey, n = 102)
Finding 7. The data dilemma: estimating the unknown in early phases
The interview study showed that 18 of the 22 cost engineers work with limited data. Cost estimates rely on assumptions, increasing uncertainty and reducing reliability. Cost engineers described this as a challenge, as missing data are replaced by intuition or experience. They also noted that additional uncertainties arise from design changes, production volumes, and factors such as material prices. Consequently, 13 cost engineers use contingencies or scenario analyses, which further reduce accuracy.
The online survey confirmed these findings: 90% reported that poor data quality hinders cost estimation and reduces its accuracy and 93% agreed that uncertainty reduces reliability (Figure 5 G–H).
Finding 8. The organizational challenge: unclear responsibility and time pressure
The interview study indicated that eight of the 22 cost engineers reported coordination-driven challenges, particularly unclear responsibilities, which hinder cost estimation. In four of the 14 enterprises, cost estimates are created once and are rarely updated because it is unclear who is responsible for maintaining them, leading to outdated results and incorrect decisions. Another challenge identified by four cost engineers was time pressure, as design engineers require cost estimates quickly to advance their work, putting cost engineers under pressure.
The online survey supported these findings: 86% of respondents stated that unclear responsibilities weaken reliability and 84% noted that time pressure reduces cost estimation quality (Figure 5 I–J).
Challenges and limitations in cost estimation (online survey, n = 102)
4.3. Expectations for AI-assisted cost estimation (RQ3)
In response to the identified challenges and limitations, the interviews revealed clear expectations for AI-assisted methods in cost estimation.
Finding 9. The performance expectation: speed meets accuracy
The interview study indicated that 16 of the 22 cost engineers expect AI-assisted methods to be at least as accurate as traditional ones. According to the interview data, deviations of ±20% are accepted in early phases with limited data, while deviations of ±10% are expected when more reliable data are available. Larger deviations result from uncertainty or missing expertise. Cost engineers highlighted the need for fast, easy-to-use tools requiring minimal user input and clear visuals.
The online survey confirmed this: 90% agreed on accuracy and 95% on usability (Figure 6 K–L).
Finding 10. The explainability expectation: traceability enables trust
The interview study revealed that 18 of the 22 cost engineers expect AI-assisted methods to deliver explainable, traceable results that build trust and let users verify or adjust parameters when needed.
The online survey confirmed this: 98% emphasized explainable results and 96% agreed that AI-assisted cost estimation should be traceable and allow users to influence parameters (Figure 6 M–N).
Finding 11. The integration expectation: AI must fit organizational reality
The interview study showed that 14 of the 22 cost engineers expect AI-assisted methods to identify cost drivers and align with existing organizational structures to ensure acceptance and practical adoption. Half of the cost engineers emphasized that integration with existing systems is crucial.
The online survey confirmed these expectations: 100% of respondents cited the need to identify cost drivers and 86% noted integration into existing systems (Figure 6 O–P).
Expectations for AI-assisted cost estimation (online survey, n = 102)
5. Discussion
This section discusses the results along RQ1–RQ3, links them to prior work, and derives implications for AI-assisted cost estimation in the early phases of product development.
5.1. Role and current practice of cost estimation (RQ1)
The results show that cost estimation in the early phases of product development is not only a calculation task but a central support for design decisions under uncertainty and limited data. Both the interview study and online survey findings (Findings 1–3) show that early cost estimation mainly relies on practitioner experience and implicit knowledge (Reference Hellweg, Brückmann, Beul, Mandel and AlbersHellweg et al., 2022; Reference Mandolini, Campi, Favi, Germani and RaffaeliMandolini et al., 2020). Prior literature highlights missing data and methodological limits in early cost estimation (Reference Niazi, Dai, Balabani and SeneviratneNiazi et al., 2005; Reference Röhrenbacher, Moerth-Teo, Schwarz, Schnöll and RamsauerRöhrenbacher et al., 2021). The results in this study extend this view by showing that practitioners see early cost estimation as a knowledge- and coordination-driven task. All 22 interviewed cost engineers emphasize the role of experience, and 99% of the 102 survey respondents confirm this. This supports earlier observations that organizational context and collaboration matter for cost estimation quality (Reference Lutz, Bodendorf, Stepanek and FrankeLutz et al., 2021). Effective cost estimation therefore depends not only on methods and tools, but also on how knowledge is shared and maintained.
5.2. Trade-offs, challenges, and limitations in cost estimation (RQ2)
The results show that early cost estimation is shaped by recurring trade-offs (Findings 4–8). The interview study and the online survey data point to the same tensions: speed versus accuracy, over- versus underestimation, and cost focus versus design freedom. For example, 74% of survey respondents state that speed is more important than accuracy in the early phases, and 83% report that strong cost focus can restrict design freedom. Since the costs of design changes increase by approximately a factor of ten in each subsequent product development phase, as described by the Rule of Ten (Reference Ehrlenspiel, Kiewert and LindemannEhrlenspiel et al., 2020; Reference Newnes, Mileham, Cheung, Marsh, Lanham, Saravi and BradberyNewnes et al., 2008), there is strong pressure to make economic decisions early. Practitioners accept reduced accuracy and use contingencies to avoid slowing down product development. The contribution of this study is to make this trade-off structure explicit based on empirical evidence.
5.3. Expectations for AI-assisted cost estimation (RQ3)
The results show a differentiated view of AI-assisted cost estimation (Findings 9–11). The online survey indicates that current use is low. Only 3% of survey respondents report using AI in cost estimation, and this is limited to later phases when drawings are available. The interview study identifies specific expectations for AI-assisted methods, including acceptable deviation ranges (±20% with limited data and ±10% with reliable data), parameter control, explainability, traceability, and integration into existing systems. The online survey assesses these expectations across a broader sample. Expectations reported in the survey are high: 90% of respondents demand accuracy comparable to traditional methods, 86% require integration into existing systems, and 96% emphasize parameter control. The results therefore do not indicate general skepticism toward AI, but rather a discrepancy between current AI tools and the requirements in the early phases. Existing tools mainly support later phases with richer data, while early phases still rely on spreadsheets and experience-based cost estimation. Compared with prior technically focused AI-assisted cost estimation research (Reference Rapaccini, Cadonna, Leoni and De CarloRapaccini et al., 2022; Reference Manuguerra, Mandolini, Germani and SartiniManuguerra et al., 2023), this study adds empirical evidence on practitioner requirements. The findings therefore indicate specific requirements for AI-assisted cost estimation in early phases, including applicability with limited input data, explainability, traceability, and compatibility with existing systems.
5.4. Limitations and future research
Participants were recruited through professional networks and the study focused on mechanical engineering enterprises, which may limit generalizability. Findings reflect self-reported practices and perceptions rather than measured estimation performance and should be interpreted as practitioner-centered evidence on current practice, challenges, and expectations. Future research should evaluate AI-assisted cost estimation tools in real product development to quantify effects on decision quality, speed, and cost outcomes, and to examine how implicit knowledge and cost drivers can be captured and how practical forms of explainability can foster trust and adoption.
6. Summary
This paper examines why cost estimation matters in the early phases of product development in mechanical engineering and what practitioners expect from AI-assisted methods. The study combines a qualitative interview study with 22 cost engineers from 14 enterprises and a quantitative online survey with 102 respondents, including design engineers, cost engineers, and managers from more than 30 mechanical engineering enterprises. The results show that early cost estimation is a key enabler of design decisions but is constrained by limited data and therefore relies strongly on experience and implicit knowledge. In addition, cost estimation in the early phases is shaped by recurring trade-offs between speed and accuracy, over- and underestimation, and cost focus and design freedom. The study further consolidates the dominant practical challenges that drive these trade-offs, including limited data availability and quality, uncertainty, time pressure, and unclear organizational responsibilities. Only a few practitioners reported using AI in cost estimation and its use is mainly limited to later phases where detailed drawings are available. At the same time, practitioners expect AI-assisted methods to provide accurate cost estimates while being fast, explainable, traceable, and easy to use, and to integrate into existing systems and product development processes. Overall, the paper provides practitioner-centered empirical evidence on cost estimation practice and derives practical requirements and implications for designing AI-assisted methods in the early phases.
Open access
