Understanding user perceptions of LLM-powered data access for business intelligence: A task-technology fit perspective in fintech

Thumbnail Image

URL

Journal Title

Journal ISSN

Volume Title

School of Business | Master's thesis

Authors

Date

2025-12-30

Department

Major/Subject

Mcode

Degree programme

Master's programme in Information and Service Management
Tieto- ja palvelujohtamisen maisteriohjelma

Language

en

Pages

81

Series

Abstract

Business intelligence faces persistent challenges, with data access remaining concentrated among technical specialists, and nearly 70-80% of projects failing to deliver expected returns. Large Language Models could offer a potential solution via natural language interfaces for data access through SQL translation, potentially eliminating technical barriers. However, technical capability does not guarantee adoption where verification and integration requirements may matter more than accuracy. This research investigates stakeholder perceptions of LLM-powered data access through Task-Technology Fit theory in a fintech startup context. A functional prototype using locally deployed LLMs (Gemma 3:4b via Ollama) was developed and evaluated through 30 queries across complexity levels. Semi-structured interviews with five stakeholders explored perceived fit and adoption barriers. Findings reveal that perceived fit varies substantially by use case. Simple operational queries achieved 83% success with strong acceptance, whilst advanced queries succeeded only 22%, raising verification concerns. Stakeholders valued operational workflows over strategic analytics, particularly multi-transaction lookup addressing immediate friction points. However, standalone architecture emerged as a critical barrier, with stakeholders favouring workflow-integrated deployment. The research extends Task-Technology Fit theory by establishing transparency and verification as core fit dimensions for algorithmic systems. Practically, findings suggest lightweight local LLMs deliver meaningful value for bounded use cases, challenging assumptions that accuracy improvements alone drive enterprise adoption

Description

Supervisor

Liu, Yong

Keywords

large language models, artificial intelligence, business intelligence, task-technology fit, text-to-SQL, data democratisation

Other note

Citation

Permanent link to this item

https://urn.fi/URN:NBN:fi:aalto-202601151321

Collections

[gradu] Kauppakorkeakoulu / BIZ