Artificial intelligence may be advancing rapidly, but for many large enterprises, the biggest obstacle to putting AI into production is not the model itself. It is getting AI access to the data that already runs the business.
That challenge is at the center of Unfold, an AI enablement platform co-founded by Idan Shuster, who also serves as its chief product officer. In a HumanX Amsterdam conversation with Omri Hurwitz, Shuster discussed his cybersecurity background, the company’s evolution beyond security use cases and why enterprise systems—from legacy mainframes to modern applications—can become a bottleneck for AI adoption.
From Cybersecurity To AI Infrastructure
Shuster spent roughly a decade in cybersecurity, moving from hands-on penetration testing and offensive security work into product management at Varonis. He also served in Israel’s Unit 8200 before working in cybersecurity services in Israel and the U.K.
That experience shaped how he approaches enterprise software. Working directly with technical teams gave him an understanding of vulnerabilities and systems, while consulting exposed him to the different stakeholders involved in security decisions.
“So I think both of them are kind of shaped the way into being a good product manager when understanding also like the business objectives, but also like the hands-on side of things.”
The company’s original thesis was closer to cybersecurity. Unfold initially focused on helping security teams obtain more of the data they needed for security operations and fraud analysis. That changed after a conversation with a healthcare company’s CISO, who introduced the team to the organization’s CIO.
The CIO had a broader problem: extracting data from proprietary healthcare systems to support AI initiatives.
“We thought, Okay, it’s interesting, it’s a bit of a different focus, but let’s try to hear from the CIO,” Shuster said. “And once we kind of talk with this one, the mind was open to this kind of space.”
The Enterprise Data Problem
Unfold’s premise is that enterprises already possess enormous amounts of valuable data, but the systems containing that information were not necessarily designed to make it easily accessible to AI.
“So, like everyone is saying, we believe that the bottleneck of adopting AI within large enterprises is the quality of the data that you are having,” Shuster said.
He argues that the challenge is often less about the underlying data than the systems surrounding it. Enterprises may rely on mainframes, proprietary healthcare platforms, ERPs and other applications that were built long before today’s AI infrastructure existed.
Unfold connects to those systems and extracts data for AI initiatives, including delivery into data lakes or directly into AI-agent workflows.
The company typically begins with one system. Shuster said its “Unfolding” process generally takes around seven days, with human verification remaining part of the process because of the importance of enterprise data.
Why Context Matters
The word “context” has become ubiquitous in the AI industry, but Shuster believes it can mean very different things depending on how a platform approaches enterprise data.
For Unfold, context includes understanding how an application actually operates—not simply exposing its underlying database.
“If you let the agents interact directly with the data, sometimes it doesn’t make sense,” he said.
A database may not provide a straightforward representation of what users see through an application’s interface. Enterprise-specific workflows and customizations can also determine how information is actually interpreted.
Unfold therefore maps the interfaces, stakeholders, custom components and underlying data structures of an enterprise application before delivering normalized data for AI use cases.
The objective is what Shuster calls “AI ready data”: a complete picture that combines usable data with the context required to interpret it.
From Data Access To Business Outcomes
For Shuster, the business case ultimately goes beyond making information easier to access.
He pointed to healthcare organizations acquiring dozens of clinics each year. Integrating each new clinic’s existing software can take months, creating a constraint on how quickly the organization can expand.
In another example, a large retailer uses mainframes across multiple entities and needs to perform fraud analysis on the data within those systems. Access to that information can become directly connected to the company’s ability to expand into additional markets.
The company’s next phase is focused on scaling these use cases across large enterprises, particularly Fortune 500 organizations with complex combinations of legacy and modern technology.
Shuster believes the problem will eventually extend beyond today’s largest companies.
“I think that every company that has existing for at least five, 10 years and have both modern software and legacy software eventually will have this problem.”
For Unfold, the bet is that enterprise AI adoption will depend not only on increasingly capable models, but on whether organizations can finally connect those models to the systems and information that make their businesses work.
