RPA promised to solve complex business workflows. AI might take its job

Any modern enterprise is an extremely complex operation behind the scenes, which requires someone or something to coordinate the thousands of small tasks from payroll to invoicing that take place every hour. Over the last several years, enterprise software tools designed to automate those processes were seen as the answer to that problem — until the generative AI boom changed everything.

Robotic process automation (RPA) allows companies to configure software "robots" to automate mundane, rules-based tasks across applications and systems, and companies building RPA tools have received a lot of investment over the past several years. However, the technology is reaching an inflection point as enterprises increasingly look to get more than just productivity gains from their prior investments.

According to Forrester, the market for RPA software and services is expected to grow to $22 billion by 2025, but growth is slowing down as companies increasingly look to AI to solve these problems. After all, RPA is nothing but a tool that helps automate repetitive computer tasks and business processes in an organization, increasing productivity and saving costs; just like AI promises to do.

“RPA has proven to be a flexible and effective driver for both digital transformation and efficiency, making it a popular choice for enterprises seeking to enhance their operational capabilities," said Vijay Pandiarajan, vice president of Salesforce Automation in charge of the MuleSoft RPA product. "It has emerged as a valuable tool that enables organizations to automate repetitive tasks and streamline processes, thereby enhancing the overall digital experience for customers.”

Like many emerging enterprise technologies, RPA appeals to CIOs looking to move faster without increasing headcount, said Amit Saxena, general manager and vice president of ServiceNow's Automation Engine. However, like many emerging enterprise technologies, RPA requires specialized skills to implement properly and can be very expensive at a time when IT managers are being asked to cut costs.

"We’ve seen customers re-frame their thinking about RPA, which was a cutting-edge technology at one time," Saxena said. "In the early days of the RPA boom, we saw customers rushing to adopt bots to automate repetitive, rules-based tasks, which served their purpose for a while. But now, RPA [has evolved] from a back-office technology to a driver of enterprise-wide digital transformation when paired with intelligent automation."

One bot at a time

RPA at first seemed like a godsend for companies that wanted to quickly modernize their legacy systems.

Applications built decades ago often come with very basic user interfaces designed by engineers for engineers, and many were built as a monolithic block of code targeting on-premises data centers, Pandiarajan said. RPA allowed those companies to extract crucial data from those systems and manipulate it with modern tools.

"For instance, insurance companies often rely on critical policy management systems that are outdated but essential for their operations. RPA can simplify the interaction process, making it easier for insurance agents to provide policies to customers without the need for extensive training on the legacy system,” Pandiarajan said.

RPA also helped companies streamline day-to-day operations, such as onboarding new employees, said Rudy Kuhn, lead evangelist at Celonis.

When a company makes a new hire, "the employee is created with their master data in Workday, and as soon as they have signed the contract via DocuSign, they are created in all systems such as Mail, Slack, [and] CRM, and are sent their access data. Without any manual work, they are part of the company and can start working,“ he said.

Most RPA tools are not best-in-class when it comes to other capabilities such as rules engines, process modeling, and connectivity solutions.

Another Celonis customer in the energy industry uses RPA in combination with process mining to identify how customer satisfaction scores can lead to better sales outcomes, Kuhn said. Unhappy customers are more likely to balk at price increases, and identifying those customers allowed the energy provider to customize its sales proposals rather than raising prices across the board, he said.

But in the process of solving some common business problems, RPA introduced some new business problems. For example, RPA requires integrating various components, none of which it excels at individually.

"Most RPA tools are not best-in-class when it comes to other capabilities such as rules engines, process modeling, and connectivity solutions," Pandiarajan said. "To unlock the true power of RPA, it is crucial to combine bots with best-in-class integration, workflow management, and API management for enhanced security and governance."

RPA is also a scripted solution that lacks the intelligence to handle errors or exceptions effectively. That makes it difficult to use inside companies that generate a lot of unstructured data.

“The limitations of earlier RPA projects, which include high operational costs and technical debt, have led enterprises to seek more advanced, cost-conscious solutions capable of handling more complex business needs with greater efficiency," Saxena said.

AI … fixes this?

Like most enterprise software companies, RPA vendors are experimenting with generative AI technologies. "Generative AI is poised to amplify the accessibility and scalability of RPA, mitigating the predominant obstacles to entry, namely the need for specialized developers and the risk of bot failure," Saxena said.

Alex Astafyev, co-founder and chief business development officer at ElectroNeek, agreed that generative AI will make it much easier to use RPA technology inside companies that have their expensive software developers committed to other projects.

"While many RPA platforms follow a low-code approach, thus allowing non-tech users to build automation bots, the knowledge of variables and programming logic might be needed in certain cases. Integration of AI lowers the barrier even further," he said.

In the near future, it is conceivable that you could ask a bot about the status of a customer's package in the fulfillment process, and the AI would understand the process and provide real-time updates.

Generative AI technology will also allow RPA systems to deal with complicated problems described with natural language inputs, Pandiarajan said.

“In the near future, it is conceivable that you could ask a bot about the status of a customer's package in the fulfillment process, and the AI would understand the process and provide real-time updates," he said.

But while generative AI should improve the effectiveness of RPA inside enterprises, experts warned that there are still a few pitfalls that buyers should keep in mind when considering introducing the technology.

Automating tasks might lead to faster business outcomes, but companies need to consider whether those business flows make sense to automate or if they even need to be there at all, Kuhn said.

"For example, if a bottleneck is identified with process mining, the first question should be whether the bottleneck can be eliminated. In my humble opinion, elimination is the best solution for challenges in processes," he said. 

Sometimes that's easier said than done, especially when it comes to the checks and balances that are required by risk management strategies, Kuhn said. "But you can try to simplify the task, and once it has been simplified, you should standardize it so that the task is always as simple." 

Companies should also ensure that they design their RPA workflows in a way that can accommodate shifting business priorities down the line, Pandiarajan said.

"From an architectural perspective, we want to ensure everything is broken down into reusable components wherever possible," he said. That way, when things inevitably break in the future or when technology evolves, you can swap out parts of the automation rather than rebuilding the entire automation."

And no company can afford to ignore security concerns when bringing automation in their mission-critical business functions, Pandiarajan said.

"If you have all these building blocks, how do you keep them secure? A very common model for RPA bots is to gate access to the bot using an API gateway, so you need best-in-class API security and governance to be able to ensure that the bot is used correctly and protected from bad actors," he said.