Strategic Prediction with SAP RPT-1: A Comprehensive Guide

This tutorial uses a single local CSV file. Two prediction targets are demonstrated simultaneously in one API call.

You can access the DATASET from the GitHub repository.

• Download Full Dataset

NOTE: If you download the ZIP file, extract it and navigate to the DATA folder. Place the file in your designated location for further use.

Payment Transaction Dataset — column reference:

Rows where a target column contains [PREDICT] are the prediction rows. All other rows with known values act as context rows that teach the model the pattern.

Sample rows from the dataset:

Transaction ID    Customer Type    Industry    Region           Currency    Invoice Amount    Days Late    Risk Score
TXN AZT67X0T      Small Business   Education   Middle East      USD         30214.02          [PREDICT]    [PREDICT]
TXN SY4HFYMZ      Individual       Healthcare  Middle East      USD         1045.84           14           Medium
TXN RWO7A9Z2      Small Business   Education   Europe           GBP         17920.85          10           Low
TXN C18O8PW7      Small Business   Education   Europe           GBP         12768.45          [PREDICT]    [PREDICT]

Note: Rows with known Days Late and Risk Score values are context rows — the model learns from these. Rows marked [PREDICT] are the prediction rows the model will fill in.

This tutorial is accompanied by two Jupyter notebooks that provide a complete, executable version of all prediction steps covered in the Python SDK and JavaScript SDK sections. Download them from the tutorial’s GitHub repository before starting the hands-on sections.

Note: Replace the # placeholders above with the actual GitHub repository links once the notebooks are published. Each notebook follows the same step sequence as the corresponding section in this tutorial — credentials setup, dataset loading, context/prediction row split, model call, result parsing, and CSV export.

This step is run once before using any of the prediction methods. The same payment_transactions.parquet file is shared across Python SDK (advanced), JavaScript SDK, and Bruno — no need to create it separately for each method.

Run the following Python script from your project folder:

import pandas as pd

# Read Excel
df = pd.read_excel("payment-delay-data.xlsx")

# Convert problematic columns to string
df["Days Late"] = df["Days Late"].astype(str)
df["Risk Score"] = df["Risk Score"].astype(str)

# Save as Parquet
df.to_parquet("payment_transactions.parquet", index=False)

print("Parquet file generated successfully!")

Requirements: pip install pandas pyarrow openpyxl

Why Parquet? Parquet preserves column data types natively and is significantly more compact than CSV for large datasets. It is the required format for the /predict_parquet endpoint used by the JavaScript SDK and Bruno.

Row ordering rule: Context rows must appear before [PREDICT] rows. The script validates this and warns if the ordering is incorrect.

With the deployment running and credentials configured, you are now ready to send prediction requests.

Regardless of which access method you use, the quality of your context rows directly affects prediction accuracy.

You have now covered all four ways to work with SAP RPT-1, from the point-and-click AI Launchpad for deployment management, to the Python SDK,JavaScript SDK and Bruno REST API for full programmatic control.

Useful references:

• SAP RPT-1 Product Page
• SAP Help Portal — SAP RPT-1
• SAP Community — Step-by-Step Getting Started Guide
• SAP Cloud SDK for AI — JavaScript RPT Documentation
• SAP-samples/sap-rpt-samples on GitHub
• SAP Community — Building a Hello World with AI Core