Marknadsintelligens i en ny era : en fallstudie om AI inom marknadsintelligens

Sammanfattning

ABSTRACT Date: 2026-06-08 Level: Master thesis in Industrial Engineering and Management, 30 ECTS Institution: School of Innovation, Design and Engineering, Mälardalen University Authors: Nour Battal Title: Marknadsintelligens i en ny era - en fallstudie om AI inom marknadsintelligens Tutor: Kristian Sandström Keywords: Artificial intelligence, Competitive Intelligence, LLM Research question: RQ1: How can an AI-driven market intelligence process be designed to ensure accuracy, completeness, and consistency? RQ2: How do different LLM models perform in terms of precision and analytical level in producing competitor analysis? RQ3: How can iterative methods improve the quality of AI-generated market intelligence? Purpose: The purpose of this study is to develop and evaluate an AI-driven market intelligence process and explore how different large language models perform in competitor analysis. Method: The study applies Design Science Research as an overarching research approach and is conducted as a case study at ABB Marine & Ports. A mixed methods approach is used, combining quantitative scoring of three quality dimensions (accuracy, completeness and consistency) with qualitative evaluation of the generated outputs. Three language models (GPT-4o, Claude Sonnet 4.6 and Gemini 2.5 Pro) are evaluated across two prompt versions and three competitors using identical source material. Conclusion: The study concludes that the quality of an AI-driven market intelligence process depends equally on prompt design and model selection. Claude achieves the highest performance across all three quality dimensions, while GPT-4o prioritizes source accuracy at the expense of analytical depth, and Gemini exhibits recurring accuracy problems. Two iterative methods, iterative prompt refinement and self-refine, demonstrate measurable effects on output quality, though their effect varies between models as self-refine reinforces each model's existing strategy for handling uncertainty. While automated review methods are advancing rapidly, the need for human review remains, since responsibility for the decision cannot be delegated to AI.