Arcplace has been working with documents and document-based processes for 20 years. Everything started with classic scanning and text recognition using OCR. A few years ago, machine learning methods for document recognition, classification and data extraction were introduced. This made it possible for the fi rst time to automatically structure document intake.
Over the past two years, Arcplace has been intensively focusing on generative artificial intelligence in the context of document processing. The focus is no longer solely on recognising patterns, but on understanding content and relationships. Today, systems do not merely interpret documents; they understand their context and the connections that emerge from it.
Machine Learning Automation with Limitations
Machine learning recognizes patterns and relationships, but not context. For a system to recognise and process invoices, contracts or other document types, it requires large volumes of training documents that must initially be manually validated to ensure quality. Every new language, every layout and every invoice variation results in additional training effort. Despite this effort, classification results, meaning the recognition of document types, often reach only around 80 to 85 percent. The smooth processing of complex invoices remains between 20 and 30%.
Changes in document design quickly lead to errors and require retraining. These systems operate efficiently, but they do not understand context, and modified document structures must be laboriously retrained. Generative AI works fundamentally differently. It has been pretrained on vast amounts of data and already understands a wide variety of document types, industry terminology, layout formats and phrasing. It no longer requires a manual training phase. Instead of searching for positions, it understands content and interprets what it means. It knows what an invoice number is, recognises payment terms even when expressed in free text, and understands that “payable within 30 days” means the same as “net 30 days”.
Arcplace Thinks Ahead: AI Reviews, Compares and Understands
Generative AI opens up new possibilities that go far beyond simply extracting data from individual documents. Entire dossiers can be analysed intelligently. In banking credit processes, for example, generative AI can automatically verify whether the information in applications, identity documents and salary statements is consistent and plausible. It can recognise whether names match, whether income has been transferred correctly, or whether documents contain contradictory information. What is still often done manually and with great effort today can be supported more efficiently, transparently, and in an automated way in the future. There is also significant potential in handling unstructured customer correspondence. Generative AI can recognise emotions, urgency and intent within a message and automatically generate response suggestions that comply with an organisation’s communication guidelines.
This marks the next phase of input management: AI as an intelligent assistant that not only delivers data, but understands it and reacts accordingly
What Does This Mean for Arcplace Customers?
Generative AI transforms input management not only technologically, but economically as well. Customers benefit on multiple levels:
- No training effort required, and therefore no timeconsuming compilation of sample data by the customer
- Reduced project effort, leading to shorter project timelines and lower project costs
- Higher automation rates thanks to straight-through processing of over 80 percent, resulting in lower operating costs due to reduced manual validation
- More stable quality, even with new document layouts, including the detection of irregularities
Conclusion: Lower costs, greater efficiency, higher quality and a system that thinks along instead of merely processing.
