By Ralph Gammon, Senior Analyst, Software
October 28, 2024
At last week’s annual Forward customer and partner event, UiPath announced its new strategic direction focused on agentic automation. On a broad level, UiPath is attempting to transition its robotic processing technology for automating repetitive tasks to more dynamic automation of less structured processes. Its new agents are designed to leverage AI to automate decision making in multi-step processes. In a metaphor, UiPath compared this to giving brains to its dumb robots.
Founder and CEO Daniel Dines described the initiative as the start of UiPath’s second act. “Act one was focused on structured information and rules-based processes,” he said during his keynote address at the MGM Grand in Las Vegas. “Ten years ago, we started with precision computer vision models for screens, so in a way AI was at the foundation of our technology. From there, we went on to IDP and created dedicated models for processing invoices, POs, and more—as long as information is structured or semi-structured, we could create a model to deal with it.”
Dines explained that UiPath’s Act 2 leverages AI to address unstructured information and processes. “Our limitation has been the unstructured parts of a project. Historically, we’ve had to isolate the structured parts and put humans in the loop for unstructured parts. Now we are leveraging AI to try and reduce HiTL.”
UiPath is not the first software vendor to discuss automating processes with agents. Its competitive advantage is a large number of customers already running its automation platform in enterprise environments. “We have made RPA reliable and scalable at the enterprise level,” said Dines. “We already have the ability to run technology that imitates people. Gen AI falls into the same category. Robots are low skilled, and agents more highly skilled, but they both belong in same platform, because the unstructured parts of business processes are married to the structured parts.”
At Forward, UiPath was demoing its new AutoPilot for Anyone technology, which features a natural language interface for building out automated processes. (UiPath had previously introduced AutoPilots for Testing and Coding that also utilize natural language prompts.)
“Agentic automation is all about how you control Generative AI,” said Dines. “An agent receives a set of instructions through a prompt—the user describes a goal. Then the agent has access to a set of tools it can dynamically choose from to achieve that goal. Orchestration is the key; it enables agents to pull in information from multiple sources and kick off sub-automations, as well as bring in humans when they are needed. Agents also have memory, and they can learn from instances where human interaction is needed.”
An example presented at the conference involved utilizing an agent to resolve a dispute related to invoice processing. In the demo, when information captured from an invoice didn’t match information on the related PO, the AI-powered agent was able to read through relevant vendor agreements to see if there was a policy in place to allow the invoice to be paid anyway. If not, the dispute could be escalated to a person. The person would then follow up with an e-mail to the supplier asking about the
discrepancy. If a similar problem occurred in the future, the agent would be able to automatically compose a similar e-mail to the same contact.
Crossover with IDP
There is strong crossover between agentic automation and Intelligent Document Processing (IDP). IDP has been one of the first widely deployed use cases leveraging AI. But, in early IDP applications, AI’s use has been primarily limited to document classification and extraction. The widespread introduction of Generative AI and LLMs into the enterprise software mix has the potential not only to extend automated extraction to a wider breadth of document types (especially complex, unstructured docs), it also potentially enables AI to be leveraged for decisions and to complete tasks, thus extending AI from document capture to end-to-end process automation.
Infosource has always seen broader automation as the end game for what started out as Capture and has evolved into IDP. Aligned with our vision has been the recent growth of case management type applications in the Capture & IDP software market. Case management involves collecting information that leads to a decision, historically made by a person. This includes applications such as onboarding (customer, client, patient, etc.) and claims management.
For the five-year period 2019-2023 (as recently published in our 2023-2024 State of the Global Intelligent Document & Data Processing Market Report), Case Management was the fastest growing of the use case segments we track, with a 15% CAGR. In 2023, it made up a little more than a third of the market for IDP software revenue.
Source: Infosource State of the Global Intelligent Document & Data Processing Market Report
Case Management is also a natural starting point for agentic automation due to its workflow-driven nature. One of the growth drivers for Case Management use cases in IDP has been the ability to utilize AI to apply automation to a greater breadth of document types. Adoption of Gen AI and LLMs will only widen that breadth and the introduction of multi-modal AI models into agentic automation applications will make automated processing of unstructured input more diverse.
UiPath has been extremely successful in expanding from automating processing of structured information (RPA) into IDP, emerging as a market leader in just a few short years. The ISV seems to have a handle on enterprise automation, and back-office focused agents are a natural next step for them. That said, it’s a big step and it brings UiPath into a space where multiple markets and technology leaders are currently converging, so the stakes are high.
I’ll close with a statement that UiPath CMO Bobby Patrick made during his portion of the keynote. “This year’s Forward event represents a seminal moment in the history of the company. The first Forward was held in New York City in 2017, and that was pivotal because we were just starting out and much smaller than our competitors at the time. But this is the most pivotal Forward, because we are now going to make AI actionable and put AI into workflows where it has been lacking until today.
“In 2017, we had three pillars, which were ease of use, scalability and security. Our fourth pillar was a vision for AI. When we outlined our strategy, at first it was robots for every person. However, it turns out it will be an agent for every person that will augment every worker with decision making AI capabilities.”
