The project involved approximately 330,000 documents and more than 220,000 items to be reviewed. The solution enabled a significant share of the work to be automated, allowing human effort to focus particularly on cases requiring interpretation and validation.

The challenge: a large and partly handwritten dataset

The purpose of the AI solution developed by Monad was to identify the correct documents within the extensive dataset and extract several pieces of information required for further processing, such as dates and information recorded in the documents.

The complexity of the task stemmed primarily from the heterogeneous nature of the material. The archive contained documents spanning several decades and based on a wide variety of document templates. Information had been entered both digitally and by hand, and even a single document could contain both.

In addition to the documents that actually needed to be processed, the dataset also contained other documentation, such as attachments and related documents. The solution therefore needed not only to read the documents but also to determine which material was relevant to each particular case.

Automation handles the volume, people focus on the exceptions

The solution was built around a human-in-the-loop approach. AI automated a large amount of repetitive document review and data extraction, while human validation was designed to remain as smooth a part of the overall process as possible. To support this, a modern review interface was developed in which the extracted information could be quickly approved or corrected. The interface was developed iteratively together with the customer to ensure that it supported the actual workflow as effectively as possible and made processing the large volume of material as efficient as possible.

– Monad responded to our needs exceptionally quickly and immediately gained a strong understanding of both the challenges in the material and our business needs. The solution was developed iteratively together with us, and the work adapted flexibly as our requirements became more precise during the project. The collaboration demonstrated both strong technical expertise and a pragmatic approach to solving problems, says Tiina Ylitolva, Head of Development, Business and Customer Applications at Elenia.

Relevant documents were automatically identified from approximately 165,000 files. Most of the results were validated by a person, while approximately 32,000 cases – almost one fifth – could be processed entirely automatically.

This division of work meant that experts did not need to spend their time manually screening the entire dataset. Automation carried out the labour-intensive initial processing, allowing human attention to be focused on the cases where it was genuinely needed.

The confidential nature of the material also placed strict information security requirements on the implementation. The solution was designed so that the data remained within Finland throughout the entire processing workflow. In addition to the functionality of the AI solution itself, careful consideration therefore had to be given to where and how the data was processed.

As a result, a large and structurally inconsistent document archive could be processed systematically in a way that combined the speed of automation with human quality assurance.