The purpose of this article is to explain how to process bills automatically. Specifically, how to manage data extraction and obtain supply and consumption information useful for numerous processes. The following are the major topics covered::
- Application scenarios
- Use case description and problems
- Extracting relevant information
- Processing alternatives
- MyBiros: a modern approach to IDP
Let’s see in which scenarios it can be useful to process bills automatically.
Having the information (in a structured format) in the bills available proves useful in many processes and application scenarios, for example:
- Processing and control of all energy bills received and availability of data to support supply tenders;
- Production of analyses, reports and KPIs preparatory to energy efficiency activities;
- Automatic accounting and reporting of all invoices;
- Supporting the definition of forecast budgets and monthly expenditure allocations.
Use case description
Bills are semi-structured documents, in fact each provider defines the format to be used. Typically, the information contained in the various formats is very similar, if not the same. Depending on the type of supply, the same provider may change the bill format over time. The complexity of the bills and the large amount of different formats lead to many complications that penalize traditional solutions. The latter see their own limited accuracy on these documents and consequently also the degree of automation of the entire system. Here are some examples:
- Document acquisition is not standardized. The bills to be processed may be acquired from different sources, which means that the formats in which the documents are received are heterogeneous. The most common formats are digital PDFs, scans and photographs. Formats heterogeneity complicates processing. An example is document captures using smartphones, which often produce blurred or rotated images that are therefore difficult to read. Such problems reduce the applicability of traditional approaches and penalize the efficiency of traditional data entry. To solve these problems, image normalization steps are required, which increase the pipeline complexity and decrease the generality of the approach.
- The formats of bills are many (at least one per provider and per type of supply), the key information is numerous and their position varies in the documents. Thinking in an international context, in addition to the change of format, it is also necessary to easily manage the change of language and units of measurement used. The number of useful formats and field positions to be considered can therefore also become very large. A further complication is the common change of bill format over time.
- Several interesting information is contained in tables or images, this further complicates the process.
These are just some of the problems associated with automated bill processing. The article follows by describing methodological alternatives for solving the problem. For ease of discussion, the article will deal with Italian energy bills without loss of generality.
Relevant information to extract
The most important information to be extracted involves consumption, supply details and contract holder information. These can typically be found in different formats and units. Another aspect that complicates the use case is the amount of relevant information, in fact we are talking about more than 25 different fields. The main pieces of information of interest are listed below: tariff, type of consumption, total consumption, cost of energy raw material, reference period, POD, total amount to be paid, provider data, consumption bands (such as f0,f1,f2,f3), recipient and holder data, supply data (voltage, committed power, etc.).
To correctly process such a document, several functions are therefore required to be used in synergy: extraction of key-value information, interpretation of tabular data and classification of the type of bill (gas, electricity, etc.).
Manual data extraction from energy bills (but the same applies to any bill) is expensive, time-consuming and error-prone. The processing steps require skilled people who can identify the relevant information in the document and extract it consistently from sometimes even complex layouts. Some challenges and issues related to manual processing include:
- Cost issues
Although manual data extraction from documents may work for small companies with limited processing volumes, it becomes expensive as the company expands. In many cases, manual data extraction involves hidden costs in addition to those associated with hiring more employees to perform the data extraction. Adding more employees to the workflow often leads to coordination problems, which in turn can lead to errors, especially in data identification and data entry. Data validation itself is a critical step that adds to the cost component of processing. In fact, the information extraction without verification steps can have error rates of up to 4%. The 1-10-100 data entry rule is well known in data entry back offices: checking the accuracy of data at the point of entry costs about $1, cleaning up errors by re-checking the entire batch of data costs $10, and escaped errors cost the company $100 or more.
- Time issues
Manual data extraction is time-consuming, especially for global supply chains, as multiple checks and approvals are required and teams in different countries often have to be co-ordinated. The integration of the various processing and verification steps can also become complicated due to the various people involved in the hierarchy and the required approval levels.
- Human Fallibility
The data entry process is repetitive, tedious and can be demoralizing. Furthermore, energy bills do not have a standard format. Although much of the information listed in the previous sections is present in all documents, each vendor uses a different format with significant spatial variability of data. Manual data extraction is laborious given these factors. Language variations between the place of issue and the place of delivery can also pose a significant challenge in understanding the data. All these variables contribute to an increased likelihood of introducing errors into the process itself.
Traditional OCR Solutions
Bills processing with traditional OCR techniques and template matching/regex is a highly inadvisable and costly approach as it is necessary to have ad hoc rule sets and templates for each document type. The formats are many and the vendors are potentially undefined in number. The languages to be considered are often numerous for a solution that must work in processes with a global scope. This makes the number of rules or templates needed decidedly numerous and constantly changing according to the new formats and countries to be considered. This leads to a high setup and maintenance cost of the solution and often poor performance. In addition, the maintenance and configuration of the solution must be carried out by trained and technically skilled resources.
In general, all the problems presented in the use case description afflict both manual and traditional approaches. This has led to the need for higher-performance solutions that solve the complications described so far. Thanks to recent developments in the field of AI and in particular Deep Learning, it is possible to achieve higher quality results and to cut time and costs at every step of the pipeline. Starting from OCR capable of learning, improving over time and transcribing even manually written documents to semantic analysis and interpretation of tabular data (and much more). The set of techniques based on artificial neural networks for comprehensive document processing is commonly called Intelligent Document Processing.
Intelligent Document Processing (IDP)
A modern approach based on Deep Learning techniques is the best choice for solving such problems. In fact, the possibility of using the best Computer Vision techniques for analyzing and reading the document, and NLP for understanding natural language, makes it possible to solve previous problems. It is not necessary to adapt the solution each time (writing new rules or configuring new templates) but simply to have a sufficient amount of data belonging to the process to instruct the system.
Another advantage is the possibility of applying the same approach to solving different tasks, such as key-value data extraction, tabular data extraction and document classification. Such an approach can also benefit greatly from the human validation step, which not only corrects errors made by the system, but also enables continuous learning of the algorithm, allowing it to improve over time and calibrate itself to the specific process.
Compared to traditional solutions, the maintenance and evolution of the system is also simplified. In fact, adding a new field to be extracted, a document category to be classified, or wanting to add a new language among those supported does not involve writing code. The collection of new documents will suffice and the subsequent retraining of the system can be easily followed even by non-technical resources. Finally, the most effective IDP solutions allow for unprecedented accuracy of results, far surpassing traditional approaches
myBiros and benefits
myBiros is a high-performance, user-friendly and versatile Intelligent Document Processing solution that enables automatic document processing. Core functionalities are information extraction and automatic document classification. All this is offered via a prebuilt set of ready-to-use APIs with pre-trained templates for the most common use cases and the possibility to re-train the entire pipeline (both OCR engine and document interpretation system) for custom cases.
By exploiting advanced deep learning techniques that analyze multimodal features, it is possible to process all document types in a single solution. The system uses pre-trained models and data-augmentation techniques and can therefore be trained with a small volume of data, allowing processes involving a small volume of documents to be automated.
This solution includes a scoring mechanism: the system reduces false positives by enabling the possibility of reviewing low confidence data while minimising errors. Interaction with a human user enables the system’s errors to be corrected while continuing to train it so that it does not repeat past mistakes (Human in the loop and continuous learning). Finally, the high scalability of the cloud-based architecture makes it possible to process highly variable volumes of documents without having to allocate expensive resources in advance.
Additional features include the ability to process tabular data, identify image artifacts and the ability to process heterogeneous and multi-language documents in a single pipeline.
The features mentioned so far allow myBiros to perform optimally in the processing of bills. Succeeding effectively and quickly in identifying all relevant information. If you are curious about how myBiros works in order to simplify bill processing, contact us and try our demo. We are ready to help you!