rel-salt SALT sales database
Database Description: The Sales Autocompletion Linked Business Tables (SALT) database is an enterprise sales order relational dataset released by SAP AI Research. It models end-to-end business transactions in an ERP system, centered on sales document headers and line items, and linked to customer and address master data. The dataset captures structured information such as sales offices and groups, plants and shipping points, payment terms, shipping conditions, and Incoterms classifications. Each sales document and item is timestamped by creation time, enabling temporal modeling of business processes. SALT is designed to support multi-table relational learning tasks, particularly categorical prediction problems in real-world supply chain and order fulfillment settings.
Database Statistics:
| Domain | Medical |
| Num of Tables | 4 |
| Num of Rows | 4,257,145 |
| Num of Columns | 31 |
| Starting Time | 2018-01-02 |
| Validation timestamp | 2020-02-01 |
| Testing timestamp | 2020-07-01 |
Database schema:
To load this relational database in RelBench, do:
from relbench.datasets import get_dataset
dataset = get_dataset("rel-salt")
References:
[1] SALT: Sales Autocompletion Linked Business Tables Dataset.
Dataset License: CC-BY-NC-SA-4.0.
Entity Classification Tasks
item-plant
Task Description: For each sales order item, predict its plant (production/storage facility).
Evaluation metric: MRR
item-shippoint
Task Description: For each sales order item, predict its shipping point (dispatch location).
Evaluation metric: MRR
item-incoterms
Task Description: For each sales order item, predict its item-level international commercial terms.
Evaluation metric: MRR
sales-office
Task Description: For each sales order, predict the sales office responsible for managing sales activities for the relevant products and geographic region.
Evaluation metric: MRR
sales-group
Task Description: For each sales order, predict the sales group, i.e. the subdivision within the distribution chain that handles the customer and transaction.
Evaluation metric: MRR
sales-payterms
Task Description: For each sales order, predict the customer payment terms (payment deadlines/discounts).
Evaluation metric: MRR
sales-shipcond
Task Description: For each sales order, predict the shipping condition (logistics terms).
Evaluation metric: MRR
sales-incoterms
Task Description: Predict the header-level Incoterms (international commercial terms) for each sales order.
Evaluation metric: MRR