Daniel Liechtenstein, CEO and Co-Founder of Hypercore – Interview Series

Daniel Liechtenstein, CEO and Co-Founder of Hypercore, is a fintech entrepreneur and financial strategist with a background spanning lending technology, software development, business consulting, and large-scale financial planning. He co-founded Hypercore in 2020 after previously serving as a Business Development Partner at Articode, where he focused on fintech software projects and financial services clients, and working independently as a financial and business consultant to startups and growing companies. Earlier in his career, Liechtenstein held senior financial planning and strategy roles within the Israel Defense Forces and Israeli Ministry of Defense, where his responsibilities included managing major ICT budgets, evaluating technology investments, negotiating large-scale agreements with global technology vendors, and contributing to cloud infrastructure and strategic procurement initiatives.

Hypercore is a technology platform built for private credit funds and non-bank lenders, designed to modernize the operational infrastructure behind lending. Its end-to-end platform centralizes loan data and supports the full lifecycle from origination and pipeline management through servicing, funding-source management, reporting, and maturity, while automating many of the calculations and workflows traditionally handled through spreadsheets and disconnected systems. More recently, Hypercore has expanded its focus on AI-native loan administration, using AI agents alongside structured loan data and professional oversight to automate recurring servicing activities such as facility onboarding, calculations, payments, reconciliation, notices, and reporting. The platform is designed to give lenders real-time portfolio visibility, auditable workflows, and the infrastructure needed to scale operations without proportionally expanding administrative teams.

When you started Hypercore, what did you think was missing from the way private credit firms managed loans after closing? And how has that idea changed as AI has developed?

We started Hypercore because there wasn’t really a loan management system built for private credit.

A lot of systems handle standard lending products reasonably well, but private credit is not standard. It involves different structures, different payment mechanics, covenants, amendments, waterfalls, lender groups, and more. No two deals are the same, and firms rely far too heavily on spreadsheets and manual processes.

Our idea was to build infrastructure that can represent the loan properly and manage it throughout its entire lifecycle.

This idea evolved when we realized that, once we had that infrastructure underneath, AI could do something useful with it. If the loan is represented correctly in the system and you know its current state, AI can understand an event, apply the loan terms, start the workflow, update systems, prepare the output, and bring a person in when human judgment or approval is needed.

This realization is what led us to start providing loan administration services alongside the software. We’re using the infrastructure we’ve already built, together with AI agents and our operations team, to run the work.

The infrastructure has to come first but, once you have that, you can use AI to change how post-close operations are completed.

Most of the conversation around AI in private credit has been about underwriting and due diligence. Why do you think the bigger opportunity could actually come after the loan closes?

Underwriting is the obvious place to start because there’s a lot to read and analyze, and AI is very good at that.  However, that process only takes a few weeks. After close, you are responsible for the loan for years, having to deal with payments, rate changes, amendments, covenant tests, draws, notices, reconciliations, reporting, and more. Plus, every time the portfolio grows, all of that work grows with it.

If AI saves an investment professional an hour during underwriting, that’s useful. But if it reduces the amount of manual work required to run a portfolio every day, that’s huge.

What makes servicing a loan over five or seven years harder to automate than analyzing it when the deal is first being underwritten?

If you ask an AI system to summarize a credit agreement, it’s a pretty basic task. Somebody checks the AI’s answer and moves on.

But servicing doesn’t work like that. This month’s calculation may be impacted by an amendment from last year, a rate reset from three months ago, and a payment from last week. The state of the loan, and in fact the loan itself, keeps changing all the time. The original agreement is just the starting point. You also have to keep on top of amendments, waivers, elections, and anything else that was agreed upon outside of the original document.

This is not easy!  It’s not just a matter of reading documents; you have to maintain a live financial instrument over several years and ensure that every new event is applied correctly. If you make a mistake early on, it can flow through future calculations for months or even years before anyone notices.

A lot of managers still keep their own “shadow book” to check the work of their administrator. Why has that become so common? And what would have to change for firms to feel comfortable giving it up?

If an administrator sends you a calculation without any info on how they came up with it, you’d probably calculate it yourself too! That’s basically what the shadow book is. It creates duplicate work, but it exists for a reason; the manager needs to know that the numbers are right.

To eliminate shadow books, managers will need much more visibility into the work, including what inputs were used, what terms were applied, and what, if anything, has changed.

If the manager and administrator are working off the same underlying loan record, there’s no need for shadow books. However, I don’t think anyone is going to give them up just because a new provider tells them they can. That trust will need to be earned with accurate results over a period of time.

There’s a big difference between AI that can read a credit agreement and AI that can actually carry out servicing work. What does a system need before you would trust it to execute those workflows?

It requires multiple things, including:

• A reliable loan record
• Knowledge of previous activities
• Access to the systems where the work is completed
• Awareness of when a task is finished and what to do if something doesn’t look right

Take a payment, for example. Obviously, the AI has to calculate the payment, but it doesn’t end there.

The payment also has to be processed, notices may need to be distributed, systems need to be updated, a human may need to review it, and the cash needs to be reconciled afterwards.

And the tolerance for mistakes is very low. That’s why our team is accountable for the administration service that we offer. The AI is doing some of the work underneath, but we’re not pretending the software itself is somehow responsible for the outcome.

Once AI starts touching things like payments, allocations and investor reporting, how do you decide what it can handle on its own and where a person still needs to step in?

Honestly, it depends on what happens if the system is wrong. There are plenty of situations where AI can do the work automatically, like pulling information, reconciling records, identifying anomalies – that sort of thing. But if you’re moving cash or sending something externally, you need to have a person step in.

Having said that, I also think there is too much focus on whether there is a human approval step and not enough on what that person can actually see. If I give somebody a number and an approve button, that’s not a lot of help. They need the ability to understand where the number came from without wasting an hour redoing the calculation.

In short, if a rate changes, the system needs the ability to provide the old rate, the new rate, the source document and the inputs. Then a human can actually review it.

As the system proves itself on a workflow, you can automate more of it, but don’t do so just because a new model has been released.

Private credit loans can be incredibly bespoke. How do you automate around that without asking an AI model to effectively reinterpret the deal every time something happens?

AI is useful when you’re reading the documents and figuring out how the deal works. A 200-page agreement plus amendments is exactly the kind of thing AI can help with. Then, once you understand the structure, AI can help map the deal into the loan model so it is represented properly.  After that, the calculations are deterministic.

If the loan has a waterfall, a PIK component, multiple tranches or a pricing step-down, those rules are in the system. We’re not asking a language model to reread the documents every time a payment comes in! Managers must be able to reproduce and explain every calculation; that’s impossible to do if the model reinterprets the deal every time something happens.

Investment risk naturally gets most of the attention in private credit. What kinds of operational risk do you think managers tend to underestimate?

Managers underestimate the little things: an amendment that gets reflected in one place but not another; a reconciliation that happens later than expected; an employee unknowingly working from an old version of the terms.

It’s generally not a single colossal failure, but a small issue that stays wrong in the background for a while before it starts affecting other things.

For instance, if you calculate something off the wrong term, that feeds into a payment, and then shows up in reporting. By the time someone notices, you’ll have to review months of work.

AI can help by shrinking that window. If you can reconcile every day instead of at quarter-end, you can find problems a lot more quickly. If every amendment gets picked up and reflected in the system immediately, you reduce the chance of people working from different versions of the loan.

It’s not as exciting as the idea of AI making investment decisions, but operationally it makes a big difference.

For a firm that has years of loans sitting across legacy systems, spreadsheets, PDFs and email, what is actually hard about moving to a new operating environment?

Figuring out what the correct data is.

A firm might have terms in a credit agreement, balances in a different system, amendments in a folder, payment history in an email, and even some important context that exists only in the memory of a single team member!

So first you have to reconstruct the loan as it stands today. Then you have to compare that with what the existing system says. Sometimes the numbers don’t match and you have to figure out why. Maybe something was mapped incorrectly during the migration?  Maybe there was a manual adjustment years ago? Or maybe the old system was just wrong. Whatever the case, you can’t just ignore the difference.

We recommend running the old and new environments in parallel for a while, comparing the outputs, investigating the differences,  and getting everything in shape before completing migration.

It’s not the fastest way to migrate, but on a live loan book, speed shouldn’t be the main objective.

If AI allows firms to run much larger portfolios with smaller operations teams, how do you think that changes the role of the traditional loan administrator?

Administration used to mean manual work, but if AI can do most of it, the loan administrator’s new role is to make sure the numbers are right and quickly correct any issues that arise.

Managers may actually take on more of the work themselves, depending on how many people they have and whether they can deliver the work accurately and transparently. Many firms will still prefer to outsource though. Smaller operations teams only work if the knowledge of previous team members is actually captured somewhere. If you just take people out of the process without doing that, you haven’t necessarily improved anything.

Thank you for the great interview, readers who wish to learn more should visit Hypercore.