Our Investment in Interloom
Building the foundations for AI-first enterprise automation
Every year, the State of AI Report covers significant leaps forward in the field. Our most recent instalment charted GPT-4’s stunning performance against popular NLP and human task benchmarks, the use of multi-modal models in self-driving cars, and new foundation models for robotics.
However, when we step outside the research lab, it can be harder to see the impact of all this progress. This is particularly stark in the world of knowledge work. Despite AI systems being able to match or outperform humans on many comprehension and reasoning tasks, very little basic work has been automated. Instead, the ‘cutting edge’ in enterprise is a series of underwhelming copilot demos.
Currently, when a large enterprise wants to automate processes, it’ll usually pay consultants who will present a series of neat process flow-charts they generated by dictating how they think employees should work in a perfect world. If advanced, they might have used software to generate said neat process flow-charts. Unfortunately, process automation by hard-coded workflow usually breaks down in the real world. This is because most enterprise processes vary in thousands of subtle ways, branch across different (and often custom) software tools, and involve judgment calls that are intuitive to humans but difficult to codify. This is why, so far, 70% of enterprise processes have never been automated at all. Not even badly.
It's against this backdrop that we’re excited to lead a $3M day 1 financing round for Interloom as it emerges from stealth today. The Interloom team is rejecting the industry standard of piling more layers or plug-ins on top of legacy technology. Instead, they’re building a cloud and model-agnostic foundation for the era of AI-first enterprise automation.
Their system pools all the data from across an enterprise and infers how processes work from historic behavior, conversations, and task notes. Instead of telling employees how they should work, it observes how processes actually work. This is the complete inverse of the hard coded workflows that are currently imposed on enterprises. Interloom does this through a mixture of AI-powered task mining, the application of their proprietary knowledge graph, and AI agents.
The Interloom team is perfectly placed to combine these complementary technologies, bringing deep domain experience earned through many years at the coalface of enterprise automation. I first met Fabian (CEO) and his team when they were building Boxplot, a company that created AI-augmented knowledge graphs for the enterprise. They had bootstrapped the company to a point where it was serving Fortune 500 customers.
Within weeks of our first conversations, they were bought by Hyperscience, in the company’s first ever acquisition. Hyperscience was the first company to solve the long-standing challenge of AI-enabled handwriting recognition and had built up a customer base of multinationals and government agencies, most recently securing a $100M Series E. Boxplot’s graph-based operating system became central to their automation offering.
At Air Street, we believe that every sector of the economy is going to be rebuilt AI-first. The era of add-ons and plug-ins for legacy technology will be short-lived. As a result, companies working on enterprise automation and the translation of human knowledge for AI will play a critical role. That’s why we’ve already backed Adept, V7, and Intenseye. We’re looking forward to working closely with Interloom as they build out their engineering team and gear up to launch their first product later this year.
The team is hiring! Check out openings here.