As part of its business model, Bulkio receives the margin between the price paid by the shipper and the price at which the job is sold to the matched carrier. However, the current pricing strategy is susceptible to potential human bias and relies heavily on historical performance. Furthermore, notable price variations exist across the different transportation lanes. This poses a considerable risk to the profit margins.

Therefore, Bulkio wants to improve its insights into the carrier prices, i.e., the minimum prices for which they can buy transport capacity from a carrier. These insights can be used to optimize the margins and to determine the relevance of incoming shipment requests.

To reduce these risk factors, Bulkio aims to integrate AI-driven recommendations based on historical purchasing data and external market conditions. Through collaboration with the client’s subject matter experts, multiple hypotheses were formulated to assess the potential influence of various factors on the carrier price.

Subsequently, a carrier pricing model was constructed by combining the most impactful factors. In this process, several AI models were iteratively evaluated to accurately estimate the carrier price distributions. Here, each order involved the prediction of 9 prices corresponding to different degrees of confidence (10%, 20%, …, 90%) on price acceptance. We utilized Kedro for constructing a modular data science pipeline, managing tasks such as input data preprocessing, price estimating, and generating business outputs.

The initial AI model for the prediction of a single price already leads to a high performance, explaining 88% of the variation in carrier prices while accounting for prices differences across lanes. Rather than providing a single price prediction, the final delivered model generates 9 distinct prices allowing for a more informed decision-making process by evaluating the confidence in price acceptance using prediction reliability.

The model is to be integrated in the company’s applications, in order to facilitate (near) real-time assessment of incoming orders, and supporting the decision-making process for optimal pricing strategies. The impact of various factors on (individual) model predictions are described using explainable AI techniques to increase model interpretability & adoption by the end-users.