Optical Character Recognition (OCR) Technology in FSI core processes
Optical Character Recognition (OCR) is the technological process of recognising and converting both handwritten and printed characters into editable and searchable data. It has two primary functionalities: eliminating manual data entry and extracting information automatically. For example if you wanted to digitalise and edit a paper contract, you could either spend a long time keying in the document, or you could use a scanner/photo and OCR to convert the file within seconds into an actionable file.
OCR technologies now achieve a very high level of accuracy in character recognition, of over 99%. The challenge today is to locate and extract items of data in printed or scanned documents by identifying the relevant combined strings of information that need to be extracted. This is specially challenging in the case of unstructured documents and tables. Natural language processing (NLP) and machine learning can be used to identify and extract data from documents and directly enable functions that would otherwise not be possible (i.e. cross-validation or summing of numbers in tables). Many OCR vendors already make use of these techniques to improve functionality of the OCR tools they offer.
In digitalisation projects, OCR technology is often used together with workflow tools to automate processes and reduce manual work. These tools collect, retrieve, process, edit, archive or forward data and documents.
OCR is just a means to an end, and it needs to be combined with advanced analytics software to add real functional value, by integrating data extracted by OCR engines from documents with AI-powered systems, for many potential use cases, such as fraud detection, regulatory compliance or process automation.
Figure 1. Illustrative OCR processing steps
Figure 2. Example of a user interface in the extraction stage (Swiss tax declaration with automatically extracted values for validation)
Selected use cases for OCR technology
Credit risk management process in banking: Credit assessments can be extremely time consuming, as they make use of financial documents that are original, photocopied or scanned. This involves inputting each line item manually into an IT system for the purpose of the assessment. This process has a fairly high risk of input errors, with only limited transfers of data, with the result that the financial analysis for credit purposes may be unreliable. OCR combined with AI can not only digitalise scanned financial statements but can also turn them into readable and searchable datasets that are accessible across IT systems. This can automate part of the work of the credit analyst. In this process the first step is to digitalise financial accounting statements using OCR. Machine learning algorithms then teach the system to differentiate the underlying patterns in the accounting statements, to identify and extract the required data. This achieves a time saving of around two to four hours per credit review, and the credit analyst is also able to focus more on the actual credit risk analysis and to make better-informed lending decisions.
- Automated loan/mortgage processing: During the mortgage/lending processing, OCR can enable the automation and processing of a diverse set of documents associated with lending such as tax declaration, payslips or salary certificate, ID/passport, etc, thus improving the customer journey and optimizing the process. This is especially a benefit when aiming for a high Straight-Through-Processing in the lending decision, such as on lending platforms.
- Customer Onboarding and KYC process: OCR can in the same way help to automatically read, route, and process documents typically required in KYC processes, enabling faster and better customer service from the first interaction during account opening and customer onboarding.
- Insurance claim processing: During analysis and processing of insurance claims, OCR can reduce or even eliminate human error and speed up the processing and handling of these claims.
- Travel expenses documentation: Businesspeople who travel may have to document and register their expenses for refund and audit purposes. They may have to enter their expense details into a portal for checking by a member of the accounts staff. This process can be cumbersome and time consuming. As an alternative, OCR technology can be used to copy data from receipts, approve or reject it and transfer it to an app or portal directly.
In general, a well-developed/trained system can easily reach accuracy levels above 90% when extracting data from SMEs Balance Sheets of SMEs, client onboarding and other various documents, meaning that automation is often more accurate than if the process is done manually, but at a fraction of the cost and time. Overall, OCR technology can save time by cutting down on or eliminating manual processes, improving productivity and reducing the likelihood of errors or fraud.
Big players in the OCR market
A number of vendors offer ready-to-use OCR. Some of the main products in the market currently are ABBYY FlexiCapture, ABBYY Vantage, Google’s Vision AI, Amazon Textract and Microsoft’s Computer Vision.
Deloitte and ABBYY: Use case for a D-SIB retail bank in credit processing
Our client, a leading D-SIB, initiated several digitalisation projects in the credit sector, including the development of a mobile B2C mortgage loan platform. This app allows end customers to apply for mortgages remotely, simplifying and speeding up the application and approval process. Since the credit lending process in Switzerland still involves a large number of paper forms, the bank sought a solution for document identification and automated data extraction by the end users themselves, in order to relieve customers of the tedious task of manually inserting the necessary data manually for the loan application.
Deloitte was responsible for building these capabilities and in partnership with ABBYY, was able to deploy within a short-time an OCR solution using ABBYY FlexiCapture, to extract relevant data with high accuracy (above 90% after training) from tax returns, salary certifications, national identification documents, as well as foreign residence permits and pension fund statements. Deloitte applied its expertise in the technical implementation of OCR solutions and its knowledge of the regulatory requirements and current best practice in Swiss lending processes. It was important to apply a reverse engineering approach to identify and simplify the data points required for the credit decision making in alignment with the client’s credit risk appetite and model. to develop a platform for processing mortgage applications quickly with a flexible configuration of input and output interfaces, which allowed for a seamless integration via API to mobile users providing an intuitive customer journey to mortgage applicants.
Figure 3. Workflow Example of OCR-App integration into a Swiss mobile mortgage application process
The data gathering and extraction process is as follows: The customer either uploads a scan of the required document or takes a photograph with the smartphone directly through the mobile mortgage application platform. The uploaded document is then analysed in FlexiCapture and automatically classified. Depending on the document, FlexiCapture extracts certain fields containing the required relevant information. Next, the accuracy of the character recognition is tested and if it is above a certain threshold (in this case 90%) the data is automatically exported for further use in the mortgage application. If the accuracy test falls below the threshold, the user has the option to review the extracted information and correct it manually.
At Deloitte, with our extensive experience implementing the latest technologies and our expertise in the financial services industry we can help you to increase efficiency in your processes, reduce costs and unlock value. Do not hesitate to contact us for any further questions.