Introducing Document Information Extraction Premium Edition - Unleashing the Power of Large Language Models and Generative AI

[tomasz_janasz]

Updated: 03/2025
Quick access to assets

• Get a glimpse of the new features in only 25 min: Jump-Start Your Document Processing with Gen AI.
• A customer reference with 6M processed documents: City of Hamburg. 
• 2025 pricing update: introduction of consumption tiers and free tier availability.
• Best practices and hints for optimizing prompts with Premium Edition.
• Dedicated step-by-step tutorials to kick start on SAP BTP Trial.
• Check also SAP Discovery Center, SAP Help Portal, BTP Service Description Guide (section 7)

---

Foundation Models, a technological breakthrough in AI, have ushered in a new paradigm with disruptive capabilities. The rise of applications like ChatGPT, built on these foundational models, has rapidly caught the world's attention. Particularly in the context of intelligent document processing (IDP) it allows to extend the pre-defined capabilities with emerging capabilities that have been either unforeseen or very difficult to implement.

Our product lineup for IDP: Document Information Extraction, premium edition is a groundbreaking addition to our software suite that is set to revolutionize the way businesses extract valuable insights from unstructured data. With the power of Large Language Models (LLMs), the premium edition has unlocked a whole new level of capabilities, enabling organizations to streamline their operations and further increase their process-related efficiency. The premium features in the premium edition encompass:

Auto Extraction of Unstructured Data
Gone are the days of laborious annotation and template creation. The premium edition will offer a schema-based extraction of unstructured data by leveraging the unprecedented capabilities of LLMs. With a simple description of required fields, the solution will automatically extract and organize data, eliminating the need for manual intervention and drastically reducing time-to-value. See demo 1, demo 2, demo 3 and documentation.

Extended Language Support
In an increasingly interconnected world, language barriers should never hinder adoption. That's why we will introduce extended language support for over 65 languages. Businesses will be able to effortlessly extract information from documents written in different languages, enabling faster go-to-market strategies, and expanding global coverage of their IDP functions. See documentation.

Extensions of SAP Standard Schemas
To cater to the diverse needs of our customers, the premium edition will extend the support for SAP Standard Schemas. This enhancement will allow for a seamless integration with SAP systems and their data models, providing quick extensibility and higher business value. See documentation.

Instant Learning Framework
The premium edition will also come with immediate improvements in accuracy based on feedback data from users. By harnessing the power of LLMs and GenAI, the solution will ensure better accuracy in data extraction, minimizing errors and maximizing the reliability of extracted information. This will lead to higher automation rates and hence, greater operational efficiency and productivity. See demo and documentation.

Document Classification
The premium edition lets you use the LLM capabilities to classify your incoming documents smoothly with embedded classification capabilities. See demo and documentation.

The benefits of Document Information Extraction, premium edition are far-reaching. By drastically reducing time-to-value, businesses will be able to quickly extract insights from unstructured data. With the global coverage and faster go-to-market strategies, companies can stay ahead of the competition. Additionally, the quick extensibility and higher business value provide a solid foundation for further growth and innovation.

Document Information Extraction, premium edition represents a significant stride forward in our commitment to providing relevant, reliable, responsible business AI that delivers real-world results. Let's embark on this adventure together! Feel free to connect with us.

Disclaimer:
The intellectual property belongs to SAP and the content is copyrighted. We kindly remind you that the roadmap-related statements are not a commitment and are subject to change.

---

Learn more
Read more about the news of Document Information Extraction on the help portal!

What is Document Information Extraction?

Document Information Extraction is one of the SAP AI Business Services on the SAP Business Technology Platform (SAP BTP). This ML-enabled service is available through the Cloud Platform Enterprise Agreement (CPEA) and also in the Pay-As-You-Go (PAYGO) model.

Tutorials & Learnings

• See all available tutorials here!

• OpenSAP course for SAP AI Business Services

Blog posts:

• Simplify Business Document Processing with SAP AI Business Services

• Free Tier Option for Document Information Extraction Premium Edition

• Business document processing through standard SAP integration

SAP Community Page:

• SAP AI Business Services

[gaurang_gujar]

Hi Tomaz,

 

That is a great news , I believe that will game changing in the world of OCR with immense capabilities.

 

Will DOX Premium edition will also be a part of  SAP Build Process Automation Service License ?

 

Regards,

Gaurang

[tomasz_janasz]

Hi Gaurang,

this is a commercial aspect that I cannot comment on yet. Please note, that GenAI is an emerging technology and we are currently building up know-how around the commercial and legal implications of it specifically in the enterprise context.

We will keep the community posted in that regard in the upcoming months.

Best regards,
Tomasz

[AlexDong]

hi tomasz.janasz,

Exciting news! I instantly tried with trial account but it seems the document cannot be uploaded, even with several browsers..

Is there any limitation here?

Thanks and best regards,

Alex

[tomasz_janasz]

HI Alex,

the "pending" status might be misleading. You need to run through the wizard and the upload happens as the last step.

Regards,
Tomasz

[AlexDong]

Hi Tomasz,

Exactly. Now I got it.

Thanks for the reply!

Best regards,

Alex

[mcarunnie]

Hi,

I don't see the AI aspect of it. May be I'm missing something. I'm able to create template and upload documents and get the values as per the template. What is the AI part of it?

[tobias_weller]

Hi Arun,
For this new functionality, you don't need to create any templates. Instead you only define your schema with the names of the fields you would like to extract. All the rest is being taken care of by AI.

Best regards,

Tobias

[mcarunnie]

Hi Tobias,

Thanks for your response.

Yes, I realised it when I tried without using template. If I want to correct the results, I see using template provides an option.Will using 'Auto' functionality provide some flexibility to re-wire the results?

My second query is Can we make the results more accurate?

[tobias_weller]

Hi Arun,
At the moment, you can try to improve the results by adjusting your schema definition, we listed some ideas here.

But we also plan to deliver an automated improvement functionality based on user feedback (i.e. your corrections).

[mcarunnie]

Hi Tobias,

Thanks again!

I have one last question : How reliable it is if we want to derive common format but have several input formats. Our goal is to convert several input formats into one uniform format.

[tobias_weller]

Hi Arun,
That's exactly the purpose, to be able to process diverse documents without having to provide every layout upfront for training.
For exact estimations how well it works in your case, we would need to look at some of the samples or you could directly test it.

[keean_ferreira]

Hi there. I would like some further clarity around this "Premium Feature". Will we still need to manually configure every field we wish to extract? We work with Manufacturing Batch Records, they have hundreds of fields, and there are differences between each doc. Will this feature still require us to input every field we want extracted? Thanks in advance.

[tobias_weller]

Hi Keean,
You need to once configure the fields that you would like to extract. Is there any superset of fields in your scenario or a generic way of defining them so that they fit to all documents?

So far I have not encoutered a scenario as you described, so please feel free to reach out with more details.

Best regards,

Tobias

[keean_ferreira]

We could likely define generic categories, however, it would still be very time consuming and I fear that we might lose precision if we take that approach? Is this tool capable of scanning the document and extracting all information it detects?

[tobias_weller]

So far we do not offer such a functionality, but I will note it down to consider it for future enhancements. However, I'm curious how would you process the result of such a functionality, wouldn't you still need to map the extracted information into some DB table?

[keean_ferreira]

Awesome, thanks for letting me know! Yes exactly, we need to find any variances between the records, so the information would be extracted into a form that the computer can analyze, like a database, CSV, or JSON. Then the app should detect the differences between records and point them out to us for further inspection. It would be great to be able to have some graphs and charts representing the information visually as well. This is more or less what we are looking for. Would we be able to leverage foundational models within the AI Foundation playground to design an app like this ourselves?

Best regards,

Keean.

[00022111734]

Hi Tomsaz

 

Thanks for Insightful Blog well Explained

[ThomasFeyer]

Hi @tomasz_janasz,

Is there any information when the premium edition will be available in "Pay-As-You-Go for Partners" BTP accounts? According to the Discovery Center, it is not yet available and we would really love to try it out.

Thanks and Best Regards,
Thomas

[krishnam_prasanth]

Hi @tobias_weller ,

Can a user of Document Information extraction able to train or map the fields from an invoice to get the required extraction results. How can training able to do for the premium edition of DoX service.

[tobias_weller]

@krishnam_prasanth we are currently working on a feature that would allow the service to automatically learn from feedback data. This feature is planned to be released in Q3, see Roadmap Explorer.