What is an invoice reader and how to use it?

One important financial document that is common to all businesses is the invoice.

The larger the business, the more types of invoices it receives in the course of its operations. The digitization of these varying formats of invoices into a unified system for easy storage and access becomes imperative to the functioning of the establishment.

That's where invoice readers come in.

The evolution of invoice readers

The era of filing and processing paper invoices, in which all physical invoices are manually analysed, the data written by hand into large ledgers and balanced, is long gone. Today, even small-scale businesses have some form of digital invoice processing system. Although many companies still deal with paper invoices, there has been a slow increase in the use of digital invoices among businesses. Some of the formats of digital invoices are:

• Visual Digital Format Invoices: JPG, PNG, GIF (picture formats), TIF (from scanning software) and PDF: These are simply digital images of the paper invoice.
• Data Digital Format Invoices:
  • Unstructured - The data cannot be automatically read from the document into accounting systems. These are free form texts such as contracts, letters, articles, and memos that may double as invoices in some unstructured, small businesses.
  • Structured – The data is in structured form and may be as Spreadsheets (e.g., Excel), tables in word processors such as Word (.doc), HTML XML Data PDF EDI (EDIFACT) and CSV.

The evolution of invoice reading has been rapid in recent years:

The first-generation of invoice reading – manual: Each invoice in whatever digital form it is, is manually read, and relevant data extracted and stored in a uniform system that bypasses the format variations of the various invoices received. This is old fashioned, involves intensive human labour, and is time-consuming, error-prone, and unsuitable for large companies that deal with many invoices on a daily basis.

The second-generation – invoice reader software: Also known as the invoice recognition software, the data from digital invoices, irrespective of their original format, are captured by software based on recognition of key data fields. The data thus read is stored in easy-to-access applications such as a spreadsheet or a database. Optical Character Recognition software or OCR are used for this purpose. While better than manual data entry, OCR could be restrictive in that developers must set up rules and templates to capture data and a certain amount of manual intervention is needed to check for accuracy.

The third-generation – AI-based invoice readers: Artificial intelligence-based invoice readers can intelligently capture relevant data with minimal errors due to the continuous learning processes of the AI tool. The feature of continuous learning in AI systems allows the reading software to adjust to all formats of invoices and gives it a universality across the company’s platforms.

Automated invoice readers

Software used to automatically read invoices can work on either second generation or third generation technology. Since each invoice holds key data that are used in accounting resource planning, and decision-making within the business, accuracy in data extraction is essential. The data thus read from invoices are usually then transferred to ERP, accounting, or data analytics platforms used by the company for subsequent processing.

A good invoice reader software must have the following features:

• The capability of extracting data that may be structured, poorly structured and/or unstructured in the original invoice. The coherence of data extracted from these various sources is eased through the use of AI-based data extraction.
• The capability of extracting data from multiple sources and formats of invoices
• The capability of converting the extracted data into multiple readable/editable formats for subsequent use.
• Data security - since most of the data read from invoices involve finance, they can be highly sensitive and the software used for automated invoice capture must be able to ensure safeguarding financial data from theft, hacking, and mismanagement.

Advantages of the invoice reader software

• Accuracy of data: Automation of data extraction from invoices can eliminate many of the human errors caused by fatigue or oversight.
• Time savings: Manual invoice reading is time-consuming, and automation can save much of the time spent by employees in mundane repetitive activities.
• Employee reorientation: The time available to the employee due to automation of invoice reading can be rerouted to productive tasks that can enhance their skillset and the company’s bottom line.
• Centralization of data: The data captured by the invoice reader can be stored in a centralized location and therefore will be accessible to all stakeholders of the company.
• Security of data: The possibility of introducing checks at various levels of the automation process initiated by the invoice reader can enhance data security.
• Scalability: As the business expands, it is cumbersome to have a manual system for invoice management. An automated invoice reader can streamline the process, leading to scale-up enhancements.

Modes of automated invoice reading

The reading of relevant fields in invoices of various formats is not trivial. Despite the progress made by AI and machine learning in recent years, the identification of complex patterns in invoices is challenging, but modern invoice readers have been steadily improving in this regard.

With known formats of invoices, e.g. from long-term clients who have not changed their invoice formats drastically, zonal OCR and keyword-based pattern matching can enhance accuracy and reliability in invoice reading.

Zonal OCR for invoice readers

The zonal OCR software can be trained to identify the structure and hierarchy of a known invoice through code or API. PDF invoice readers typically fall under this category. The OCR engine splits the document into physical “zones” that could correspond to a particular field. These zones are determined through the design of appropriate OCR templates. These zones are usually location-based, as shown in the following figure, in which, the user simply draws a square around data that must be extracted. Then, rather than reading the page as a single entity, the data in the specified zones are extracted as specified in the template.

OpenCV, Tesseract, and Python are some zonal OCR systems that can be trained to pick out specific fields from a scanned document. The invoice2data python package, for example, reads data from defined fields in invoices. It extracts structured data from PDFs using a template system. Other OCR libraries can also be used for python invoice readers.

Python invoice readers and PDF invoice readers can also extract line-items from invoices, which can be useful because the product information can be stored along with the classic invoice data such as date, number, and amount. This is especially useful to obtain fine-grained data that must be fed into a subsequent ERP system.

Pattern-matching and keyword searching invoice readers

Instead of or in addition to zonal OCR based invoice reading, intelligent filters can be used to isolate specific data that may be present in varying locations in invoices. These keyword filters work by checking for specific data forms (like numbers or currency symbols) in the entire document and searching for keywords around it to categorize the numbers into types like date, quantity, amount, etc. For example, when there is a dollar sign (“$”) in a pdf invoice, the reader can be trained to search for the words “Amount due” or “total due” or “total” immediately next to the sign and pick up the numbers that follow the dollar sign, to save under an appropriate handle such as “Total amount due”.

Keyword-based extraction is suited to read the metadata files such as total, date and number, and is not particularly suited for line items in invoices.

Challenges to automated invoice reading

Invoice reader software could fail when fine-grained table data are to be extracted from an invoice, the layout of which is unknown at that time. Zonal OCRs could fail in extracting data from semi-structured documents, in which the fields to be extracted are not in the same position in all the documents. The extraction of text from complex data fields, such as multi-line postal addresses is also challenging. Another difficulty that many invoice readers face is in the extraction of sequential data fields (e.g. continuing product numbers in the same invoice or receipt).

A solution to the above problems is to adopt a hybrid model in which an additional layer of human data validation is included in the invoice capture step. While the computer can do a large fraction of the job of invoice capture, the manual intervention can be kept minimal, only to validate the extracted data, thereby not adding to human labour time and effort significantly.

Another solution to the invoice reading challenges is Electronic Data Interchange or EDI. In EDI, instead of companies exchanging invoices in a format recognizable to humans alone, transactional data are exchanged between companies in a machine-readable format. The machines, in effect, “talk to each other”. This can obviate manual intervention. This however is not a universal solution yet because a majority of businesses worldwide still deal with invoices either in paper formats, or other human-readable digital formats like PDF.

Artificial Intelligence-based readers can also circumvent many of the above problems. Nanonets is an OCR software that leverages AI & ML capabilities to automatically extract unstructured/structured data from PDF documents, images and scanned files. Unlike traditional OCR tools, Nanonets doesn’t require separate rules and templates for each new document type.

Things to remember when adopting an invoice reader software

Companies that seek to adopt an automated invoice reader system must consider the following factors before launching:

• The infrastructure and IT resources required to support the invoice reader
• The financial commitment involved in setting up and running the system
• Integration with other systems within the company
• The levels of automation and human intervention required/possible within the business
• Availability of know-how within the company and customer support from the maker of the software
• The levels of data security required
• The level of access – this would decide where the data would be stored – in a local machine, a central server or the cloud.

Conclusion

The increasing digitization of the financial world necessitates changes in workflow structures and the use of tools that keep companies competitive. Invoice reader software can help companies spend less time on mundane activities such as manual invoice management and instead focus on their core competencies of customer care, innovation, expansion, and productivity.