Connect with us

NEWS

Marble Raises $3.75 Million to Map Work for Buyout Owners

Marble Cognition raised $3.75 million to watch how staff work and turn those steps into AI agents that private-equity owners can run.

Published

on

Marble Cognition has raised $3.75 million to watch how staff do office work, then give those jobs to AI agents. Heartcore and Seedcamp co-led the Munich pre-seed, with angels from Datadog, Personio, McKinsey, Mistral AI and Lovable.

The product is a desktop app that records screens, clicks and keystrokes in business software, then builds process maps those agents can run. Named customers already include Mutares, a Munich buyout firm that buys companies in trouble, along with EGYM and Enam.

Heartcore and Seedcamp Put $3.75 Million Behind Watching Work

Felix Rösner, co-founder and chief executive, told backers the round closed in under two weeks. Valuation was not disclosed. The money is meant to scale the computing that runs Marble’s models, grow the engineering team in Munich, and launch in the United States.

Rösner and Christophe Schmit, the chief technology officer, founded the company in late 2025 after meeting at Munich’s Center for Digital Technology and Management. Rösner had worked at 11x, which builds AI sales agents. Schmit, a former MIT researcher, had worked at Enam, which Marble also lists as a customer.

Every company we talk to wants AI to take over real work. But nobody can tell the AI how that work is actually done. So we solve that first. Our Coworkers watch how the work happens before they take it over. They do it the way your best people do.

Felix Rösner, co-founder and CEO, Marble Cognition

That line is the pitch. Software that can run a task is no longer rare. A reliable account of the task, including the messy bits people leave out when they describe their own jobs, still is.

Private Equity Already Has a Line on the Product Page

Marble’s site treats buyout firms as a customer type, next to insurers and finance teams. The work it offers them is value-creation planning, post-merger integration, and running portfolio companies. That is a different buyer from a software team looking for a chatbot.

Mutares SE & Co. KGaA sits in the same city. It is a listed holding company that buys businesses in transition, tries to stabilize them, and sells them on. It has been pushing further into the United States, including through a Chicago office. For a firm that has to understand a newly bought company before it can cut costs or prepare an exit, a map of how the work is actually done is the asset, not the chatbot that follows it.

The only numbered result Marble has put on the round is from that world, and it is not Mutares. At an unnamed private-equity-owned financial-services firm in the United States, a Marble Coworker was set up to draft settlement offers for a specialist to approve. Seedcamp said the agent cut handling time per case by 90%, and that the team handled 50 percent more cases in the first month with no new hires. Those figures are Marble’s, from one deployment.

Whoever holds that map decides which jobs get handed over next. Buyout owners already pay for that kind of visibility when they send in operating partners. A product that builds the map by watching staff, then keeps running the steps, is a way to keep the know-how after the people who held it move on.

Rösner said the company also works with large tech firms in Munich and Germany, private-equity firms in Europe, and PE-owned professional-services shops in the United States. He wants the book split evenly between the US and Europe. EGYM and Enam sit on the tech side of that mix. The quantified case, and the product line on the website, sit on the buyout side.

What the Desktop App Records

Marble installs a desktop client on the machines where the work happens. It watches business applications, takes in the screen, the clicks, and the keystrokes, and turns that stream into a structured process map that an agent can read the way it would read code. The agents, called Marble Coworkers, then run the process in the same tools the team already uses, with a person still in the loop when a judgment call is required.

Rösner argues that most vendors walk in with a use case and a hours-saved slide and have no idea whether that use case is a real bottleneck, how the work runs, or what the saving would be. Marble says it starts from the current state and only builds where the observation shows a real load. On its site, the company says that in a few weeks it can show a team how the work actually runs and which pieces to hand over first.

Humans are bad narrators of their own jobs. Ask someone how they use a spreadsheet and you usually get a cleaned-up version, not the missing files, the side channel in email, and the delay while they wait on a colleague. Seedcamp, which previously backed UiPath, is underwriting that observation as the missing step before another wave of automation software.

The People on the Other Side of the Screen

A product that learns by watching staff has to clear a higher bar than a chatbot. Marble’s home market makes that concrete. German works councils typically have a say over technical systems that monitor how people behave or how they perform. Rösner said change management is the hardest part of the sale, because watching staff will always draw more pushback than most tools.

The company has written a set of limits and put them on the product site. Personal data is supposed to be stripped before a person or a model sees it, including when the system runs in a customer’s own environment. Rösner said private data is stripped on the local network before it reaches a language model, and that records for German customers stay in Germany, with other European customers stored in Europe.

THE STATED PRIVACY LIMITS

  • Who hits record: The company says employees decide when Marble records and can pause capture at any time.
  • What is in scope: The app is meant to watch business software only, not private apps, private browser tabs, or activity outside working hours.
  • What gets hidden: Names, addresses and account numbers are redacted on screen before processing, as in Marble’s sample invoice INV-5203 for €1,860.00, where the work stays readable and the person does not.
  • How it is studied: Analysis is supposed to run in aggregate, with nothing traced back to one person, and staff able to see outputs, delete their own data, and block apps.
  • Where it runs: Marble says it is SOC 2 certified, offers GDPR-aligned EU hosting, and can keep full anonymization inside the customer’s own environment.

Those are the company’s rules, not a regulator’s finding. Staff who are asked to train a system that may later do their tasks have a reason to test every one of them. Adoption still depends on people agreeing to be observed, which is a different sale from a licence for a model.

McKinsey’s 57 Percent Meets Processes Nobody Wrote Down

Marble leans on a November 2025 McKinsey Global Institute paper, Agents, Robots, and Us, which found that today’s tools could in theory automate about 57 percent of US work hours. Digital agents account for 44 percentage points of that ceiling and robots for 13. McKinsey presents the number as technical potential, not a forecast of jobs lost, and notes that most organisations are still in pilots.

Rösner’s claim sits beside that ceiling, and it is not a jobs forecast either. Among the companies Marble works with, he says about 90 percent of core processes have never been written down. They live in inboxes, spreadsheets, and people’s heads. If that share is even roughly right, the bottleneck is not the model. It is the missing picture of the work.

THE FIGURES BEHIND THE PITCH

Measure Figure
Core processes never written down, at companies Marble works with about 90 percent
US work hours current technology could automate, in theory (McKinsey, Nov 2025) 57 percent (44 percent agents, 13 percent robots)
Drop in handling time per case, PE-owned US finance firm (Marble, one site) 90 percent
Rise in cases handled in month one at that firm, with no new hires 50 percent

The two 90 percent figures are not the same thing. One is Rösner’s reading of how little is documented. The other is a time cut on settlement-offer prep at one PE-owned US finance shop. The 50 percent is throughput at that same shop, still with a specialist signing off. None of those numbers has been checked by an outside auditor in the materials Marble has put out with the round.

A UiPath Backer Returns to the Same Problem

Reshma Sohoni, a partner at Seedcamp, tied the cheque to an older bet. Seedcamp backed UiPath, which turned recorded desktop steps into software robots, and Synthesia. She said most companies want to automate more and that mapping how the work gets done has always been the hard part. Marble, in her telling, fills that gap and gives companies context for AI projects.

Max Niederhofer, a partner at Heartcore, went further and named the asset.

The scarce input for automation is not the models, which are commoditising. It is building a precise record of how work gets done. Marble learns a whole company’s processes, anonymises the data, derives SOPs and deploys agentic AI against them.

Max Niederhofer, partner, Heartcore

That is the second-order product hiding under the coworker branding. Once a firm has standard operating procedures derived from live work, it can run agents against them, brief a new hire faster, or make a portfolio company less dependent on the people who used to carry the process in their heads. Models get cheaper. The record of how this company, this week, actually closes the books does not.

Angels around the round fit that reading. A Datadog board member knows monitoring at scale. A Personio founder knows HR software inside European companies. Executives from McKinsey, Mistral AI and Lovable add consulting, model, and product-building experience. None of those angels was named in the announcement.

The Money Goes to Inference, Munich and the United States

Marble is still a young company. It has not published a headcount, a valuation, or contract sizes for EGYM, Mutares or Enam. A named customer at pre-seed usually means an early deployment, not a mature rollout. The US launch is a plan, though the financial-services case shows Marble has already done work there.

Rösner groups the competition in three piles: consultancies that map processes by interviewing everyone, vertical AI shops that send their own engineers to do the same and then build the tool, and software firms that already sell process mapping. Marble’s wager is that watching the desktop beats all three, if staff will accept the watcher and if the map is good enough on the ugly exceptions, not only on the happy path.

The next test is not another model release. It is whether a works council, a US operations chief, and a portfolio manager will let the app stay on the machine long enough to draw the map, and who owns that map once it exists. The $3.75 million is there to keep the watchers running while Marble finds out.

Harry is the editor of IAQABA, an independent publication he owns and runs. A decade in journalism, beginning as a reporter and now as the editor of his own titles, has left him with a clear test for what deserves a story: it has to change what a reader knows or decides, and it has to rest on something he can point to. That rules out recycled press releases, forecasts with no data behind them and rumours that no document supports. It leaves room for a great deal, and the site covers news, business, science and technology alongside sports, entertainment and lifestyle, with travel, auto and gaming given the same standard rather than lighter treatment. Sources are primary wherever possible: the regulator's filing, the company's own statement, the transcript, the dataset, or the product on Harry's desk. Figures are checked before they are published and rechecked if a reader questions them. Mistakes are corrected under a published policy. Readers across the world can reach him directly at support@iaqaba.com.

Click to comment

Leave a Reply

Your email address will not be published. Required fields are marked *

Trending