Our co-founder and CEO Marcin Zarakowski on the capacity ceiling in blockchain forensics, what CoTrace does today, and the North Star behind it.
Recoveris started building an AI investigator because the work stopped fitting inside the number of qualified humans available to do it. Our investigators trace losses on crypto assets, check the provenance of funds, and produce Source of Funds analysis, and the queue for all three kept growing faster than the expert pool that services it. That gap, more than any model architecture, is the reason the company is building what our CEO Marcin Zarakowski calls the digital asset investigator.
Why did Recoveris start building an AI model at all?
Demand rose from two directions at once, and they pull on different skills. Scams and crypto-related crimes have been surging, which drives tracing and recovery work for victims and law enforcement. At the same time, traditional financial players and VASPs keep adopting digital assets, which drives provenance and Source of Funds work for compliance teams that need a defensible answer before they onboard a client or release a transfer.
Marcin is blunt about where that left the firm: it became impossible to service all the demand with existing capabilities. Hiring more investigators does not close a gap of this shape, because the constraint is the supply of people who can actually do cross-chain tracing to an evidentiary standard. So the answer was to automate the investigative workflow itself, and to build the tools that make each investigator’s hour go further.
The second realisation is the one that turned tooling into a model. Every investigation Recoveris runs generates feedback, structured data, and expert judgment, and Marcin’s view is that it would be a waste not to use it. That corpus became the training set for a domain-specific model built for blockchain investigations rather than a general-purpose assistant pointed at on-chain data.
What happens when tokenized assets outgrow human compliance?
The capacity problem gets worse before it gets better. As value moves on-chain and the market adds tokenized financial instruments and tokenized real-world assets, the industry inherits a new set of compliance obligations attached to instruments that settle in seconds and move across chains and bridges without asking permission.
Human-led compliance cannot answer that at speed, and the binding constraint is headcount. There are simply not enough experts to trace these assets and produce proper Source of Funds analysis at the volume tokenization implies. This is our clearest contrarian read on the market: most compliance roadmaps still assume a human reviewer at every meaningful decision point, and that assumption breaks the moment tokenized instruments reach retail scale. The solutions that survive will automate the investigative workflow and the compliance process around it, and they will employ AI to do it.
What is CoTrace, and what can it do today?
CoTrace is the product Marcin says he is most excited about, and it is where the Recoveris model sits. It is a domain-specific AI model for blockchain investigations, trained on the firm’s own casework and then fine-tuned using a combination of large language models, AI agents, and the corrections of investigators who are working experts in blockchain intelligence. That last input is the one competitors cannot buy, because it is the accumulated judgment of people who have taken cases through to freeze and recovery.
In practice CoTrace works alongside the investigator rather than behind them. It supports graph building by pointing to the likely next asset-transfer path to follow, which is exactly the step where an inexperienced analyst loses a trail through a mixer or a bridge. Marcin’s position is that CoTrace already enhances investigator capability on both sides of the practice, blockchain investigations and forensics on one side, Source of Funds analysis on the other.
Does Recoveris really want to replace human investigators?
Yes. The ultimate goal is a model that can fully replace humans doing blockchain investigations, and is the company’s North Star.
The ambition is a coverage position. The requests keep arriving, the caseloads keep growing, and no realistic hiring plan meets them. Our CTO Oleksii Koshlatyi made the same point with a single figure in his own piece on the technology, noting that Binance alone received 71,000 law enforcement requests in a single year. The roadmap between here and there runs through human-led investigation today, a co-pilot working alongside analysts on live cases, and finally a Digital Investigator that runs the case under human oversight.
What makes the Recoveris approach hard to copy?
Marcin’s answer to the differentiation question has three parts that only work together. The first is the investigators themselves, who bring the domain expertise. The second is the case volume, complex matters arriving from very different types of clients, which means the model is exposed to a wider variety of typologies than any single-sector practice would produce. The third is that Recoveris builds products, so the suite that automates the investigative workflow also serves clients across traditional finance, VASPs, and law enforcement.
Everything the firm learns while servicing those clients feeds back into training the domain-specific model. A software competitor can replicate an interface in a quarter. Replicating the corpus means running the same cases, and that takes years. Our ZondaCrypto investigation, where our analysis showed the exchange’s hot wallet had fallen 99.7% from 55.7 BTC in August 2024 to 0.18 BTC in March 2026, is the kind of work that produces the data a vendor cannot license.
Why is law enforcement still behind the criminals?
The largest single obstacle Marcin identifies is an understanding gap. Criminals adopt novel and scaling technologies very quickly, and the public sector and law enforcement agencies lag behind them, which means conversations with those agencies often begin before there is a shared understanding of what blockchain intelligence can do.
So a meaningful share of the work is education. Recoveris takes prosecutors, investigators, and compliance officers through what the tooling actually does before it can be useful to them, and the payoff is straightforward: the same unit closes more cases with the same headcount once it knows how to use blockchain intelligence services and the products companies like ours build. That is unglamorous work and it is a real part of the job.
The takeaway
The bet Marcin is describing is that in blockchain compliance, the binding constraint is expert capacity, so the firms that matter in five years will be the ones that turned real casework into a model rather than the ones that hired fastest.
If your organisation needs investigative capacity it cannot staff internally, or you want to see how our products handle it, contact us at [email protected].
