Big companies have invested big bucks in recent years into software designed to automate routine back office tasks, many of which involve simply cutting and pasting data from one software application to another or using drop down menus to populate database fields.
Known as “robotic process automation” or RPA, these kinds of software “robots” aren’t AI. Some are little more than souped-up versions of Excel macros, recording mouse movements and keyboard strokes. Others use “if-then” rules to help software complete a workflow.
And yet businesses are now estimated to spend more than $6 billion per year on RPA software, according to technology analytics firm Forrester Research, a figure that is growing at a double-digit percentage clip. UiPath, one of the leading players in the RPA field, is valued at $13.5 billion. Appian, Blue Prism, and IBM also offer RPA solutions.
Now a tiny startup based in Munich, Germany, called Interloom thinks it can disrupt this entire market by rebuilding process automation on top of the new wave of large language models and generative AI assistants. Air Street Capital, the London-based venture capital boutique that specializes in early stage AI investments, just gave the company $3 million in seed funding to start making good on that vision.
Fabian Jakobi, Interloom’s cofounder and managing director, is a serial entrepreneur. In 2021 he sold Boxplot, his last company, to Hyperscience, a New York-based AI software company that specializes in extracting data from unstructured documents. Jakobi thinks similar AI methods can be used in the future to extract information from video recordings, call logs, notes, and more, enabling AI software to learn how professionals actually work. Then AI agents, based on the same underlying AI methods that power today’s large language models, can be used to automate many parts of these tasks.
This could allow for much higher-value tasks to be automated than can be addressed with today’s RPA, which only works well for tasks with highly routinized and repeated workflows. According to Jakobi, current RPA tech is capable of automating about a third of business tasks — a limitation that helps explain why reports by consulting firms EY and Deloitte have found that a majority of RPA projects either fail completely or never live up to their potential.
Rather than starting from idealized workflows, Jakobi says that AI software can be trained on what a company actually does in real world situations. The AI can intuit what the right workflow is for that particular situation, instead of adhering to an overly standardized and rigid template.
He gives the example of drafting and sending out a non-disclosure agreement as part of commercial deal. With today’s RPA, a company might create a rule that any deal worth more than $100,000 requires that the counterparty be sent an NDA. The workflow of filling out the NDA template for this particular deal is automatically handled by the software robot.
The problem with such rules is that they are far too inflexible to capture real business logic, Jakobi says. What about a deal valued at $98,500? Most businesses would probably still want that covered by NDAs even though it is below the threshold it set for the robot. Modern LLM-based software excels at capturing, from past data, tacit knowledge, which includes a lot of professional judgment about things like when an NDA is required.
The tasks most suitable to this kind of automation would include procurement and risk assessments, customer onboarding, mortgage and insurance claims processing, and managing import and export logistics documentation, according to Interloom.
While Jakobi says humans will still be needed to act as a quality control check on Interloom’s software robots, he believes that for many processes, AI robots such as the ones they are building will be able to increase the output an employee can produce in a given amount of time by 30 times.
Interloom, which currently employs just ten people but is rapidly hiring, plans to target Germany’s large “mittelstand”—the medium-sized industrial companies that are a bulwark of the German economy—as its initial customer base, with plans to expand globally, including to the U.S., soon.
“Every sector of the economy is eventually going to be rebuilt AI-first,” Nathan Benaich, Air Street’s founder and general partner, said. “From their experience at Boxplot and Hyperscience, the Interloom team have a uniquely intimate understanding of automating complex workflows. This makes them best-placed to build the process infrastructure that will underpin AI-enabled productivity gains for business.”
