At Gathr, innovation drives our mission to deliver cutting-edge solutions for our clients. Recently, our tech team has been working on a groundbreaking feature: the ability to process scanned bank statements. Why is this an exciting development for our clients? We’re breaking down barriers and providing access to financial tools for a broader spectrum of individuals, so that we can truly power financial access in Africa. This innovation isn’t just about improving our services; it’s about empowering individuals by enhancing affordability tools. To unveil this development, we sat down with David Beretta, the lead engineer spearheading this initiative. Let’s delve into the intricacies of scanned statement processing and its profound implications for our clients.
How does processing scanned bank statements enhance our affordability tool for our clients?
The Scanned Statements enhancement will allow our clients to extract affordability data from scanned bank statements which makes up over 30% of all statements uploaded to our system. Previously these statements were rejected and may have led to client drop-off. Catering for these statements will allow our clients to capture more business.
Why did we decide to incorporate scanned bank statements into our platform?
We noticed a large percentage (>30%) of bank statements that were uploaded to our system failed because they were either scans or images of bank statements. We decided to focus on implementing scanned statements for Capitec bank initially because the majority of the bank statements uploaded to our system are Capitec statements, the bank boasts the largest customer base in South Africa, totalling 22 million customers.
Could you walk us through the process of how we cater to scanned statements?
When a user uploads a bank statement, we initially check to see whether the document contains readable text or if it is an image (or pdf of images). If it’s an image, we use an OCR (optical character recognition) tool to extract text from the first page of the statement to determine the bank and read account information. Once we know the bank and the type of statement (i.e., 1-month statement, transaction history, etc.), we run some initial validations. If the validations pass the document is pushed into a job queue for asynchronous processing by a worker service.
During the processing job, an AI-powered OCR tool is used to extract all the transactions from the statement. We then perform further validations on the document, before processing the transactions and storing the data.
Once processing is complete, we provide our clients with raw transactional data, summary reports, and insights to understand affordability and spending habits.
What challenges have you faced while integrating this feature?
One of the biggest challenges with OCR is that it is not perfect. Extracting accurate data, particularly the transaction amounts, is crucial to the process and critical for our clients. We have had to implement numerous checks and redundancies in reading the OCR data to increase accuracy. This does, however, come at a cost of performance.
How does scanned statement processing address the needs of lower LSMs?
Lower LSM individuals are more likely to provide hard copies of statements printed from ATMs or bank tellers due to their lack of access to computers and online banking, with only 10% of South Africans having access to a computer at home. These hard copies are then scanned and uploaded to our system, mostly by an agent or sometimes by the customer. With this new optimisation we will now be able to process these documents.
What measures did the team take to ensure a smooth experience for scanned statements?
Scanned statements support will be available within the existing affordability module. This means that clients will be able to use the same API endpoints for uploading statements and retrieving data from statements, so long as they have the Scanned Statements feature enabled. No additional development or integration work is required to leverage this feature.
How does this expansion align with our company’s mission and goals?
Supporting scanned documents will allow us to cater for a large segment of the market – particularly lower LSM – that were previously unsupported by our system. This means that more people will have access to financial products that were previously excluded, aligning with our goal of furthering financial inclusion for all.
What are the benefits of this feature for our clients?
Our clients will be able to upload any type of bank statement, be it scanned or regular, and receive fraud checks, validate accounts and retrieve valuable insights from the data on these statements.
Could you discuss any notable technological innovations implemented during the development of this feature?
We have developed several proprietary cross-validation techniques that allow us to increase the accuracy of extracted text from bank statements.
How does our team handle potential inconsistencies in scanned statement formats?
We use an AI tool to pre-process and de-skew documents before extracting text. We have also built some in-house cross-validation algorithms to ensure the accuracy of the data.
How do you envision the utilisation of scanned statements evolving in the future?
The technology around OCR is improving rapidly, particularly as it leverages the growth in AI. Text extraction is becoming faster and more accurate and can also be more easily customized/fine-tuned to specific types of documents. As the technology improves, we expect it to be used by more clients as they will have more trust in the data that is being extracted.
Beyond Capitec, what’s next?
We have already begun adding support for FNB, Nedbank, and Standard Bank statements. We have ABSA in the pipeline too. We will eventually support all the banks for scanned statements that our system currently supports for regular pdf statements. This is part of our roadmap for the next few months.
As we continue to push the boundaries of innovation, the integration of scanned statement processing stands as a testament to our commitment to empowering clients and advancing financial inclusion. Through the expertise and dedication of our engineering team, we look forward to further enhancements that will shape the future of financial technology.
