UPDATED 21:30 EDT / OCTOBER 13 2019


Hot bots: The payoffs and pitfalls of robotic process automation

Irish health insurer Laya Healthcare Ltd. needs to clean up thousands of accounts each month that carry balances that are too small to bother collecting. “A customer may have a balance of $150.12 and pay $150 because they can’t remember the correct amount,” said Information Technology Director Ian Brennan. “We’re not going to chase them for 12 cents, so we apply a credit to zero them out.”

With 586,000 customers, the company has thousands of such tasks to address each month. It’s a mind-numbing job that few humans would covet. “Our financial system didn’t have the functionality we needed, and the vendor wanted a king’s ransom to build it,” Brennan remembered.

Instead, Laya assigned the job to software robots developed by Blue Prism Ltd. For the past 21 months, Blue Prism bots have been logging into the debt ledger, identifying small-debt accounts and zeroing them out. “The ledger is much cleaner and our debtors’ team can focus on real outstanding debts without bothering with the shrapnel of the job,” Brennan said.

Laya Healthcare has now automated more than 100 tasks this way, jobs that used to the take anywhere from five minutes to six hours per day for humans to process. For example, it has taught bots how to do basic checks on simple claims, in the process automating about 15% of claims that were formerly done by human workers. “We’ve unlocked human time, which is the most valuable resource we’ve got.” Brennan said.

The technology Laya uses is called robotic process automation, and it’s one of the hottest new categories of software today. RPA uses intelligent software “bots” to perform repetitive tasks that have resisted automation through scripting, macros and other more established forms of automation. In most cases, the work involves switching between multiple applications and performing simple tasks such as copying and pasting among them or running search queries and gathering results.

RPA has elements of artificial intelligence, but while AI attempts to simulate human intelligence broadly, RPA is more of a tool for automating tasks. The software is becoming more sophisticated, though. The latest generation of tools even monitors users’ keystrokes as they interact with applications and identifies tasks that are candidates for automation. Some observers consider RPA to be a sort of gateway drug for AI.

Beyond scripting

Unlike scripts, which are typically limited in complexity or are specific to a single application, RPA actually mimics the interactions between human and computer, often switching among multiple applications. It’s best applied to what Gartner Inc. calls “swivel chair integration” in which humans perform rote work that involves a low level of decision-making.

Image source: CFB Bots

Image: CFB Bots

An optimal use case for RPA might be transferring data from a legacy accounting application that lacks data export functionality or application program interfaces to a new enterprise resource planning system. The only way to move the data is by re-keying it, a slow and error-prone process. “RPA is better for systems you don’t want to touch,” explained Rephael Sweary, co-founder of WalkMe Inc., which makes a platform for understanding and improving user experience.

An RPA bot can duplicate the human-computer interactions and flag exceptions that genuinely require human decisions. When combined with optical character recognition, the technology can be useful in transferring data from paper documents into digital systems.

As promising as all this sounds, Gartner estimates that most RPA implementations have so far underperformed expectations. Reasons range from poor choice of tasks to cultural resistance to overestimating the intelligence of bots. Early adopters and industry watchers told SiliconANGLE that having a plan for RPA is essential, even if the plan is not to do much planning at all.

A bot for every worker

The RPA market is still in its infancy, but it’s growing fast. Gartner expects global RPA revenues to total $1.3 billion this year, up 63% over 2018. The analyst firm forecasts that AI-based automation will reduce the need for employees in business shared-service centers by 65% by the end of next year. McKinsey & Co. estimated in 2017 that one-third of activities in about 60% of occupations worldwide could be taken on by such technologies.

“I believe eventually every employee will have a robot,” said Craig LeClair, a Forrester Inc. analyst and author of the recent book “Invisible Robots in the Quiet of the Night.”

UiPath's Patrick: "We are just getting started." Photo: SiliconANGLE

UiPath’s Patrick: “We are just getting started.” Photo: SiliconANGLE

One reason for all the excitement is the rapid rise of UiPath Inc., the Romanian software firm that last spring raised a $568 million funding round at a stunning valuation of $7 billion. The company is reportedly on a $200 million annual run rate and counts more than 400,000 customers of its freemium and paid products. UiPath’s user conference, which takes place next week in Las Vegas, is expected to draw more than 3,000 attendees.

Company executives say even their own customers have barely begun their automation odyssey. “Not a single customer of ours is over 1% penetrated in the potential of software robots to augment their workforce,” Chief Marketing Officer Bobby Patrick said in an interview with Wikibon analyst David Vellante. “We are just getting started.”

UiPath is far from the only RPA player attracting big bucks from customers and investors alike. Two leading rivals are Blue Prism, which went public in 2016, and Automation Anywhere Inc., which has raised a total of $550 million. Other pure-play combatants include NICE Ltd.AutomationEdge Inc., SAP AG’s Contextor, Nintex UK Ltd., Epiance Software Pvt Ltd., Kryon Systems Ltd., ActiveOps Ltd.’s OpenConnect and Softomotive Ltd. There are also more than a dozen software providers that include RPA features in other products.

Automation in action

Retirement insurance provider Security Benefit Corp. has identified about 300 processes that can be automated with UiPath. “That’s between probably 12,000 and 15,000 hours of time savings that we will get on an annual basis,” Jean Younger, vice president and Six Sigma leader at Security Benefit, said in an interview on theCUBE, SiliconANGLE’s streaming media platform at the UiPath Forward 2018 conference.

Younger targeted areas where RPA can relieve overworked staff, such as in investments. “They couldn’t get another headcount, so I gave them a headcount with a bot,” she said. “It’s doing all the processes that they’ve only been able to do on a monthly basis every day.”

At its best, RPA can enable business functions to be performed that would be impossible with available human resources. ConsenSys Inc., a company that builds enterprise and consumer platforms and applications based primarily upon the Ethereum blockchain protocol, uses RPA technology developed by Troovo Technologies Pty Ltd. to power a corporate travel management service called Smart Hotel Rate that ensures businesses get the discounts they’ve negotiated with hotels.

Customers submit a list of their negotiated terms in a spreadsheet. “They might have 1,500 agreements with different properties all over the world,” said John Packel, who co-developed the software. “There are dozens of columns for multiple types of rooms, seasons and blackout dates.”

That data needs to be compared to hotel reservations while also taking into account promotions and short-term offers. The Troovo bots pull real-time data in from Travelport Worldwide Ltd.’s travel commerce platform and converts it into JavaScript Object Notation for comparison using Smart Hotel Rate’s proprietary algorithms.

“It couldn’t be easier,” Packel said. “If we had to establish a separate data feed for every new customer, we’d never have gotten this off the ground.”

Great expectations

Like any technology moving up the early stages of the hype curve, RPA has had its share of inflated expectations. Definitional clarity is one. Some vendors of automation tools are now chasing a hot market by rebranding their products as RPA without incorporating the screen interactions or machine learning function that make RPA distinctive.

“What’s new is the way these automated solutions are accessing applications on our desktops through the same interface you and I would,” said Tony Abel, managing director in the supply chain solutions practice at Protiviti Inc.

RPA adopters apply the term “bot” to describe the way processes are automated, with the bot acting as a proxy for the human user. The distinction is important because a bot can be trained to take on multiple tasks and even improve processes over time, much like a human. Bots can also interact with each other and make simple decisions based upon the data passed between them, something that’s a challenge for scripts. “If you can create a bot that mimics what a human does and link that to other digital components you’ll have a worker that gets smarter over time,” said LeClair.

Gartner distinguished analyst Cathy Tornbohm defines four basic criteria for what should be considered an RPA tool. One is that the software has its own metalanguage and an interface that business people can use. RPA tools should require limited IT integration, although IT should be involved in their implementation, she recommends. The tool should sequentially run multiple processes with one bot rather than tying the bot to a specific task. Finally, the software should be able to record the user’s interactions and encode them in ways that they can be edited and improved.

Paving the cow paths

Forresters LeClair: "Eventually every employee will have a robot." Photo: SiliconANGLE

Forrester’s LeClair: “Eventually every employee will have a robot.” Photo: SiliconANGLE

Forrester’s LeClair adds another critical element: “You need central management, a control tower that lets people understand the activities of digital workers and find opportunities to dispatch new bots,” he said.

Such oversight was critical to the approach Synchrony Financial took to implementing RPA across its organization, said AI Lead Greg Simpson. “We had seen a lot of people in the industry turn on a lot of robots and not get much out of them,” he said. “We wanted to understand how to redesign our processes to take into account the fact that both bots and humans would be performing them. In many cases, the human jobs need to be redesigned as well.”


That recommendation was shared by several early adopters of RPA who were interviewed by SiliconANGLE. One of the risks of any kind of automation is “paving the cow paths,” or casting inefficient processes in concrete without first redesigning them. “Poor processes can be automated, but it is better to improve the process first,” wrote Tornbohm.

RPA and other forms of automation can have the unintended consequence of stifling innovation if not accompanied by a process review. “If you use spackle to patch bad processes it can remove your incentive to move to something new,” said LeClair.

Riley Adams agreed. In his previous role as a financial analyst for an energy firm, his company adopted RPA to automate a tedious process of populating quarterly earnings statements. A CPA who’s now a senior financial analyst at Google LLC, Adams spent about a day training an RPA bot to transfer numbers between two applications. The result was a savings of roughly 24 hours of work each quarter, as well as a reduction in errors.

“Assess the process before automating it,” he said. “Rethinking it end-to-end could result in a better outcome than simply automating the already inefficient steps taken for the work task.”

Synchrony's Simpson: "You don’t want to automate things you can eliminate altogether." Photo via Twitter

Synchrony’s Simpson: “You don’t want to automate things you can eliminate altogether.” Photo: Twitter

Synchrony has put in place a standardized tool for bot design, a control room for monitoring performance and standardized rules for introducing bots without creating errors. “We probably have fewer bots implemented than other companies, but we have the controls in place, and we looked at the process holistically with the people in mind,” Simpson said.

Synchrony uses software from Automation Anywhere in conjunction with a process re-engineering approach based upon Six Sigma techniques. The intention was to “get the current process mapped out and look for pieces that don’t make sense anymore,” Simpson said. “You don’t want to automate things you can eliminate altogether.”

Among the applications the financial services company has put in place is a machine learning algorithm that determines fraud risk and routes the results to bots for action. “The bot can check other systems, fill out different fields and reduce cycle time by gathering information so the investigator can understand the problem rather than gathering data,” he said. “We want people to focus on the cognitive part of the process and bots to focus on the busy work.”

Tripping points

For all that, RPA projects have failed, usually for one of a handful of reasons. One is overestimating the capability of the tools, by failing to heed what LeClair calls the “rule of five. If you get beyond five decisions or applications you probably need another technology like business process management or decision management,” which are used to streamline and automate complex processes, he said.

LeClair cited the example of one telecommunications client that was disappointed with the performance of its RPA project. Although the application was well-suited for automation, there were 90 different decision points in the process. “The bot went on forever,” he said. In their current incarnation, RPA tools “aren’t built for advanced decision management.”

RPA providers agreed that users should set an upper threshold on complexity, beyond which more sophisticated programming is required. “Usually our sweet spot is five to 10 steps,” said Sam Fahmy, chief marketing officer at RPA developer WorkFusion Inc.

But there are plenty of tasks that fall within that limit. Automated passport readers in an airport “can handle 99.5% of check-in requests,” Fahmy said. “We believe there are thousands of applications like that: Extract data, act on data, make a decision and provide an outcome.”

WorkFusion customer Dave Brajkovich agrees. Mississauga, Ontario-based Polaris Transportation, where Brajkovich is the chief technology officer, moves freight back and forth across the U.S.-Canada border every day, incurring a blizzard of customs-related paperwork in the process.

“It’s receiving paperwork, organizing it, filling in missing information and then moving that data to brokers who input it into their systems that go out for clearances,” he said. “It’s a manual process that could take anywhere from minutes to days because many hands have to touch certain things.”

By optimizing processes and then applying RPA, the shipping company reduced human touches by up to 60% and improved throughput by 85% in some cases. Error rates are also down. “We’ve gained anywhere from two to three hours in a day,” Brajkovich said.

Polaris Transportation's Barjkovich: "

Polaris Transportation’s Bjajkovich: “We’ve gained from two to three hours per day.” Photo: Dave Brajkovich

Polaris expects machine learning to amplify the payoff. It’s used to train bots to scan multiple systems such as the company’s enterprise resource planning, email, databases and even Google searches to pull back information human operators need. The bots are also getting smarter about reading forms and spotting data without human intervention.

Need for structure

A second fail is attempting operations on the wrong kind of data. Bots can currently only interpret tags or structured fields, LeClair said, although machine learning algorithms are steadily improving at interpreting and tagging unstructured data.

Data preparation can be a laborious process when many different variables are involved, but it’s essential to ensuring that bots run efficiently. Although simple RPA projects can go live in a matter of weeks, major banks have been known to devote a year to trial and testing, wrote Gartner’s Tornbohm.

Polaris Transportation spent more than three months training its bots to process customs forms, with about one-quarter of that effort devoted to tagging, Brajkovich said. “It was labor-intensive, but the more tagging we did, the more effective the bots became,” he said.

A third common setback is failure to specify all the steps in an interaction. Although RPA vendors endeavor to make their software accessible to business users, said Laya’s Brennan, “the inability of people to describe what they do in their day job is astonishing sometimes. They’re so used to doing the job that they don’t see the exceptions, and getting that level of detail is a challenge.”

Self-discovery features in the latest generation of RPA tools can help by recording tasks as they are currently being performed, but hard-coding those scripts into repeatable processes requires an understanding of program logic. “I wouldn’t let business users do it at my company,” Brennan said. “IT people tend to be more logical and ordered. You need that mindset.”

It’s also easy to put too much faith in bots to perform quickly and reliably. The reality is that they’re fundamentally pretty stupid. When a bot encounters an unexpected condition, it should be instructed to flag the error for human attention rather than continue with its task. “If a bot places bad data into a system, it can propagate through hundreds of other systems,” said Synchrony’s Simpson. “They can go really fast.”

Synchrony applies a rigorous testing process to new bots that has them process records one at a time rather than in batch. It’s slow, but the methodology is intended to minimize the risk that unreliable processes could be executed at high speed. “We want resilience, graceful exits and optimized efficiency,” Simpson said.

RPA has also introduced some unique new security issues. Because bots may log on to a half-dozen or more applications in the course of completing their tasks, access controls need to be maintained externally to the process or granted at the most basic level. Synchrony keeps its passwords in an enterprise vault from BeyondTrust Corp. and requires bots to request permission to retrieve a password from the vault and return it immediately after use.


Sidebar: Automation advice from the front lines

Developers and early adopters offer several guidelines for companies that are considering RPA.

Start small and show quick results. Although RPA is best deployed as part of a bigger process overhaul, gaining buy-in can be difficult. Ask colleagues which mundane tasks they would most like to jettison.

“Sometimes I’ll pave the cow paths almost to get into a department and show the value,” said Laya Healthcare’s Brennan. “I tell people to give me their worst process and I’ll automate it.”

Choose use cases wisely. Repetitive tasks with a limited number of decision points and data types are best. Choosing scenarios that are too complex can lead to failure and make RPA that much harder to sell internally.

Don’t let perfection be the enemy of good enough. Although analysts advise that processes be overhaul before automating, the reality is that there isn’t always time. Some users said they’ll put up with inefficiency for the sake of gaining acceptance.

“If you have to log in to 10 systems and it’s extraordinarily difficult to scale down to three, then leave it at 10 and let the bots at it,” said Polaris’ Brajkovich.

Be aware of costs, however. Prices can range from $1,000 to $16,000 per bot, and some vendors require multiyear licenses, according to Gartner.

Establish key performance indicators. As acceptance grows, so will demand. The criteria for judging success may be more than simple hours saved. Consider the value of functions that are being automated and how those freed-up skills might be best applied.

“Everybody will want a bot,” said Synchrony’s Simpson. You need to prioritize by value to the organization.”

Don’t forget qualitative payoffs.

“Understand if you’re improving the accuracy of the outcome,” said Mike Morper, vice president of marketing at Veritone Inc. “You should a benchmark of what accuracy is going in and coming out.”

Apply good programming practices. Most of the practitioners we interviewed involve professional developers in training and testing. Part of that process is documenting exactly what each bot does so anomalies can be quickly diagnosed.

Establish visibility and control. Successful deployments will quickly expand to dozens or even hundreds of robots. Having a high-level view into what each is doing and when reduces duplication and error risk. It also improves efficiency, particularly when multiple bots are working on a task and waiting for handoffs from others.

Laya Healthcare has a “superbot” that supervises others. “It looks at the master schedule and sees how many records needs to be processed,” said Brennan. It can assign the right number of digital workers to the tasks that need to be done.”

Anticipate resistance. Even though the jobs RPA automates are the ones few people want to do, any automation invariably raises concerns. Expect it and prepare a response.

“There was resistance off the bat, but we communicated to everyone what we were doing, what the expectations were and what opportunities there would be for people to enhance their skills,” said Brajkovich. “We held to that, and people for the most part had a smile on their faces.”


Bots versus jobs

The biggest impediments to overcome, though, may not be technical at all. Robotics and AI have been a popular subject of doomsday scenarios in recent years around visions of mass unemployment and economic dislocation. That doesn’t enhance their appeal as coworkers. Pushback was a common concern cited by early adopters.

There is reason for concern, particularly in industries with large numbers of people performing repetitive tasks. Wells Fargo & Co. recently forecast that 200,000 jobs could be lost to robotics in the US banking industry over the next decade. Forrester Research last year estimated that RPA will threaten the jobs of at least 230 million knowledge workers. Its most recent survey found that 30% of RPA use cases were in finance and accounting, LeClair noted.

However, fears of employment armageddon have yet to materialize, particularly in a nearly full-employment economy. Early adopters who were interviewed were unanimous in asserting the jobs aren’t being lost in their companies. The tasks that are being automated are the ones few people want to do, anyway. In most cases, they said, workers plenty of other things to do, most of them more fulfilling. “Some people are working 70 hours a week, and if I can take 10 or 20 hours away from them, they are loving us,” said Security Benefit’s Younger.

A 2019 Protiviti survey of 450 RPA users found that companies that embrace the technology actually experience higher employee engagement and lower attrition. “It isn’t about reduction so much as planning for growth and hiring fewer people in the future,” Abel said. “While there’s often angst early on, those concerns usually fade away.”

What won’t fade away is the impact software robots will have on the way we work and interact with technology. The chat messages that seem to pop up on every webpage these days are the tip of the iceberg. The bot is out of the bag and the workplace will never be the same.

Featured image: Franck V./Unsplash

A message from John Furrier, co-founder of SiliconANGLE:

Your vote of support is important to us and it helps us keep the content FREE.

One-click below supports our mission to provide free, deep and relevant content.  

Join our community on YouTube

Join the community that includes more than 15,000 #CubeAlumni experts, including Amazon.com CEO Andy Jassy, Dell Technologies founder and CEO Michael Dell, Intel CEO Pat Gelsinger and many more luminaries and experts.

“TheCUBE is an important partner to the industry. You guys really are a part of our events and we really appreciate you coming and I know people appreciate the content you create as well” – Andy Jassy