Value Assessment of AI-Based Procurement : A Case Study of Autonomous Negotiations in Indirect Tail Spend

Sammanfattning

Background. Autonomous negotiation agents are increasingly deployed in indirect procurement to address tail spend and optimize workflows. However, the economic and organizational value of the systems is difficult to assess as the benefits are partly intangible, and negotiated outcomes may not be realized in downstream transactions. Objectives. This thesis examines how negotiation-level outcomes from AI-driven autonomous negotiations can be used to evaluate economic and organizational value in indirect procurement tails spend. Furthermore, the thesis proposes a context-specific value assessment framework for the studied implementation. Methods. The thesis applies a single-case quantitative explanatory design using secondary data from an ongoing deployment of Pactum AI’s procurement module. The empirical basis comprised 148 concluded negotiation cases extracted from Pactum negotiation logs and linked to the firm’s ERP purchasing data to distinguish negotiated from realized outcomes. The analysis combined descriptive and comparative statistics with logistic regression to estimate agreement probabilities and linear regression to model discount outcomes. Results. The agreement rate was 64.9%, but agreement was not equivalent to commercial improvement: approximately half of the agreement cases achieved a positive discount, while the median discount and median savings across agreement cases were zero. Among positive-discount cases, savings were highly concentrated, with the top five cases accounting for 63.15% of total positive savings. ERP matching verified 83.3% of agreement cases, indicating that a notable portion of negotiated outcomes could not be confirmed as executed at the time of extraction. Supplier feedback was on average favorable (mean 3.52 out of 5), while outcome satisfaction was lower (2.70). Conclusions. Negotiation-level outcomes can support evaluation of economic and organizational value when interpreted as layers of indicators with varying evidentiary strengths rather than as a single measurement. The proposed framework treats agreement as a threshold condition, enabling the commercial improvements, which appeared to be skewed and context-dependent. Furthermore, the ERP-validated outcomes and supplier-facing indicators represented organizational evidence. The findings indicate a carefulness in relying on average discounts and agreement rates alone when governing autonomous negotiation in indirect tail spend.