Large banks all over the world are piloting the use of robotic process automation, also known as bots, and getting results. But the efficiency gains have come at a cost.
Bots are small pieces of code designed to do one task formerly done by a human, such as opening an application, capturing data, and entering that data in another application. Financial institutions are using them in account opening, operations, anti-money laundering compliance, customer service, and elsewhere. Bankers on three continents report faster booking of transactions, expense savings, improved risk controls and better management of the workforce.
"The technology works," said Dean Mazboudi, cofounder of Deutsche Bank Labs, the bank's innovation arm. His company has been doing proofs of concept with robotic process automation for a while, mostly in the back office.
"In every case we experimented with, we saw positive results," Mazboudi said. "We saw pure improvements in the process by virtue of putting better controls around a process." (The bank is using WorkFusion's software.)
The bot experiments come at a time when banks are starting to use various other kinds of artificial intelligence to improve fraud detection, to identify cybercrime, to handle compliance chores, and to make lending decisions.
This new technology comes with unique implementation and deployment challenges. Bots raise security concerns, for instance. They also require a computing environment in which they can be programmed and tested and need to be designed precisely and be able to address exceptions. And bankers can't do this alone — they'll need help from operations, IT, and outside experts.
At Cooperative Bank in the U.K., if your card is lost or stolen, a bot will take down the details and cancel the card.
At Barclays Africa, a chatbot is working to engage customers in the hopes of bumping up the new account activation rate, which has been 68%.
"Africa is a young continent, most people are under the age of 24, the median age is 19.5 years old, and they're well equipped with mobile phones and heavy users of Facebook messenger and other chat channels," noted Yasaman Hadjibashi, chief data officer at the bank.
In June, the bank began sending new millennial customers an SMS text message immediately after they opened their account, encouraging them to ask questions. Basic questions about branches or products are answered with information from the bank's website.
In a pilot of 1,000 millennial customers, the percentage of activated accounts rose to 73% from the usual 68% of those who open accounts.
"Millennials love this stuff," Hadjibashi said. "It's an evolution for how chatbot technology is going to enter the banking system."
Routine transactions — customer onboarding and offboarding, mortgage processing, HR processes — can easily be done by bots, said Ian Barkin, co-founder and head of strategy at Symphony Ventures, a firm that implements robotic software. Most of the processes banks offshore could be given to bots, he suggested. It might be less politically treacherous than shifting the jobs overseas.
But the top use for bots in U.S. banks is in the back office. "There's a huge unmet need for streamlined operations," noted Alastair Bathgate, CEO of Blue Prism, whose robotic process automation software is used by BNY Mellon and others. "IT doesn't have unlimited resources."
Early experimenters have found bots have special needs.
Process redesign. Before bots can be deployed, existing processes need to be examined, understood, and maybe changed. "You don't want to automate bad processes," Mazboudi said.
For instance, a loan process today might be triggered by an email instructing the loan operations group to reset a loan or make a disbursement. Yet emails can get lost, deleted, and left unread. Instead of training a bot to open and extract data from email, it might be better to rethink the process.
Getting people to buy in. To program a bot correctly, you need the people currently doing the task to explain how they do it. But people who know they may be replaced by a bot may be less than enthusiastic about the project.
Mazboudi said he's freeing experts to focus on higher-value activities.
"And that's a good thing," he said. "Many times these low-level workers, it's a miserable job. Over the long term these individuals can do more complicated things or things that have more value." The turnover rate in operations jobs is about 40% on average, he noted.
Adam Devine, vice president of WorkFusion, said the company's software can learn to perform tasks just by watching someone.
Getting other executives support bots is also a challenge. "People don't understand robotics unless they see it," one bank executive said. His team created a video that showed a process of getting a tuition payment sent to a school getting shortened from 22 days to a few hours, with the use of bots. This made the benefits of bots clear to many colleagues.
Managing identities. Bots access applications the way humans do — by entering a user name and password. How do you keep those credentials completely secure and how does the software distinguish between bots and malware?
One large U.S. bank has created "robot IDs" — specially created credentials that only the bot knows and no employee has access to. They're in a secure vault. Each bot can only access applications tied to the process it's assigned to.
Deutsche Bank relies on "persona" IDs that are connected to human bot supervisors, who are responsible for the bots' actions.
"A bot is not something you can hold accountable; traceability and accountability have to go back to a human," Mazboudi said. "You can't tell a regulator, 'I didn't do it, the bot did it.'"
Security, period. Like any process that uses software running on the internet, bots face cybersecurity threats. A large South African bank that had just gone live with bots on a web server recently had to suspend the project due to fraud that had nothing to do with robotics. The bank hopes to have the project back up within a month.
Need for IT help. "If any consultant tells you that you can do this without IT, good luck, you won't be connecting to any backbone any time soon," said an executive at the South African bank. And bots need computing environments to be programmed and run in. "Unless the IT guys can spin up an environment quickly, it's an absolute roadblock," the executive said. His bank will use Docker container technology for this.
Good implementation help is hard to find. Although there are many consulting and business process outsourcing firms eager to help banks implement bots — one banker said he'd been pitched 26 times — finding the right help is hard. "You want someone with experience and this is all new to banks, so nobody has that deep experience yet," an executive at a large U.S. bank said.
Bots do only what they're told. Deutsche Bank is considering customer service chatbots but is treading lightly, because bots' capabilities are still narrow. "We can develop a bot to automate responses to password resets, but that bot will not be able to look up your account balance," he said. A bot could be taught to authenticate the user and give her an account balance, but it would not know the customer has other accounts with the bank unless it was specifically trained to look for them.
"This is where machine learning is today," he said. "It's not learning, it's machine learning. It's based on historical data, data and patterns you are able to translate for the machine. It's got some ways to go before it becomes more aware."
People will get laid off. "One thing that's inevitable is that jobs will shift," the South African banker said.
Headcounts will be reduced, but some people will learn new skills, such as bot supervision and bot programming, and might find the work more interesting.
"We've been losing jobs to machines since the wheel was invented," Barkin said. "This is just another wave of it."
Editor at Large Penny Crosman welcomes feedback at firstname.lastname@example.org.