Skip to Content
Product Information
Author's profile photo Daniel Stoerkel

Announcement: Invoice Object Recommendation Becomes Part of Data Attribute Recommendation (Part 1)


In a world where large volumes of data need to be processed and managed by organizations, the SAP AI Business Services portfolio comes in handy. By applying machine learning algorithms, these AI-powered services available on SAP Business Technology Platform can (semi-)automate and optimize manual, repetitive and error-prone tasks, and therefore realize more intelligent business processes. 

In this blog post, let’s focus on two of these services: Invoice Object Recommendation and Data Attribute Recommendation
You will get to know how both services are becoming one as of December 2021, and why this is beneficial for SAP customers.  


What is Data Attribute Recommendation 

Data Attribute Recommendation (part of the SAP AI Business Services portfolio), supports with the creation, maintenance and management of structured data by predicting and classifying categorical entities, e.g.  product categories, and predict numeric values such as ratings and numbers, using text, numbers, and categories as input. It can be used to enrich missing attributes of data records and classify incoming hierarchical information through state-of-the-art automation workstreams and processes. 

It is a generic classification and regression service that applies machine learning to all data records independent from the business problem. Some common use cases are: 

  • Get suggestions of material class and its characteristics when creating new material requests 
  • Get international trade commodity code predictions when adding a new product 
  • Solve master data inconsistencies 
  • Obtain a price estimation for a product based on its description 


What is Invoice Object Recommendation 

Invoice Object Recommendation follows the same logic, with a stronger focus on procurement and finance use cases. Invoice Object Recommendation proposes G/L (general ledger) accounts, cost centers and profitability analysis dimensions (also known as CO-PA dimensions) for incoming invoices without a purchase order reference. 

This is a specific business-related classification service for financial data that applies machine learning to classify invoice line items. Common use cases are: 

  • Prediction of finance related invoice objects where there is no purchase order attached 
  • Prediction of invoice objects in the material management area where no purchase order can be found (and therefore also no invoice objects can be pre-filled) 


Merging Invoice Object Recommendation with Data Attribute Recommendation 

As SAP continuously aims to enhance the performance of its services offered to its customers and given the fact that both services are enhancing and complementing each other perfectly, SAP has decided to combine these AI capabilities and their business focus within one common SAP AI Business Service going forward. The above explained functionality of Invoice Object Recommendation continues to be available as a business blueprint template* in Data Attribute Recommendation. Customers who want to automate use cases – which have been formerly implemented using Invoice Object Recommendation – can now use the Invoice Object Recommendation business blueprint available within the Data Attribute Recommendation service. 

It leverages the Data Attribute Recommendation extended capabilities which will directly benefit SAP customers. Customers can now bundle resources and head for a one stop shop for all their classification and regression tasks. The service offers full model lifecycle and more flexibility in running and training models simultaneously. Further details on benefits will be shared later in this blog. 

The service’s functionality will continue as it is, just not as an independent solution. However, there are small but significant changes behind the scenes in data and model management that will greatly benefit SAP customers without reinventing the wheel. 

In this article, these changes are explained in more details, including explanations on how the data and model trainings are handled and how the APIs endpoints interact with each other. 


Benefits from Using Data Attribute Recommendation Capabilities 

As mentioned before, Invoice Object Recommendation users will be able to experience a broad range of benefits by using the Invoice Object Recommendation business blueprint within Data Attribute Recommendation. Please find a short summary of those benefits below: 

  • One single solution for classification and regression tasks  
  • Full model lifecycle  
    • Multiple models at the same time 
    • Models can be deleted at any time 
    • On demand deployment/un-deployment 
    • Flexible number of predictions as output 
  • Flexible input/output structure 
    • Not restricted to G/L account & cost center only 
    • Custom features can be taken into consideration 
    • No currency conversion as mandatory pre-processing step for a customer is needed any longer 
  • Single DAR instance can be used for multiple use cases (when creating multiple dataset schemas) 
  • Train multiple models simultaneously 
  • Using more generic service that would fit multiple use cases 
  • Lower price 
  • Increased file upload size (single upload instead of batch uploads) 
  • Classify and predict a flexible number of invoice objects (such as General Ledger accounts, Cost Centers,…) 


Roadmap and Future Releases 

We are also continuously enhancing the Data Attribute Recommendation service with additional features and performance optimization. Besides that, the service is also capable of solving regression tasks, there are numerous innovations planned for the upcoming year of 2022 which can be found in the SAP Roadmap Explorer. 



As the Invoice Object Recommendation business blueprint is part of Data Attribute Recommendation, now the Data Attribute Recommendation pricing applies. We are happy to share that the Data Attribute Recommendation service is cheaper to run than the Invoice Object Recommendation service. The table below shows and compares the earlier pricing of Invoice Object Recommendation with the current one of Data Attribute Recommendation: 

  Invoice Object Recommendation** Data Attribute Recommendation**
Metric: Blocks of 1000 records  Up to 14 blocks, 250 €  Up to 5 blocks, 320 € 
From 14 blocks, 188€  Up to 100 blocks, 64 € 
Up to 250 blocks, 32 € 
Up to 500 blocks, 19 € 
Up to 750 blocks 13 € 
From 750 blocks, 10€ 
Models/additional deployed models  N/A  0.91/hour (if more than 3. No limit to trained, but undeployed, models) 

Feel free to use the Pricing Estimator to simulate the cost associated to your future usage. 

Read our technical blog to learn more about the technical implications of this merge. 

Please use our dedicated Q&A section in SAP Answers to ask questions about both DAR and IOR going forward 

Visit our SAP Community page

Read our product documentation 


*) A business blueprint is an entity that encapsulates the business logic of a specific use-case and is responsible for mapping a business problem to a ML task 

**) as of 2021 

Assigned Tags

      You must be Logged on to comment or reply to a post.
      Author's profile photo Abhishek Chowdhury
      Abhishek Chowdhury

      Having spent a week on data attribute recommendation i feel the tool does a decent job of predicting the labels, however the following questions still linger in my mind:

      1. Choice of machine learning algorithm must be delegated to business and not automated. There are situations where a precision is preferred over accuracy or over recall.

      2. I see DAR is giving me important attributes like precision, recall but the confusion matrix is missing.

      3. What about correlation between independent variables? Is it possible to get such statistical inferences from the data?

      Selecting the machine learning models and getting the predicted labels is pretty much the last step. There are lot of intermediatory steps pertaining to statistics.

      I see there are 3 ways SAP is addressing Machine Learning

      1. SAP HANA cloud - I see that you need a script server to call the Predictive Automated Libraries (PAL and APL). Not available in free trial.

      2. SAP Analytics cloud - Can i use existing out-of-the box algorithms such as random forest, decision tree and logistic regression?

      3. SAP Data Intelligence Cloud - How is this different to HANA cloud and SAP Analytics cloud?

      Of these 3 tools in SAP, which tool should I use to: -

      1. Leverage existing predictive analytics APIs in SAP?

      2. Analyze the data myself, pick the ML model of my choice. Iterate through the results and determine which ML model best suits the business need.


      Your insights would be very useful to understand and begin the learning process. Thanks in advance.

      -Abhishek chowdhury

      Author's profile photo Sudarshan Pavanje
      Sudarshan Pavanje

      Hi Abhishek, Let me try answering your questions

      Choice of machine learning algorithm must be delegated to business and not automated. There are situations where a precision is preferred over accuracy or over recall.

      Yes I agree with you, but the main end user persona that we are targeting are the business users or developers who would have the domain knowledge but need not be an expert on the data science side. So we internally handle most of the pre-processing , validation , split logic .

      2. I see DAR is giving me important attributes like precision, recall but the confusion matrix is missing.

      Yes, as we are targeting business users , I am not sure if confusion matrix would help especially when there are more number of classes.

      3. What about correlation between independent variables? Is it possible to get such statistical inferences from the data?

      Yes, in general it is possible to compute the pairwise correlation between independent variables, but this is not provided currently as it is not a common requirement for our target persona