This week, we hosted the second edition of our SAP Document AI Workshop together with SAP, this time with an even broader group of participants from more than 50 companies across industries including energy, manufacturing, construction, food and consumer products, technology, and industrial services.
For us, the objective was very practical: put AI and automation to work on real business processes where manual effort is still high, there is little room for error, and the ROI can be measured clearly.
In the first edition of the workshop, we focused on showing what that can look like through a real Document AI use case, from the original manual process to the automated workflow. For this second edition, we wanted to build on that foundation and go further into the hands-on experience.
We worked through how DocAI can be applied inside an end-to-end workflow, how confidence levels and human review help manage exceptions, how SAP BTP and Integration Suite connect the process, and how to evaluate the economics before deciding where automation is worth pursuing.
We also gave participants access to the Document AI ROI Calculator developed by Inclusion Cloud, so they could leave with both a technical framework and a clearer view of the potential savings behind their own use cases.
Starting With a Simple Question: Where Is Manual Document Work Getting in the Way?
The workshop started with a problem that exists across industries: employees still spend significant time opening documents, identifying what they contain, locating specific fields, entering information into enterprise systems, and making sure the document is associated with the correct business record.
Invoices, purchase orders, sales orders, forms, and other documents are obvious examples.
At low volumes, those individual actions may appear minor. At enterprise scale, however, a process that takes only a few minutes per document can consume thousands of hours.
That was the lens used throughout the workshop: find repetitive, document-heavy processes where automation can remove manual steps without compromising the controls the business requires.
The session then placed Document AI within SAP’s broader BTP and Business AI ecosystem. SAP presented Document AI as one component designed specifically around document consumption and extraction, going beyond traditional OCR by using AI to understand the context of the information being processed.
A Key Advantage: You Do Not Always Need to Start from Scratch
One of the most practical parts of the workshop was seeing how much is already available out of the box.
SAP Document AI includes more than 30 preconfigured templates for common document types. During the session, examples such as purchase orders, payslips and I-9 forms were discussed. These can provide teams with a starting point instead of requiring them to define every extraction scenario from zero.
This became even clearer later in the hands-on portion.
Participants worked with SAP’s standard invoice schema, which already contained structures for typical invoice information such as taxes, payments, delivery details, suppliers, materials, descriptions and quantities. The schema could be enabled, copied into the user’s environment, activated and then used to begin processing invoices.
At the same time, the workshop demonstrated the other side of the equation: custom schemas.
For an unstructured document without an existing template, participants created their own schema and defined the information they wanted Document AI to extract. The exercise showed that companies can use SAP-provided structures where they already exist and configure their own extraction logic when their process requires something more specific.
From Document Extraction to an End-to-End BTP Workflow
Extracting information is only one part of automating a business process.
The workshop also walked through an anonymized enterprise scenario in which employees received large volumes of documents, manually identified the document type and relevant business reference, and then made sure each file reached the correct record in SAP. The process was mandatory for record keeping, but consumed substantial operational capacity.
The proposed architecture showed how SAP Document AI and Integration Suite can play different roles in the same workflow.
Document AI acts as the intelligence layer that reads and interprets the document. Integration Suite can orchestrate the surrounding flow: retrieving files from an input source, sending them to Document AI, receiving structured information back, and passing that information into SAP or another destination.
The architecture discussed in the session was designed to replace several manual and legacy steps with a more scalable flow while still allowing additional document types and integrations to be introduced over time.
The session also demonstrated that document ingestion does not have to begin with a manual upload. Mail, FTP and shared drives were discussed as possible entry points, while the Document AI workspace includes channels and outbound capabilities that can connect extracted information with other systems through services and APIs.
Context matters more than document position
Another important part of the hands-on exercises was the difference between contextual extraction and traditional template-based OCR.
Participants created fields and processing instructions that explained what information the system should look for. Those instructions could be applied at schema level or even to individual fields.
The SAP specialists explained that the quality of these instructions matters because DocAI can use LLM-based capabilities during extraction. Instead of defining only where a value appears on a page, teams can give the system contextual information about what the field represents and how it should be interpreted.
That is particularly relevant when a company receives the same type of business document in many different layouts. A purely location-based approach may require maintaining templates for each layout. Contextual extraction provides another way to approach that variation.
Accuracy, Confidence and Human Review
Accuracy generated some of the most interesting discussion during the workshop.
Participants asked what happens with handwritten information, low-quality scans or ambiguous fields. Rather than assuming that every extraction should be treated equally, the workflow demonstrated how Document AI uses confidence levels to indicate when information may require review.
During the hands-on exercise, participants could open an extracted document, compare the source with the structured fields, correct a value or its source when necessary, and confirm the extraction.
That feedback can then be used as part of the learning process as additional documents are processed.
The objective is not simply to remove a person from every step. For processes with little tolerance for error, automation also needs a mechanism for identifying uncertainty and routing those cases for validation.
Putting an Economic Value on the Workflow
Technical feasibility was only half of the workshop.
Participants also received access to the Document AI ROI Calculator developed by Inclusion Cloud, designed to help evaluate whether a potential use case makes economic sense.
The calculator starts with the existing process: document volume, processing time, labor costs, legacy software costs and other costs associated with handling a document. Those inputs can then be compared with the expected cost of processing the same workload using Document AI.
The anonymized scenario presented during the workshop illustrated why scale matters. A process handling approximately 8,000 documents per month, at roughly eight minutes of manual work per document, represented around 1,000 hours of operational work each month.
The point of the calculator was not simply to show that automation saves money. It was to help teams answer a more useful question: which processes have enough volume and cost to justify the investment?
The workshop emphasized that not every document workflow should be automated with the same technology. The economics change according to volume, pages processed, existing labor and software costs, and the amount of work that can realistically be removed.
That gives business and technology leaders a way to prioritize use cases based on measurable impact rather than adopting AI simply because the capability exists.
The questions quickly moved from demos to production
One of the strongest parts of this second edition was the level of discussion from participants. As soon as the architecture and use case were introduced, the questions moved into the issues teams encounter when considering a real implementation.
Some of the Q&A that came up during the workshop:
Can Document AI process documents in different languages?
Yes. We discussed examples including Mandarin, Japanese, and Arabic. Document AI can extract information from documents in different languages. If translation is needed, that is a separate step that can be added to the workflow using other BTP capabilities.
Is Document AI only for S/4HANA?
No. Document AI is a SaaS service and is not limited to S/4HANA. It can be used as one component in a broader architecture, with BTP and Integration Suite added when orchestration or integration is needed.
Can it process scanned PDFs, JPEGs, photos, or content coming from email?
Yes. During the session we covered scanned PDFs, JPEGs, photos, and email-based inputs as possible sources for document processing.
What happens if the image quality is poor or the system is not confident about a field?
Document AI lowers the confidence score when the image quality, context, or extracted field is ambiguous. Those cases can then be reviewed and corrected before being confirmed.
Do we need to build every document schema from scratch?
No. SAP provides more than 30 preconfigured templates, including common documents such as invoices and purchase orders. For documents that are more specific to the business, teams can create custom schemas and define their own fields and processing instructions.
What format does Document AI return, and can the data be sent to other systems?
The extracted information is returned as JSON. We also showed how Integration Suite can take that structured data and send it to SAP, a database, or another downstream system.
How is document retention handled?
Retention is configurable and should follow the company’s own policy. During the discussion, examples such as 30- and 90-day retention periods were mentioned.
What Participants Took Back to their Teams
By the end of the session, participants had moved through both sides of a Document AI initiative.
On the technical side, they had seen the position of Document AI within the SAP ecosystem, worked with both custom and preconfigured schemas, defined extraction fields and instructions, processed documents, reviewed confidence levels, corrected results, and explored how Integration Suite and BTP can extend the workflow.
On the business/economical side, they had a framework for identifying repetitive document processes, quantifying the current workload, and evaluating the potential ROI before deciding which use cases deserve further investment. That combination of hands-on experience and economic assessment was one of the explicit objectives carried forward from the first edition of the workshop.
Turning the Pressure to Adopt AI Into Value
A big thank you to everyone who joined this second edition, with participants from more than 50 companies across different industries, and to the SAP experts and speakers who shared their experience throughout the workshop. We really appreciated the opportunity to organize this session in partnership with SAP and have such an active group in the room.
If your team is being asked to show progress with AI, one of the hardest questions is usually where to start.
Document AI has one advantage that makes that conversation easier: the value is relatively straightforward to measure.
You already know how many documents your team processes, roughly how much time each one takes, and what that work costs today. That gives you a baseline. From there, you can build a PoC, test the automation on a real process, and come back internally with something much more useful than a demo: actual numbers.
And when you are trying to get an AI initiative approved, that matters.
As an SAP partner with experts in BTP, Integration Suite, and AI solutions, we can help you identify a good first use case, build the PoC, and quantify the expected ROI before you take it to a broader rollout.
If you are considering Document AI and want to see what a PoC could look like for one of your own processes, book a call with us.
