Tomáš Čupr tried to introduce AI agents at his online grocery startup Rohlik (he’s moved to chair and it’s now heading to an IPO). He didn’t get very far. He discovered that much of the company's operational knowledge just wasn't documented anywhere.
"Companies don't really understand their processes," Čupr told me. "There's so many things that live in people's heads, and people function as a sort of ‘human API’.”
His new startup, Duvo has now raised a $32m Series A led by Balderton Capital, bringing its total funding to $48 million. His bet is that he can target both large enterprises and the management consultancies advising them with a brand new approach to AI.
Existing investors Index Ventures, Northzone and Credo Ventures also participated. The funding will support Duvo's expansion into the US, initially targeting retailers and consumer goods companies.
Founded in 2025 by former Rohlik Group entrepreneur Čupr and his former head of AI, Duvo has developed what it calls an "Enterprise Execution OS", designed to automate business processes across disparate enterprise systems, including ERPs, spreadsheets, supplier portals and legacy applications.
Duvo's approach to getting at this ‘human API’ is to use AI agents to interview employees, analyse how they work and build detailed process maps. It then deploys agents to execute those processes across existing software systems, using APIs where available and interacting with screens where necessary.
But the technology also places Duvo in an interesting position relative to the management consultancy industry.
Čupr told me that the ‘process-discovery’ work traditionally carried out by firms such as McKinsey and Deloitte can cost enterprises millions of dollars. And even months to figure out. Duvo claims it can compress that work into weeks, before, crucially, automating the resulting processes.
"For consultants, we are compressing that discovery time," he said. "The discovery that would usually take many months of these junior people asking junior people and drawing these process maps, we can compress that into a couple of weeks."
Once again, we must all weep for those former junior hires, right?
So yes, that threatens a labour-intensive part of the traditional consultancy business model. And now they have a way to deliver results faster.
"Long gone are the times where a large consulting firm would walk in with the old school approach, trying to charge for the same amount of work that they used to charge," Čupr said. "Even consultancies need to reinvent themselves, and this is what we're enabling now." Got that, Deloitte?
Rather than positioning Duvo purely as a competitor to consultants, Čupr sees them as a distribution channel and implementation partner. The company now has an alliance with Deloitte Central Europe, which combines its expertise with Duvo's process-mapping and automation platform. Appropriate, given Čupr is Czech.
"We don't see it as either or," Čupr said, arguing that consulting firms can improve results by bringing broader strategic expertise, while enterprises with strong internal teams can use Duvo directly.
Indeed, Čupr’s longer-term ambition is to make enterprise AI automation something companies can buy and deploy themselves. No more expensive teams of specialist engineers.
"There's so many things that live in people's heads, and people function as a sort of ‘human API’.”
Could that mean the end of the Forward Deployed Engineer (FDE) movement?
"If the product is built well, it should be self-deployable in a way," Čupr told me, comparing the experience to buying conventional enterprise software. "I think there was maybe a brief period in human history where AI needed [large numbers of] people to be deployed. I think we're past that."
Duvo claims to have unlocked more than $100 million in value for retail-style clients including Morrisons, Holland & Barrett, Boden, Notino and Allegro.
At Notino, the European online beauty retailer, Duvo claims its platform reduced a supplier-bonus reconciliation process from 70 days to 15, cutting specialist labour from around 120 hours per cycle to fewer than 10.
The company is also betting that enterprises will increasingly demand control over where their AI models run. While Duvo uses Anthropic models, it also operates GPU infrastructure capable of running open-weight models privately.
Čupr said the distinction is particularly important in Europe, where companies are often more reluctant than their US counterparts to send sensitive business data to external AI providers.
The company has almost 60 employees, but Duvo is entering an increasingly crowded enterprise AI market where many companies are still struggling to move beyond pilot projects.
It already has several well-funded VC-backed competitors, but the closest are Orby AI, Skan AI, Mimica, Sola and Scribe.
Orby AI (US, $35m raised; NEA, Wing VC, WndrCo, Pear VC). It’s a particularly direct competitor as it’s designed to learn business workflows. Its $30 million Series A in June 2024 was co-led by NEA, Wing VC and WndrCo. Skan AI is another, given Duvo's thesis about the undocumented ‘human API’. It raised a $63m Series C in August this year (Cathay Innovation, Dell Technologies Capital, Citi Ventures and others). And that’s just a couple worth mentioning.
For its US push, Duvo has recruited former Databricks and Salesforce executive Michael Green to lead go-to-market operations and former Kmart president and Accenture managing director Spencer Insolia to head its US business. Former Aldi South Holdings CEO Christoph Schwaiger has joined as a senior European adviser.