"Unified Specialized Depository" introduced Directum Ario One intelligent services

2022: Implementation of Directum Ario One

On February 1, 2022, Directum announced that it had introduced Directum Ario One intelligent services into the United Specialized Depository. The specialized depository of OSD JSC controls, records and stores the assets of more than 70 management companies, which represent 240 mutual investment funds. Every month, a large amount of incoming documents arrive in the organization, of which about 100,000 are contracts of various types and formats. To manually enter this volume into the 1C: Document Flow system, 15 specialists are required.

Source: yandex.ru

Project Goals and Objectives

In order to reduce the labor costs of entering primary accounting documents and contracts, SD decided to introduce ready-made recognition services. The company's management opted for an intelligent Directum Ario One system that is compatible with any software class.

As part of the project, the team faced two main tasks

• Integrate Ario One intelligent services into your enterprise system
• Train AI engines to determine the type of incoming documents and extract the desired information from the text layer.

Documents with different priorities and processing dates are received by the SDD registration department. With a large incoming flow, it is difficult for a specialist to determine which sets you need to work with first. Pre-classification with artificial intelligence simplifies making this decision, " said Alexey Nikitin, First Deputy General Director of OSD JSC

The project was designed to integrate Ario One with 1C: Document Flow, which is easy to configure with the open Ario One API. SDS included in the project recognition of 135 types of documents - a total of more than 20,000 sources. This includes various types of acts, contracts, statements and orders. Next, the project team selected several criteria, and 107 types of documents took part in the final testing.

99% of documents provided by SDD included applications (room diagrams, registries, etc.), so Ario One services were tasked with correctly processing such multi-documents. Also, preparation for training has complicated the variety of formats, including archival ones, the presence of cover sheets with details, intersections by type of documents, dates in words, agreements in two languages, etc.

However, all these difficulties did not prevent the project team from performing work on time, having achieved a training speed of about 15-20 types of documents per day.

Trained models of Directum Ario One put work in test mode on the system "1C: Document flow." In total, three trial operations were carried out, and in between they improved the product. In particular, the cutting of documents was refined, a hierarchical classifier was added to improve the quality of species recognition and formatters to improve the extraction of facts. More than 3.5 thousand documents were used for testing. As a result, the customer confirmed the planned results.

Project Results

According to the results of test operation, the planned indicators were achieved:

• for cutting and classification of types of documents: 95%;
• for extracting facts from documents: 70%.

The company's employees saved about 30% of their time by entering contracts and PUDs in the SED. OSD plans to develop the system on its own using the open API Directum Ario One.

"Next, within the framework of the project, we are waiting for the integration of intelligent mechanisms with the 1C system in order to expand the number of recognized types of documents directly from the SED and train AI directly during operation. We plan to halve the labor costs of processing the incoming flow and reduce the burden on registration employees who receive up to 6,000 incoming documents daily, " said Alexey Nikitin, First Deputy General Director of OSD JSC