How AI can Streamline the Procurement Process

Sammanfattning

Background – The tendering process in the construction industry is a resource-intensive and document-heavy process in which tender documents, technical specifications and submitted bids must be analysed and compared under time pressure. The manual working methods that characterise many of the process's central stages increase the risk of overlooking important requirements, reservations and deviations. At the same time, advances in artificial intelligence and large language models have created new opportunities to support document-intensive tasks through text analysis, summarisation and information structuring. Purpose – The purpose of this study is to investigate how existing AI-based tools can be used to support the analysis and structuring of tender documents in the bidding phase of the construction industry, and to assess the potential effects of such support in terms of time savings, quality and decision support compared to traditional manual working methods. Method – The study is conducted as a qualitative and exploratory case study in collaboration with K-Fastigheter, inspired by the framework “DRM, a Design Research Methodology” by Blessing and Chakrabarti (2009). Data collection consists of semi-structured interviews with four respondents active in tendering and procurement work, as well as analysis of real project documents from the electrical contract in a project in Gothenburg. Three AI tools were evaluated in the case study: ChatGPT, Microsoft Copilot and NotebookLM. Results/Conclusions – The results show that the AI tools were able to structure and compile information from tender documents and submitted bids in a clear and organised manner, with structured summaries generated in approximately eight to fifteen minutes. The tools identified key requirements, reservations, deviations and price differences between bids, improving the comparability and transparency of the decision-making basis. At the same time, the study shows that AI-generated results consistently required verification against the original documents, particularly where formulations were ambiguous or boundary conditions were complex. AI should therefore be regarded as a qualified decision support tool rather than a replacement for professional judgement. The study also identifies the quality of the tender documents as a decisive factor for both the accuracy of AI analysis and the overall outcome of the tendering process.