SAP and DataRobot: Elevating Bill Processing with Anomaly Detection and Generative AI







SAP and DataRobot are taking their partnership to new heights by strengthening their collaboration by means of the mixing of predictive and generative AI capabilities. Now we have developed a cutting-edge partnership that may empower clients to generate worth with AI by seamlessly connecting core SAP BTP with DataRobot AI capabilities.  

For example, let’s discover how organizations can harness the ability of predictive and generative AI to streamline bill processing providing a quicker, extra correct and cost-effective various to handbook assessment and validation.

The Enterprise Downside

Proper now corporations of all sizes grapple with a typical problem:  the relentless inflow of invoices.  The substantial quantity of monetary documentation might be overwhelming, typically necessitating a military of workers devoted to handbook assessment and validation.  Nevertheless this strategy shouldn’t be solely time-consuming and expensive, but additionally vulnerable to human error, making it a fragile hyperlink within the monetary chain.  

Harnessing the potential of AI is extra vital than ever earlier than.  Companies can make use of predictive AI fashions to study from historic bill information, acknowledge patterns, and mechanically flag potential anomalies in real-time.  This not solely accelerates the validation course of but additionally considerably reduces the margin of error, stopping pricey errors. Moreover, the mixing of generative AI permits for the concise summarization of detected anomalies, bettering communication and making it simpler for groups to take swift and knowledgeable actions.

SAP and DataRobot Built-in AI Resolution

This AI utility enhances bill processing by means of a mixture of a predictive and generative AI to determine irregularities amongst invoices and to speak the problems across the invoices.

  • Leverage Predictive AI mannequin for anomaly detection.
    • Enterprise perspective: Anomaly detection may also help determine irregularities, similar to incorrect quantities, lacking data or uncommon patterns, earlier than processing funds.
    • Implementation: Practice the mannequin utilizing historic bill information to acknowledge patterns and typical bill traits.  When processing new invoices, the AI mannequin can flag potential anomalies for assessment, lowering the chance of errors and fraud.
  • Generative AI Summarization:
    • Enterprise perspective: After figuring out anomalies, it is very important talk the problems to the related crew members.  Conventional reporting strategies could also be wordy and time-consuming.  Generative AI may also help interpret and summarize the detected anomalies in a concise and human-readable format.
    • Implementation: Leverage a LLM to generate an explanatory abstract of the detected anomalies.  The AI mannequin can extract key data from the anomaly detection outcomes and supply a transparent and structured narrative that summarizes the detected anomalies and the explanations to be thought of anomalies, making it simpler for analysts and managers to know the problems. 

Structure and Implementation Overview

To attain these targets, our platforms make use of varied integration factors, as illustrated within the structure graph beneath:

Graph 1. Architecture overview for the SAP - DataRobot Integrated Solution
Graph 1. Structure overview for the SAP – DataRobot Built-in Resolution
1. Information preparation and ingestion 

Bill information is ready and parsed in SAP Datasphere / HANA Cloud.  DataRobot accesses and ingest this information from HANA Cloud by means of a JDBC connector.

Graph 2. DataRobot access to create a JDBC connector with SAP HANA.
Graph 2. DataRobot entry to create a JDBC connector with SAP HANA.
2. Function engineering and predictive mannequin coaching

DataRobot  engineers options and conducts experiments with the bill information set, permitting you to coach anomaly detection fashions that excel at recognizing invoices with irregular or irregular data.  The strategy you select might be tailor-made to your particular information state of affairs—whether or not you may have labeled information or not.  You’ve got choices to handle this problem successfully, both with a supervised or an unsupervised strategy.

On this case, we utilized historic data that had been categorized as anomalies and non-anomalies.  After information ingestion, DataRobot runs an intensive information exploratory evaluation, identifies any information high quality points, and mechanically generates new options and related characteristic lists.   With that prepared, we have been capable of conduct a complete evaluation by means of 64 distinct experiments in a brief time period.  Consequently, we have been capable of pinpoint the top-performing mannequin on the forefront of the leaderboard.  This strategy allowed us to pick the simplest predictive mannequin for the duty at hand.  

Graph 3. DataRobot Leaderboard highlighting the best performing model.
Graph 3. DataRobot Leaderboard highlighting the most effective performing mannequin.

Inside every of those experiments, you may have the chance to totally assess and gauge their efficiency.  This evaluation gives helpful insights into how every predictive mannequin leverages the options inside your bill to make correct predictions.  To facilitate this course of, you may have entry to an array of instruments, together with raise charts, ROC curve, and SHAP prediction explanations, which estimate how a lot every characteristic contributes to a given prediction. These insights supply an intuitive means to achieve a deeper understanding of the mannequin’s conduct and their affect of the bill information, guaranteeing you make well-informed choices.

Graph 4. This Lift Chart depicts how well the model segments the target population and how capable it is to predict the target, letting you visualize the model’s effectiveness.
Graph 4. This Carry Chart depicts how nicely the mannequin segments the goal inhabitants and the way succesful it’s to foretell the goal, letting you visualize the mannequin’s effectiveness.
Graph 5. SHAP Prediction Explanations estimate how much a feature contributes to a given prediction, reported as its difference from the average. In this example how the delivery Date, shipping and gross amount had an impact.
Graph 5. SHAP Prediction Explanations estimate how a lot a characteristic contributes to a given prediction, reported as its distinction from the common. On this instance how the supply Date, delivery and gross quantity had an affect.
3. Mannequin deployment

As soon as we determine the optimum predictive mannequin, we transfer ahead to transition the answer into manufacturing.  This part seamlessly merges our predictive and generative AI strategy by orchestrating the deployment of an unstructured mannequin inside DataRobot.  This deployment harmonizes the predictive AI mannequin for anomaly detection with a Giant Language Mannequin (LLM), which excels in producing textual content to speak the predictive insights.  Alternatively, you may have the flexibleness to deploy predictive AI fashions instantly inside SAP AI Core, providing a further route for operationalizing your answer.

The LLM summarizes the rationales linked to every prediction, making it readily digestible to your monetary evaluation wants. This versatile deployment technique ensures that the insights generated are accessible and actionable in a way that fits your distinctive enterprise necessities. 

Two easy python information simply orchestrate this integration by means of easy capabilities and hooks that can be executed every time an bill requires a prediction and its consecutive evaluation.  The primary file named, has the credentials to attach with GPT 3.5 by means of Azure and comprises the immediate to summarize the reasons and insights derived from the predictive mannequin.  The second file, named, simply orchestrates the entire predictive and generative pipeline by means of a number of easy hooks.   Yow will discover an instance of the way to assemble customized python information for unstructured fashions in our github repository.  

You’ve got the potential to check and validate this unstructured mannequin prior its deployment, assuring that it persistently produces the supposed outcomes, freed from any operational hitches.  

Graph 6. Validation of the unstructured model before deployment.
Graph 6. Validation of the unstructured mannequin earlier than deployment.
4. Enterprise Software

As soon as the deployment is formally in manufacturing, an accessible API endpoint turns into your bridge to attach with the deployment, seamlessly producing the exact outcomes you search in SAP Construct. 

Graph 7. SAP Build Workflow that includes a module to connect with the deployment of DataRobot via API.
Graph 7. SAP Construct Workflow that features a module to attach with the deployment of DataRobot by way of API.

Subsequent, we craft a enterprise utility for bill anomaly detection inside SAP Construct.  This utility retrieves the predictive and generative output by way of API integration and provides a user-friendly interface.  It presents the ends in a sensible and intuitive method, guaranteeing that monetary analysts can effortlessly add invoices in PDF format, simplifying their workflow and enhancing the general person expertise.  

Graph 8. SAP Build Workflow for the invoice approval business application.
Graph 8. SAP Construct Workflow for the bill approval enterprise utility.
Graph 9 - Final output generated in the business application for financial analysts to approve or reject an invoice based on the anomaly prediction and the corresponding LLM summary.
Graph 9. Remaining output generated within the enterprise utility for monetary analysts to approve or reject an bill primarily based on the anomaly prediction and the corresponding LLM abstract.
5. Manufacturing Monitoring

DataRobot maintains an oversight over the generative AI pipeline by means of the utilization of customized efficiency metrics and predictive fashions.  This rigorous monitoring course of ensures the continual reliability and effectivity of our answer, providing you a seamlessly reliable expertise.   

Graph 10. DataRobot deployment containing the predictive and generative pipeline properly monitored over time with relevant custom metrics.
Graph 10. DataRobot deployment containing the predictive and generative pipeline correctly monitored over time with related customized metrics.


In abstract, the partnership between SAP and DataRobot continues to permit organizations to rapidly drive worth from their AI investments, and now much more by leveraging generative AI.  Predictive anomaly detection and generative AI can remodel the challenges and dangers related to bill processing.  Effectivity and accuracy soar, whereas communication turns into clearer and extra streamlined.  Companies can now modernize their operations, save time and cut back errors.  It’s time to unlock the potential of this transformative know-how and take your operations to the following stage. 

Free trial

Expertise the DataRobot AI Platform

Much less Friction, Extra AI. Get Began Right now With a Free 30-Day Trial.

Signal Up for Free

Concerning the writer

Belén Sánchez Hidalgo
Belén Sánchez Hidalgo

Senior Information Scientist, Group Lead and WaiCAMP Lead, DataRobot

Belén works on accelerating AI adoption in enterprises in the USA and in Latin America. She has contributed to the design and improvement of AI options within the retail, schooling, and healthcare industries. She is a pacesetter of WaiCAMP by DataRobot College, an initiative that contributes to the discount of the AI Trade gender hole in Latin America by means of pragmatic schooling on AI. She was additionally a part of the AI for Good: Powered by DataRobot program, which companions with non-profit organizations to make use of information to create sustainable and lasting impacts.

Meet Belén Sánchez Hidalgo


Supply hyperlink

Share this


Google Presents 3 Suggestions For Checking Technical web optimization Points

Google printed a video providing three ideas for utilizing search console to establish technical points that may be inflicting indexing or rating issues. Three...

A easy snapshot reveals how computational pictures can shock and alarm us

Whereas Tessa Coates was making an attempt on wedding ceremony clothes final month, she posted a seemingly easy snapshot of herself on Instagram...

Recent articles

More like this


Please enter your comment!
Please enter your name here