Enhancing New Product Development with Large Language Models
Open Access Depositedhowever, their effectiveness is also influenced by the quality of the datasets, prompt engineering, and the underlying LLM itself. This research concludes two key findings
While LLMs offer promising capabilities in enhancing NPD, the tendency of LLMs to generate hallucinations, including incomplete, inaccurate, or incorrect outputs, poses risks to development quality and efficiency. This research investigates the integration of LLMs into early-stage NPD activities, focusing on requirements traceability and effort estimation using a multi-role architecture (Actor, Judge, Meta-Judge). In order to evaluate the hallucinations and impact, three software engineering datasets (eANCI, eTOUR, iTrust) and multiple commercially available LLMs were evaluated, including Claude 3.5 Sonnet, 3.7 Sonnet, and GPT-4o.This research demonstrates that LLM-based approaches significantly outperform traditional NLP methods, depending on the dataset, achieving improved F2 scores and resulting in improvements ranging from 6.55% to 65.04% compared to NLP methods. These improvements are projected to reduce development timelines for similar scope projects. The relationship between hallucinations and estimation quality varies significantly depending on the dataset characteristics and quality. This research provides empirical evidence that LLMs can meaningfully enhance early-stage NPD activities
first, the quality of the datasets themselves can impact the performance of the output, and second, LLMs exhibit distinct performance characteristics that should be evaluated for individual projects.
- All rights reserved
Notice to Authors
If you are the author of this work and you have any questions about the information on this page, please use the Contact form to get in touch with us.
| Thumbnail | Title | Date Uploaded | Visibility | Actions |
|---|---|---|---|---|
|
|
MontgomeryRecht_gwu_0075A_17508.pdf | 2025-12-11 | Open Access |
|