WORLDTECHMunich's Marble Cognition raises $3.75M for AI coworkers that learn workflows
Munich-based AI startup Marble Cognition has secured $3.75 million in pre-seed funding to build AI coworkers that first learn how office work is actually carried out, according to reports by Tech.eu and EU-Startups. The round was co-led by Heartcore and Seedcamp.
Funding round and investors
Marble Cognition, known as Marble, has raised €3.35 million, equal to $3.75 million, in a pre-seed round. Heartcore and Seedcamp led the round together. Angel backers taking part include a board member of Datadog, a founder of Personio, and executives at McKinsey, Mistral AI and Lovable.
Learning processes by observation
Marble argues that many business processes go unrecorded, with essential knowledge spread across emails, spreadsheets and employees' experience. According to the company, around 90 per cent of core business processes at the firms it works with are not formally documented.
The startup studies how staff carry out their daily tasks and turns those activities into structured process maps that AI systems can interpret. Its desktop app captures screen activity, clicks and keystrokes in business applications, but not private apps, private tabs or activity outside working hours. The company says everything is anonymised before any person or AI model sees it, employees can pause the recording at any time, and data is analysed only in aggregate.
Customers and early results
Notable customers named by the company include EGYM, Mutares and Enam. Marble reports that at a private-equity-owned US financial services firm, one of its coworkers cut handling time per case by 90 per cent, and the team handled 50 per cent more cases in the first month without hiring anyone. The coworker prepares each settlement offer for a specialist to approve.
Founding and plans
Felix Rösner and Christophe Schmit founded the company in late 2025 after meeting at Munich's CDTM programme. The new funding is earmarked to scale inference, expand the engineering team in Munich and launch in the United States.
Open questions
- Do the reported customer results hold beyond the single example cited?
- Will the anonymisation and monitoring measures be independently verified?