SAP-RPT-1.5 Is Now Generally Available: More Accurate Predictions, Better Explainability, and What’s Next
Predictive AI is becoming an increasingly important part of enterprise decision-making. Over the past few months, teams have been putting SAP-RPT models to work across real-world scenarios—from demand forecasting and high-volume classification to discovering patterns hidden within complex relational data.
These experiences raised an important question: How can we make predictions even more accurate while also making them easier for business users to understand and trust?
That question helped shape the next evolution of the SAP-RPT family.
We are excited to announce the general availability of SAP-RPT-1.5, bringing improved prediction accuracy, particularly for complex and high-cardinality scenarios, together with a new approach to explainability that makes predictions more intuitive for business users.
In this post, we’ll look at where SAP-RPT-1 is already creating value, what’s new in SAP-RPT-1.5, the available model variants, and what’s coming next.
Where SAP-RPT-1 Is Making an Impact
SAP-RPT-1 has already been applied to a growing range of enterprise use cases across finance, supply chain, sustainability, retail, and governance, risk, and compliance.
Combining Document AI with Predictive Intelligence
One of the most interesting applications has been the combination of Document AI and SAP-RPT-1 for high-volume, rule-intensive classification.
A great example is automated tax code determination with Clariant.
Clariant processes thousands of vendor invoices every year. These invoices are often unstructured PDFs and can be written in the local language of the supplier. Determining the correct VAT code requires consideration of several factors, including jurisdiction, supplier type, material, and place of supply.
Using Document AI to extract the relevant information and SAP-RPT-1 to predict the appropriate VAT code, the team built a working MVP in just four days and achieved 90% prediction accuracy during the pilot.
Looking ahead, the goal is to automate up to 32,000 invoices per year, with an automation rate of up to 98%, while expanding quality coverage from approximately 4% spot-check sampling to 100% of processed invoices.
Supporting Agentic Sales Order Creation
The same combination is also being explored for agentic sales order creation.
Instead of manually creating sales orders in SAP S/4HANA based on unstructured emails and documents, an AI agent can use Document AI to extract the available information and SAP-RPT-1 to predict additional fields required to complete the order.
This enables a more automated process while still leveraging the structured business context available in enterprise data.
Expanding Across Enterprise Scenarios
SAP-RPT-1 is also being applied to a broader set of business challenges:
- Payment matching: Up to 90% match accuracy without requiring historical model training.
- Working capital management: Up to 54% reduction in mean invoice approval date prediction error compared with a heavily tuned narrow AI model.
- Sustainability: Automated account and cost object determination for tens of thousands of uploaded carbon-data entries, saving several working days.
- Supply chain and logistics: Prediction of delivery delays and logistics destinations.
- Finance and procurement: Classification and prediction tasks such as payment delays and material groups.
These examples demonstrate that SAP-RPT-1 is not limited to a single business process. It is designed as a general-purpose prediction engine for relational enterprise data.
What’s New in SAP-RPT-1.5?
SAP-RPT-1.5 introduces improvements across three key areas:
Prediction accuracy, throughput, and explainability.
More Accurate Predictions for Complex, High-Cardinality Data
SAP-RPT-1.5 delivers stronger prediction performance across a wide range of classification scenarios.
On SAP’s benchmark suite covering 166 real-world classification tasks, SAP-RPT-1.5 demonstrates strong performance overall, with particularly notable improvements on high-cardinality classification problems.
This is especially relevant for enterprise tabular AI, where datasets often contain many unique values and complex relationships.
SAP also evaluated SAP-RPT-1.5 on publicly available benchmarks. On the SALT enterprise tabular benchmark, SAP-RPT-1.5 demonstrated leading performance compared with other tabular foundation models as well as per-task tuned models.
For teams using the API, the maximum number of query rows per API call has also increased from 128 to 512, making larger prediction batches easier to handle.
You can find additional benchmarking details in the SAP-RPT-1.5 model card.
Why Accuracy Improvements Matter
In enterprise scenarios, even a few percentage points of additional accuracy can have a meaningful business impact.
Better predictions can mean:
- Fewer manual corrections
- Earlier identification of risks
- Higher automation rates
- Faster processing of business transactions
- More consistent decision-making
- Reduced operational effort
Whether the use case is tax code determination, material classification, payment delay prediction, logistics assignment, or supplier delivery prediction, improved accuracy can translate directly into better business outcomes.
Explainability: Explain by Example
One of the most exciting additions in SAP-RPT-1.5 is a new approach to explainability.
Traditional feature importance can tell you which columns influenced a prediction. SAP-RPT-1.5 goes a step further by showing which historical rows most influenced the prediction.
In other words, the model can explain a prediction by example.
Imagine a sales order scenario. Instead of simply saying that material group was the most influential feature, SAP-RPT-1.5 can also identify previous sales orders that closely resemble the current transaction.
For business users, this provides a much more tangible explanation:
“This prediction is similar to these previous business cases.”
This example-based explanation is particularly useful for users who naturally reason through comparisons with past business decisions.
It can also help teams build greater confidence in AI-driven recommendations by giving stakeholders a concrete reference for why a prediction makes sense.
SAP-RPT-1.5 Model Variants
SAP-RPT-1.5 is now generally available through SAP’s generative AI hub and is priced at the same level as RPT-1.
Two variants are available:
SAP-RPT-1.5
This is the default option and was previously known as the -small model.
It provides strong performance across a broad range of use cases and is the recommended starting point for most scenarios.
SAP-RPT-1.5-large
This variant is designed for more demanding scenarios, particularly:
- Datasets with more than 100 columns
- Use cases where maximizing prediction accuracy is the primary objective
- More complex tabular prediction scenarios
More information is available in the SAP Help Portal.
What’s Coming Next: Tabular Orchestration
One of the most common questions from customers is:
How do we efficiently work with millions of historical records?
SAP-RPT models operate within a fixed context window. When a dataset contains millions of rows, selecting the most relevant historical records to provide as context becomes an important part of the prediction workflow.
Today, teams often need to manage this process themselves.
To simplify this, SAP is introducing tabular orchestration as a managed workflow on SAP’s generative AI hub, targeted for general availability in August 2026.
The concept is simple:
- Register your data table.
- Configure the prediction scenario at design time.
- Submit prediction requests at runtime.
- The workflow automatically retrieves and selects the most relevant context rows.
Tabular orchestration will also provide centralized management for model configuration and invocation, with context registration, model routing, and lifecycle management built into the workflow.
As the SAP-RPT family evolves and additional tabular foundation models become available, this orchestration layer is intended to provide a governed and repeatable foundation for enterprise predictive AI.
More details are expected closer to the September 2026 release.
Getting Started with SAP-RPT-1.5
SAP-RPT-1.5 is now available in the model library of SAP’s generative AI hub.
If you are new to the RPT model family, the RPT Playground is a simple way to explore the capabilities of the model using your own data or SAP-provided use-case samples, without requiring an initial setup.
You can explore the RPT Playground to get started.
For those ready to integrate SAP-RPT into predictive and tabular AI scenarios, our step-by-step guide walks through the process from model deployment to your first inference.
The SAP-RPT-1.5 model card is also a valuable resource for understanding capabilities, limitations, benchmarking results, and best practices for responsible adoption and governance.
Final Thoughts
SAP-RPT-1.5 represents another step toward making predictive AI more accurate, scalable, and understandable for enterprise use cases.
With stronger performance on complex and high-cardinality data, larger API batches, and the new “explain by example” capability, teams can not only generate better predictions but also provide business users with more intuitive reasons behind those predictions.
And with tabular orchestration on the horizon, the focus is shifting from simply building individual predictions to creating governed, reusable, and scalable predictive AI capabilities across the enterprise.
We’re excited to see how you use SAP-RPT-1.5 in your own scenarios.
What predictive use cases are you exploring with SAP-RPT? Share your experiences and ideas in the comments.
Read More Technology Blog Posts by Members articles
#abap