Adoption of SAP AI: SAP RPT-1
Estimating delivery cost, duration, and CO₂ emissions before a sales order is processed is a persistentchallenge. Carrier rates are often unavailable at that stage, and consolidated shipments make earlycost visibility even harder. SAP RPT-1 solves this by drawing on historical shipment data to generatereliable predictions, accessible by eCommerce platforms or other AI agents.
RPT stands for Relational Pre-trained Transformer: an LLM for tabular data, pre-trained by SAP on 2.18million real-world tables, using a Transformer architecture to understand relationships withinstructured datasets.
Unlike traditional machine learning, RPT requires no model building, no algorithm selection, and notraining cycle. It automatically determines the right task, whether classification, regression, or ranking,and returns results in real time, even when input data is incomplete.
RPT is particularly well suited when training data is scarce or low quality, when internal datapreparation expertise is limited, or when one model needs to serve multiple prediction tasks acrossthe business.
In practice, it supports four key logistics use cases: upfront delivery cost and CO₂ estimation foreCommerce, gap-filling in the transport graph where cost references are missing, loading andunloading duration prediction, and packaging dimension estimation before the load planner runs.
When to use SAP RPT-1
RPT delivers the most value when training data is scarce or low quality, when data preparationexpertise is not available internally, and when one model needs to serve multiple prediction taskswithout being rebuilt each time.
RPT in action: real-time predictive tables for logistics
Delivery cost and CO₂ estimation draws on historical transportation costs, route characteristics, andproduct dimensions to give eCommerce platforms reliable pricing and emissions data at the point oforder.
Completing the transport graph fills missing network links using historical spot carrier orders andexisting routes across different equipment types, giving planners a more complete cost reference.
Loading and unloading duration estimation combines client segmentation, regression on historicalpatterns, and observed route delays to produce accurate time predictions that close the gap betweenplanned and actual operations.
Dimensions for the load planner estimates missing packaging dimensions before the load plannerruns, based on product data and packing history, so downstream steps always operate on completeinputs
