Price: 0
Number of applications: 4
31.08.26 (inclusive)
by agreement, based on the results of the assessment of labor costs
Idea
ICT tasks
ИКТ
Intelligent control systems
Software/ IS
Primary documents are received through several disparate channels — paper, e-mail, contractors' personal accounts, and files on removable media. There is no single register of receipts, registration is carried out manually in journals and tables. The details are entered by the operator in the accounting system character by character, from the paper original or from the screen. With a volume of 3,000 to 8,000 documents per month and peaks of up to 15,000 during reporting periods, this creates a constant overload: processing one document takes 8-12 minutes, and during reporting periods the queue of unprocessed documents grows faster than it is sorted. Correctness control is performed selectively. Input errors — incorrect amount, incorrect date, or a double document — are detected after the fact, when checking with a counterparty, or at the end of a period when their correction requires cancellation and reissue. Paper originals are stored in folders by period; searching for a specific document requires physically viewing the archive and takes up to a business day. During inspections, the selection of supporting documents becomes a separate project. The processing status of a particular document is not monitored: the answer to the question of where the document is currently located and who is working with it can only be obtained by contacting a specific employee. Ready-made commercial solutions of this class have a high cost of licenses linked to the number of pages, are trained on common templates and do not work well with the forms of documents of specific counterparties, and some of them require the transfer of document images to external cloud services, which is unacceptable for documents with commercial and personal information.
Reduction of labor costs for manual input of banking details by at least 70%; at least 80% of documents pass without manual correction of fields. Reduction of the average processing time of one document from 8-12 minutes to 2 minutes or less. Reducing the time required to complete a document from receipt to accounting from several business days to one. The accuracy of document type classification is at least 97%, and the accuracy of extracting key details is at least 95%. Reduction of the number of input errors due to automatic arithmetic controls and reconciliation with reference books; elimination of duplicate documents. The search for any previously processed document takes no more than 10 seconds instead of a working day; the package of supporting documents is ready for presentation during verification within an hour. Transparency of the process: for each document, the current status, the responsible executor and the full history of changes are visible. No royalties for the amount of processing and storage of all data inside its own loop without transmission to external services.
Vladislav Antsiferov
Purpose and description of task (project)
To develop a system that automates the full cycle of processing primary accounting documents: from the receipt of the document to the transfer of ready-made structured data to the accounting system. System composition: Acceptance of documents. A single entry point for all channels: dedicated mailbox, web-based downloads, network scanners and MFPs, and API reception. Automatic separation of multi—page files into separate documents, image preprocessing - eliminating distortion, removing noise and blank pages, and determining orientation. Recognition and classification. Automatic identification of the document type (invoice, invoice, bill of lading, act, contract, payment order). Extraction of banking details: number, date, parties and their identifiers, bank details, amounts, tax, currency, payment term. Extracting the tabular part line by line. Calculation of confidence for each field. Working with documents in Russian, Kazakh and English. Automatic controls. Arithmetic verification of amounts and taxes, completeness control of mandatory details, identification of duplicates, reconciliation of counterparties and nomenclature with reference books, comparison of the invoice or act with the previously issued invoice and contract. Verification. A screen with simultaneous display of the image and shape of the props, highlighting the recognized area when navigating through the fields. The operator checks only fields with low confidence and control deviations. Operator corrections are used to retrain models. Storage and search. Immutable storage of electronic images along with extracted data, full-text search and search by any combination of details, a complete history of changes for each document. Integration and reporting. An open REST API, data transfer to the accounting system, downloading directories, and reports on volume, timing, and recognition accuracy.
Whatsapp
