Two German startups announced funding rounds on 2 and 3 November 2026. Neither is pitching a chatbot that does everything. Each is selling AI that takes one specific slice of office work off people’s hands. Munich-based Marble Cognition has raised $3.75m for agents that learn from what real employees do, and Berlin-based Conlio has secured €6.4m to handle the paperwork behind freight transport.
An agent that watches first
Marble Cognition announced its pre-seed round on 2 November. Heartcore Capital and Seedcamp put in the money, alongside business angels with ties to McKinsey, Mistral AI and Datadog. The company was founded by German AI engineers Felix Rösner and Christoph Schmitt.
The product, Marble Coworkers, works in two stages. First, a standalone desktop app is installed on the employee’s computer. It records everything the person does in the company’s software: clicks, keystrokes, screenshots. From these recordings the system builds a structured map of the process, and the AI agent then uses that map to run through the same steps by itself.
Anyone who has ever brought automation into a department will see why this matters. Often the most expensive part of such a project is documenting the process. That usually means interviewing staff, and even then half the exceptions only come to light after launch. Marble’s idea is to skip this stage altogether and learn on the job instead.
Classic robotic process automation (RPA) takes a different approach. The bot replays a scripted sequence and often stumbles when something in the interface changes. Going by the company’s description, an agent working from a process map should cope with small changes like these more calmly, though for now the only way to check is against Marble’s own case studies.
Screen recording raises questions about data straight away, and Marble appears to know it. The company says recordings are anonymised, clients can review them before they reach the models, and processing happens in the customer’s private cloud or entirely on-premises. In Germany that is the price of entry rather than a nice extra. An employer that monitors staff screens without the works council’s agreement is asking for a dispute.
The startup names EGYM, Mutares and Enam as its first clients. The most detailed case study comes from the US. Marble’s agents drafted settlement proposals for an American financial company, and a person still made the final call on every case. According to the startup, processing time fell by 90%. Staff handled about 50% more cases, and no new hires were needed.
Consignment notes, reconciliation, payments
A day later, on 3 November, Conlio announced a €6.4m round led by Piton Capital. Haufe Group Ventures, WALTER Group’s WaVe-X and FIEGE’s F-LOG Ventures joined, along with existing investors System.One, Puzzle Ventures, Booom Ventures and Yellow. One detail stands out. The round includes funds belonging to two logistics groups, which are in effect future buyers of the product.
Conlio was founded by Claudio Consul in 2023. Its system handles the documents that road freight runs on, from CMR consignment notes to proofs of delivery and carrier invoices. The AI checks them against each other, and the system then takes each case through to reconciliation and payment by itself. A CMR note accompanies most international road shipments in Europe, so there is a huge amount of paper involved. The company says more than 90% of checks are completed automatically.
It sounds dull. But this is exactly where freight forwarders lose margin. Europe’s road freight market is estimated at around €430bn, and every shipment brings a pile of paperwork with it. A mismatched weight, date or registration number means a manual check, a call to the carrier and a late payment. Conlio claims its system can lift a freight forwarder’s EBIT by up to 30%. For a company with €100m in revenue, the startup puts that at more than €1m in EBIT a year.
The ‘documents first, money later’ rule has long been standard across digital services, and online casinos are no exception. Winnings from slots at Spinlander Casino, or from the live casino section of any other online casino, are usually paid out only after identity checks, once the operator has verified the player’s documents and payment method. A freight forwarder handles an order of magnitude more paperwork, but the logic is the same. Until the documents match, no money moves.
Conlio will spend the new funding on its product, hiring and expansion into other European markets.
What the two deals have in common
On their own, these are two ordinary early-stage rounds. Taken together, they show fairly clearly where money in applied AI is heading. Not long ago, investors were happy to back yet another wrapper around a large language model. Now, judging by rounds like these, money flows more readily to systems that learn a specific process and work inside software the client already has.
The human doesn’t drop out of the loop, though. At Marble, the agent drafts a proposal and an employee decides. At Conlio, the machine deals with routine checks and disputed cases stay with people. That set-up is easier to sell. The customer doesn’t need to believe the model never gets things wrong. They only have to work out how many hours the system takes off the department’s workload.
The metrics help too. Both startups talk in numbers a CFO can check personally, such as processing time, cases per employee and EBIT. In the German and Austrian B2B market, where buying software means several rounds of sign-off, that kind of language arguably works better than any demo.
The German market has its own motive as well. The country has plenty of mid-sized companies where one accounts department or claims team serves hundreds of clients, and people for these roles are getting harder to find. Marble’s case study, in which a department took on about 50% more cases without hiring, speaks directly to that problem. And the freight forwarders Conlio is promising an extra million in EBIT run on thin margins and count every manual task.
The weak spots
For now, both companies are relying on their own data. Marble’s 90% time saving comes from a single project, and nobody knows how well it will transfer to other processes. Conlio’s claim is ‘up to 30%’, and the real gain will depend on how much manual work a given freight forwarder was doing to begin with.
Marble faces an extra risk of its own. Wherever a German company has a works council, recording clicks and screens will almost certainly need its approval, and data protection lawyers will want a say as well. On-premises processing deals with some of these issues. The rest depends on whether the startup can persuade employees that the app is learning the process rather than spying on them.
Then there is price. Neither Marble nor Conlio says publicly how much implementation costs or how quickly it pays for itself. A pilot on a single process and a company-wide rollout usually differ in cost several times over, so the first client reports will be more interesting than the press releases.
What comes next
Marble will use its pre-seed money to refine the product and sign up clients. Conlio is already scaling and plans to expand beyond Germany. If the first deployments bear out the headline figures, both could raise follow-on rounds quickly, and the logistics funds among Conlio’s investors suggest the buyers are already waiting.
A year from now it should be clear whether Marble’s 90% and Conlio’s promised EBIT gains hold up on other people’s processes, not just on pilots the companies picked themselves.