Author(s) :

Shaina Gupta.

Volume/Issue :

Abstract :

Startups are critical for global economic development, and they are often constrained by resources and affiliations, limiting their growth, unlike larger companies. Concerningly, nearly 70% of startups fail 2-5 years after their launch. The way to eliminate this discrepancy is to leverage commercial, readily available AI tools to assist with tedious tasks, so human capital and financial resources could be better spent on business development. This paper intends to use prompt engineering principles and evaluation rubrics that assess appropriate AI tools for different necessities of startups to answer the following question: How do different large language models (LLMs) vary in their prompt responses in AI-driven solutions for startups, and what implications does this have for selecting generative AI tools in small business contexts? An exclusive public dataset of prompts and commercial-LLM responses is being released that can be used by startups to evaluate the effectiveness of integrating AI tools for specific business activities. This dataset can be leveraged by startups of all types, to baseline the selection of AI tools, allowing them to allocate resources to more meaningful aspects of a business. This is being done for three business cases, including web design, market research, and business support. Each of these business cases have several prompts which are evaluated with 2-3 different LLMs to determine the optimal LLMs for different use cases. The key findings were that for certain use cases like web design, general usage LLMs like ChatGPT 5.0 produced optimal results, but in contrast, for other use cases like market research and business analysis, the specialized LLMs that provided lots of research performed better, like Claude. Therefore, the quality of results based on LLMs is on a case by case basis, but it can be extrapolated to the majority of prompts under the jurisdictions of those three use cases.