Startup aims to upend old-school RPA with large action model
With its GenAI-based platform, Orby AI aims to provide enterprises with generative process automation and AI agents to automate a range of business workflows.
Generative AI technology and its quickly sprouting offshoots have spread to far-flung regions of enterprise software, including the unglamorous but long reliable world of robotic process automation.
Now, just about every RPA vendor, including Automation Anywhere, Blue Prism and UiPath -- the so-called big three independents that have dominated the field -- is integrating aspects of GenAI into its platform to automate business processes.
But a cluster of upstarts is challenging the established vendors with native GenAI technologies. These companies aim to dramatically update RPA by infusing it with intelligence to enable enterprises to automate complex processes faster, cheaper and at greater scale than traditional RPA.
Enter Orby AI, a 2022 startup founded by Bella Liu, a former director of product management at UiPath, and William Lu, a former Google head of engineering.
While LLMs generate text, images and audio, LAMs generate processes and can act on their own using AI software agents.
LAMs are a relatively new category that sprung up about a year ago. They use neuro-symbolic programming, going beyond the neural networks underlying most LLMs to include symbolic representation to model actions and how to trigger them.
In June, Orby revealed it raised $30 million bringing its total fundraising to $34.5 million. It secured some large beta users and now is readying to go to market.
"Of course, there are market leaders that have established customer bases and can fast track embedded GenAI capabilities. But I think their existing platforms and architectures are too rigid to make a change as impactful as what Orby is bringing to market," said Mike Leone, an analyst with TechTarget's Enterprise Strategy Group.
Orby AI co-founders Bella Liu, CEO, and William Lu, CTO.
Post-RPA tech
The startup, based in Mountainview, Calif., rarely mentions the term RPA under the theory that the technology, dating to the 1990s, is outmoded. Orby doesn't want to be lumped into the RPA category.
"Orby's greatest challenge right now is being tied to RPA altogether. We've heard so much from customers that haven't seen the promised value from traditional RPA," Leone said. "Orby can deliver simplicity and intelligence that is nowhere near what traditional RPA can bring, but customers still have RPA fatigue."
Orby's platform is based on several GenAI variants, including the large action model for what it calls generative process automation and agentic technology to take action -- in Orby's case, carrying out business processes. Agentic technology refers to AI agents that can act autonomously; they have agency and can act on their own.
Orby's ActIO Large Action Model platform generates business processes for enterprises, such as invoice and contract processing and employee expense auditing.
On the user's behalf
The way Liu explains it, Orby's LAM works on a "observe, learn and automate" paradigm.
The AI observes a user doing their work on the computer and figures out their process. Then, in a few minutes, it automatically generates the scripts and code for the process and presents them to the user, who can choose whether to press a button and activate the automation or reject it.
"If you're a business user, you don't need to know anything about the code. You just do your work and show us," Liu said.
Orby's LAM is multimodal. "The input information it looks at is all your actions data. It looks at your screen, your HTML, your interactions with URLs," she said. "The model will understand the general goal and specific steps of your process and it will figure it out."
"The output of the model is that it will just generate actions on the user's behalf to do this work," Liu continued.
An Orby user
For the last seven months, JLL, a global real estate services and technology conglomerate with a transaction-heavy business, has been testing Orby for automating various applications, including invoicing.
The process involves matching an invoice to a purchase order, and if it matches, submitting it for payment. However, the process is often quite complex, with million-dollar invoices that include long documents -- sometimes 50 pages or more -- and a lot of unstructured text, such as Word documents and emails.
"These multi-page document invoices are hard for traditional optical character recognition or RPA to read," said Bruce Beck, enterprise systems CIO at JLL. "So, the large action model is reading the documents. Then, subsequent to that, it has the data and decides what actions it's going to take against that data."
JLL uses Automation Anywhere as its main enterprise RPA platform, and UiPath in small pockets of its business.
"They have some early stage GenAI, but they're not nearly as far along as Orby is," Beck said, referring to the more established vendors. "When we started testing with Orby, the document management capabilities really caught our attention. It accurately captures the right data that we want and then it auto generates code so we don't need RPA developers."
If JLL decides to fully sign on with Orby -- a strong likelihood, according to Beck -- the company will likely hold on to Automation Anywhere for some time before possibly switching over completely to Orby.
Challenges and potential
But JLL still has some degree of hesitation about going with a new vendor like Orby, Beck said.
"You always worry with a startup if their product is going to catch on or [you are] going to be out there with them as the only client," he said. "Will they succeed? Nobody wants to implement a solution just to find that the company didn't succeed and now you've got to start over again."
While Orby co-founders Liu and Lu have strong tech pedigrees and hold MBAs from elite universities, Orby is essentially the first business venture of their own.
Orby's greatest challenge right now is being tied to RPA altogether.
Mike LeoneAnalyst, Enterprise Strategy Group
That business expertise is one of the areas in which the venture capital investors in the startups come into play, said Dan Twing, president and COO of Enterprise Management Associates, a tech analyst firm.
Most notable among Orby's financial backers is Jeffrey Katzenberg, the former Walt Disney Studios and Dreamworks chairman who co-founded tech venture capital firm WndrCo. Also participating in the Series A round were New Enterprise Associates; Wing VC; and Pear VC, Orby's original seed round investor.
Meanwhile, Orby must contend with other startups in the GenAI coding automation sphere, including Writer, Tabnine, Brevian, LightOn and Tune AI as well as the big three RPA independents and larger GenAI specialists.
However, Orby and other GenAI-oriented startups have a distinct market opportunity because, Twing said, "traditional RPA is garbage technology" that is hard to install and maintain.
"RPA created armies of consultants," he said. "It created a lot of money for Deloitte and KPMG."
Another advantage for Orby is its relationship with Google, which has provided discounted cloud compute credits for the startup and promotes Orby as a partner, Twing noted.
Shaun Sutner is senior news director for TechTarget Editorial's information management team, driving coverage of artificial intelligence, analytics and data management technologies. He is a veteran journalist with more than 30 years of news experience.
