SAP-RPT-1: The first table-based AI model
Reading time: 3 min
Comparing payroll tables or predicting staff turnover from employee data – just let ChatGPT do it? It’s not quite that easy, because many Gen AI applications are text-based. However, SAP systems in particular contain huge amounts of tabular data. So how can this data be analyzed using AI? SAP provides the answer: with SAP-RPT-1 – a new table-based AI model. But what is behind the model and how can companies use it? Our Team Lead SAP AI, Maximilian Spranz, explains.
What is SAP RPT-1?
SAP-RPT-1 is a pre-trained relational foundation model (Relational Pre-Trained Transformer). This sophisticated AI model was developed to generate precise predictive insights from structured business data.
The core principle of SAP-RPT-1 is in-context learning. This enables immediate, reliable predictions without the time-consuming process of model training. Users simply provide sample data sets (context rows) in their calls.
The model uses this context to recognize patterns and solve a variety of tasks, including classification and regression tasks. This includes typical scenarios such as churn prediction, demand or sales forecasting and other business-relevant analyses.
The pre-trained relational architecture of SAP-RPT-1 provides an integrated understanding of tabular data structures, relationships and business logic. The model is resilient to data quality issues and can easily handle incomplete or changing business data and variable values. A major advantage is that it eliminates costly and time-consuming model training and manual pre-processing or feature engineering.
What distinguishes SAP RPT-1 from classic ML models?
The biggest difference lies in the efficiency and adaptability due to in-context learning (ICL) and the type of data processed. According to SAP , the prediction quality is 3.5 times higher than with LLMS models.
What requirements must be met for implementation?
The use of RPT-1 requires a modern and standardized SAP landscape. As RPT-1 is consumed as a cloud service via the SAP AI Foundation, the SAP Business Technology Platform (BTP) is the central technical enabler.
Success factors for SAP RPT 1 implementation
- Define clear objectives and KPIs for each use case.
- Connect HR, IT and data analytics.
- Involve your data protection and compliance team to ensure GDPR and EU AI Act compliance
- Ensure a clear data model.
SAP RPT 1 implementation steps
- Use case definition: Select the specific prediction problem (e.g. churn prediction).
- Data extraction & governance: Extract historical data from S/4HANA or SuccessFactors to BTP in compliance with data protection guidelines. Based on this, you evaluate data quality, governance and technical connection.
- API call: The application sends the data to RPT-1 in two parts via the API:
– Context samples: A small sample of the historical data set (e.g. 100 rows).
– Integration: The prediction results from RPT-1 are fed directly back into the target application (e.g. CAP application in the BTP). - Prediction data: The current data rows for which a result is required.
What SAP RPT 1 use cases are there in the SAP HCM area?
The HR sector is an ideal use case for RPT-1 due to the abundance of structured but sensitive data.
Prediction of the risk of employee churn (Employee Churn Prediction)
The risk of churn can be predicted using RPT-1. The goal: early identification of employees who want to leave the company. From historical information on compensation developments to performance review scores, or the duration in the current position to organizational change, you can use a lot of structured data. Based on this, RPT-1 provides a precise probability assessment for all employees. This allows HR to proactively initiate intervention measures such as salary adjustments or mentoring programs.
Recruiting precision and suitability forecast
Another important area is recruiting precision and suitability prediction. This involves predicting the long-term suitability and performance of candidates based on the initial process data. Anonymized data from successfully hired and non-hired individuals is used as a database. This includes recruiting metrics, assessment scores and the correlation to later performance scores of similar employees who have already been hired. Finally, RPT-1 provides a suitability score that enables a more objective pre-selection of candidates and reduces incorrect hires.
Conclusion: SAP-RPT-1 revolutionizes predictive analytics
Are you struggling to analyze an enormous amount of structured data? Then SAP-RPT-1 offers a way to analyze your tabular data precisely, contextually and with 3.5 times higher quality than LLMS models and to predict trends – from predicting the risk of churn to the suitability of applicants. You can compare the use of SAP-RPT-1 with the use of a highly gifted tutor who takes on a completely new subject area without any preparation: You only have to show the teacher (RPT-1) two to three solved example tasks (Context Rows) from the current test (Query Row). Based on these examples, the teacher immediately understands the topic, the rules and the associated patterns (in-context learning) and can immediately provide you with the solution to the unknown task (prediction) without having to complete weeks of study (model training) beforehand. What is needed for this? A targeted use case and a clear data model.
