Deploying small artificial intelligence (AI)-based automation projects is one thing. But implementing it in a large enterprise at scale is quite another. Yet it is a challenge that many organizations are being forced to confront due to the unprecedented adoption of AI and ML (machine learning).

According to Bobby Patrick, chief marketing officer at robotic process automation (RPA) vendor UiPath, most businesses are unprepared for AI. He said that three years ago only 2% were using ML and now it is up to 75%.

“AI ignites ideas faster, but automation is the best path to deliver AI and give its implementation some muscle,” he said. “Anything that AI can do, AI and automation can do better.”

At the UiPath Forward conference in Las Vegas last month, a variety of organizations laid out how they are dealing with the scaling of AI and automation as they seek to optimize operational efficiency, employee engagement, and infrastructure.

“Scale and size are a real challenge,” said Neeraj Mathur, director of Intelligent Automation at VMware.

His company has now automated 470 processes across 45 business functions using 270 software bots across the network. Mathur estimated that AI-based automation has already saved 1.6 million hours of manual labor. The key to his success in scaling AI-based RPA is constant testing. VMware does an automated test for every change made to software bots, automation process and related network connections, storage locations and upgrades. Over the last couple of years that has amounted to more than 800 change requests.

“Every change needs a test,” said Mathur.

Previously, the company engaged in manual testing and smoke testing (testing to ensure stability of critical functions) for its RPA program. VMware decided to move from reactive to proactive maintenance and to find a way to fix things before they broke while boosting automation uptime and the speed of software bot deployment across the network.

A continuous integration/continuous discovery (CI/CD) pipeline plug-in was created to deliver new versions of software for RPA workflow projects consisting of elements from Jenkins, GitLab and UiPath Test Suite. Result: a 96% reduction in error rates and a 15% increase in speed to production.

The Industrial Bank of Korea (IBK) attributes executive and department buy-in as key elements of its success in scaling. Automation efforts to date have replaced almost two million hours of manual work. Bots have taken process times in branches from two days to 15 minutes in some cases.

KT KyoungTae Kim, leader of Center of Excellence at IBK, said that as software robotic headcounts rise, they need a new level of management, much as would be the case if employee headcount rose sharply.

“You need to inject more control to influence ROI when robot headcounts increase,” he said.

Further, AI-based RPA requires buy-in for it to succeed. At IBK, that came first from the CEO validating automation from the board level. But that wasn’t enough. The pace of automation in departments differs sharply. Some are well ahead of others and some show little interest at first. IBK also presents its automation projects in different ways based on the mindset within individual lines of business (LOB).

“Some departments are more interested in avoiding risk rather than cutting hours so you must set your RPA metrics accordingly to get buy-in,” said Kim.

Japan's largest telecommunications company NTT Docomo wrestled with digital transformation and automation on legacy infrastructure. A legacy system structural reform project began in 2019 to enhance infrastructure systems that had been in use for more than a decade.

"With our 30-year history, there are many legacy services and app mechanisms, but if there is an important service for our customers, it cannot be easily stopped,” said Takanori Nagahara, chief of application development in the Service Design Department at NTT Docomo.

The project included migrating systems to modern platforms, introducing a containerized architecture as well as finding ways to accommodate older systems that were difficult to change. Software development processes were upgraded so releases for mobile and other apps could move from a couple of times a year to several per month.

Automated testing plays a part here, too, shortening the time it takes to develop and release everyday services and apps from a day to one hour, said Nagahara.

“Compared to waterfall development, which took eight months for a single release, agile development releases once every two weeks and every other week, but test automation is required to maintain quality,” he said.

Streamlined processes, software bots and automation play an important part in scaling up. But they can only take you so far. They must be matched with hiring the right people and upskilling those already there in AI and RPA tools as well as making it easy for them to deploy software bots in their own areas.

Workforce consultancy Kelly released the 2023 Kelly Global Re:work Report to highlight the gap that exists between the current workforce and what is needed to operate AI at scale.

“Three in four expect AI to cause disruptive change and reshape the workforce,” said Ed Pederson, vice president of Innovation and Product Development, Kelly. “However, 80% say a lack of skills are impacting enterprise digital transformation.”

The lack of people with the requisite RPA and AI smarts leads to trouble in reaching revenue expectations, holds back innovation, delays new product releases, and slows overall progress in automation and digitalization. According to Pederson, there are up to 10,000 open RPA positions.

Pederson recommends in-house and certification training as ways to overcome this problem. According to IDC, 35 to 40 hours of training per employee in AI and automation translates into a project success rate of 80% or more. However, less than 25 hours of training sees the success rate fall below 45%.

Daniel Saroff, an analyst at IDC, agrees that training and upskilling are vital elements in being able to scale up AI and automation programs. He also stresses setting AI policy to cover compliance, transparency, accountability and data protection, as well as the need for a comprehensive Ai strategy.

“Crafting a comprehensive AI strategy with prioritized use cases is essential to aligning the organization’s efforts to business impact (both short and long terms),” said Saroff. “The AI strategy should include the rules or guidelines for genAI [generative artificial intelligence] proofs of concept (POCs) and it should incorporate the results of the POCs to improve the strategy.”

This approach buys time for the organization to upskill its workforce while making progress on low-hanging AI fruit and setting a timeline for automation scale up.