IDP automation: Banking use cases and benefits

Commercial and retail banks for years have been trying to automate document-intensive processes, but with limited success. The reason: because most processes deal with documents that contain unstructured content.

Early bank process automation attempts relied on systems based on keywords, rules, or templates, all of which fall short when confronted with unstructured content. Such approaches would have to account for every possible document type a bank may come across – an unlikely scenario given the range of documents and content involved in many processes.

Consider customer onboarding. It requires all sorts of documents to get a new customer set up correctly, including the bank’s own application forms, perhaps tax returns, statements from other institutions, credit reports and the like. You would have to spend millions paying consultants to come up with templates or writing rules for every possible document type you may encounter. Even if you succeeded for a time, as soon as a new document type came along, or an existing document format changed, the automation would break down. Money wasted.

As implemented by Indico Data, intelligent document processing technology is fundamentally different from OCR, RPA and templated approaches because it can understand document context much like a human does. That’s because the Indico Unstructured Data Platform is based on a model that incorporates some 500 million labeled data points, enough to enable it to understand human language and the context of a document.

Building an effective automation model requires having a large set of data to “train” on. It’s that trove of data that brings bring intelligence to any artificial intelligence solution. But even the commercial or retail bank would be hard-pressed to come up with that much data and effectively train its own models.

Indico uses an AI technology known as transfer learning to create custom models that can tackle virtually any downstream task – including customer onboarding, loan processing and other common banking processes. The end result is it takes a relatively small number of documents to train the model – usually just a few dozen. What’s more, you don’t need data scientists to make it all work. Rather, business professionals on the front lines train the automation model – those who know the processes best. (For a deeper dive on this point, check out our Intelligent Document Processing page.)

While intelligent document processing often works well on its own, some commercial and retail banking processes can benefit from the combination of IDP and robotic process automation.

RPA is effective at automating deterministic business processes that involve highly structured data. So, it’s useful for automating repetitive, labor-intensive tasks, often with greater accuracy than humans – because software robots don’t get tired. IDP, meanwhile, is able to automate processes that involve unstructured data, making it complementary to RPA.

One example of how RPA and AI can work together in banking is automating lockbox processing, which involves matching payments, in the form of checks, to invoices. An IDP tool is used to “read” invoices and checks, extract relevant data and translate it to a structured format, then hand it off to an RPA tool for input to a downstream processing system.