Let your DMS work for you.
– Thanks to artificial intelligence

With regard to ECM systems, development focuses on machine learning (ML) methods as a subfield of artificial intelligence. Our ML methods automatically learn models of the input data in the supervised learning process. In this process, the algorithm is shown sample data – in this case, business documents – for which the desired answer is known. The algorithm calculates “its” current answer based on a mathematical model. If this is incorrect, the model is adapted in such a way that the error is reduced in the following run. The mathematical representation of the model makes it possible to decide, for example for the text of a document, whether it is an invoice or a delivery bill. The most important difference to classical, rule-based methods is the fact that the algorithm “learns” the model fully automatically and exclusively on the basis of given sample data. If sufficient training data is available, the entire training process can be automated. The manual creation and maintenance of complex rule sets is completely eliminated.

Our machine learning techniques learn models from sample data: Business documents are translated into an abstract representation. This representation is used to create a mathematical model of the documents that enables predictions about the documents. The learning process involves reviewing many thousands of examples for which the desired answer is known. At each step, the process makes a prediction. If this is not true, the model is adjusted to reduce the error in the future – the procedure learns. For this to work, the algorithm needs sample data, lots of sample data. Here, you benefit directly from the d.velop platform: Due to the large amount of completely different data that you and other users process every day with our solutions, our ML processes “see” a large variety of examples every day, which help them to become better. You benefit directly from this, because better recognition performance automatically leads to more time savings for you.

Is the machine replacing the human being? We strongly believe in the performance of highly skilled workers. Therefore, our goal is to support you, not replace you. We relieve you of tedious, repetitive work steps. In order to give you more time for the essentials – the things that the computer can’t simply do for you. For example, we provide you with an “AI colleague” who reviews document inventories for you in advance, makes recommendations on how to file the documents and helps you to transfer essential information from the documents to the system more quickly. This often works fully automatically. But you retain control at all times and can overrule the AI.

My document obviously says “invoice”. So why is it not recognized as an invoice? – Our procedures see very many, very different examples. The volume of documents is so heterogeneous that it is not possible to create or derive simple rules that, for example, assign a document to a certain category, even if this should be obvious at first glance. Sometimes this leads to “wrong decisions” that are difficult to understand, as in the example above. A document that succinctly contains the word “invoice” is not recognized as an invoice. This can have a simple reason. Many documents contain the term “invoice” but are not; order confirmations and reminders, for example, often refer to a future or previously issued invoice. To rule out systematic errors, we continuously monitor the recognition performance of our processes and intervene where necessary to maintain maximum reliability for you.

When you use AI-powered services, your documents and the metadata you store about them are used to further train the models. This improves recognition performance – completely automatically and without you having to do anything. For this to work, we have to temporarily cache some of your data. But this only happens for as long as is absolutely necessary. Of course, this fact only affects our AI services – in our document repository, your data will continue to be stored securely for the long term.

Your data is stored and transmitted exclusively in encrypted form. We treat your documents per se as personal data. In this respect, the principle of data economy underlying the GDPR applies to us. Your data will not be kept longer than necessary to provide our services. In particular, we use upstream mathematical transformations to ensure that the final AI model does not allow any conclusions to be drawn about the input data. In this way, your data is secure – and remains so even with extensive AI use.

AI is currently a very active field of research. Innovation cycles are measured in months – sometimes only in weeks. Today’s methods make it possible to make statements that were considered unthinkable just five years ago. At the same time, current research results find their way into widespread application extremely quickly. Every day, we use our know-how to make these advances directly useful for you. In this way, we continuously improve our models. Which leads on the one hand to more automatically recognized information and on the other hand increases the recognition accuracy.