AI OCR organizes documents with KSeF, e-mails, PDFs, scans and photos: reads fields, classifies document type, suggests design or cost category and transmits exceptions to man.
The administration and finance department receives documents from many sources: KSeF, e-mail boxes, PDF files, scans, photos, forms and shared folders. Documents concern invoices, contracts, orders, protocols, accounting notes and project attachments.
The employees manually recognized the type of document, rewrited the data, selected the cost category, indicated the project or MPK and decided who the document should be sent to. With a larger volume, the risk of delays, mistakes, duplicates and lack of information which documents need to be corrected increased.
The most important difference is not the technology itself. The team moves from manual arrangements, sheets and messages to a process with clear statuses, inputs, case ownership and a place to check for exceptions.
The scope of modules can be narrowed down to MVP or expanded in subsequent stages after testing with users.
Central register of documents from e-mail, files, forms and manual upload.
Reading fields, tables, amounts, dates, counterparties and quality control documents.
Suggestion of document type, cost category, design, MPK and next step of the process.
Human panel for validation of reading, correction of fields and decisions with low certainty.
Statuses, case owner, comments, acceptance and audit trail of the document.
Transfer of data to ERP, accounting, BI or sheet and process quality report.
We check the types of documents, sources, volume, scan quality, required fields and decisions that the AI should not make on its own.
We define fields, statuses, levels of certainty, exceptions, roles and moments of mandatory human control.
We test the reading on documentary samples and determine which fields can be automated and which need validation.
We add classification, suggestions of categories and logic of passing low certainty to the queue of verification.
We combine the process with ERP, accounting, email, folders or dashboard through API, webhooks or export.
We measure read accuracy, number of corrections, classification errors, operating time and most common exceptions.
In any similar project, automatic steps must be separated from human decisions. Thanks to this, the system helps reduce manual work, but does not hide errors, exceptions and responsibilities.
Determine which data is primary and how we handle missing data, duplicates and inconsistencies.
Each case should have a stage, a responsible person and a clear moment of reaction.
Integration, import, AI or automation errors should be visible in the dashboard or report.
High-risk steps, financial documents and contentious decisions require human approval.
In financial processes we recommend human control for accounting decisions, corrections, low certainty and exceptions. AI may suggest classification and prepare data, but high risk decisions should be approved.
Yes, but quality depends on resolution, lighting, document layout and legibility. Before valuation, it is worth checking the sample of documents and determining the acceptable level of corrections.
You can save adjustments and use them to correct rules, dictionaries, prostheses, classifications or future models. This requires quality control and clear rules for the use of data.
Yes. The data can be transmitted via API, webhook, CSV or an exchange file. The range depends on the possibilities of ERP and required fields.
First from audit of document types and samples of files. Then it is worth to implement MVP for 2-3 most common types of documents and only after quality measurement develop further classes.
SmartCodeIT can help you design your process, prepare MVP and implement an application tailored to your organization. We start with a short conversation about data, tools, exceptions and business purpose.