Citi and TD Bank are exploring ways to build upon simple applications of robotic process automation with optical character recognition (OCR) applications.

“Like most organizations, we started with the very basic [RPA] stuff and now we are focusing on all three aspects, from the front-end to the mid-office to the back-office,” Manav Thiara, who leads digital, automation and process simplification at TD Bank, said during a panel at the Banking Automation Summit today. “Now we are really on this thin line, which is graying the area between where we start with RPA and then move that automation toward more artificial intelligence and machine learning.”

Specifically, TD is looking at OCR to operate alongside its bots to automatically process and digitize documents containing handwritten notes or signatures. As the bank walks that fine line, Thiara said it will need to be diligent from a data ethics perspective to make sure the data is being used as it’s meant to be. TD Bank is the Cherry Hill, N.J.-based subsidiary of the Canadian megabank TD, which has $1.3 trillion of assets.

In the recent years since TD has jumpstarted its automation journey, the bank has been able to automate most back-end processes, resulting in tens of millions of total savings, Thiara said. Its anti-money laundering programs, for example, process more than one million transactions every month using bots. And during the pandemic, TD rapidly created RPA for its front-end operations, specifically introducing a new virtual assistant that uses conversational AI to provide real-time responses to COVID-specific questions. It also leveraged automation to build digital applications for customers to apply for small business loans and payment deferrals.

Meanwhile, Citibank’s efforts to bolster automation with additional technologies comes down to converting documents into data, because “once you have [digitized] data, then you have the possibilities of navigating it through algorithms and learning,” said Tapodyuti Bose, global head of digital channels and data for Citi Treasury and Trade Solutions.

The $2 trillion dollar bank is using OCR to digitize its decades-old cache of documents, then metadata tagging them to organize and manage the resulting data sets.

Many documents in the B2B space are bespoke, compared with the consumer side, Bose said, and contracts are typically customized, with the legal language differing between companies. But while this has made OCR technology indispensable in digitizing documents, the bank has not been able to reap benefits of RPA within Citi Trade and Treasury Solutions, he added.

“Our personal experience from an RPA, [is that it] is really a step automation kind of a tool, which banks its success on a highly structured, repeatable process, which runs on predicted data sets,” Bose said. “We have been largely disappointed in use of RPA. It has often cost us more than it is has saved because of the complexity of the B2B process.”

For RPA to work well, it requires a “non-trivial” amount of effort to integrate it, so that it understands the data and the business process, Bose said. Plus, it has limitations that can make asynchronous processes challenging. So, in many cases Citi is trying to eliminate rather than automate manual processes, “otherwise you’re trying to throw good money after a bad process,” he said.