The Underwriting Moat: Why Data Infrastructure, Not Model Choice, Decides the Next Decade of Private Credit

By Steve Iskander, Founder and CEO, Intrepid

In this article

• Why the model cannot be the moat
• The four properties of a defensible data layer
• Compounding is the whole point
• What this means in practice

I have spent six weeks making one argument about lending data infrastructure from six directions. Here it is in one place.

The durable advantage in private credit is not the model you underwrite with. It is the data infrastructure underneath it.

Start with why the model cannot be the moat. Capability is converging and the cost curve is falling. Whatever you can access this year, the fund down the street can access next year for less. An advantage any competitor can purchase on a twelve-month lag is a feature, not a moat.

Now the six pieces.

The pipeline, not the model. The time and error in a credit decision cluster before the analysis, in assembly and verification, not in the reasoning itself.

The context problem. The material that most often changes a credit decision lives in prose and documents. It gets compressed into a memo and the nuance does not survive the hand-offs.

Speed is a data problem. Reported gains from automated underwriting measure how quickly information becomes usable, not how clever the model is. Hiring more underwriters scales assembly work you should not be paying for.

Underwriting does not end at close. The same live data that shortens origination is what surfaces deterioration early, and regulators are now explicitly interested in the quality of that visibility.

Throughput, not capital. With the asset class projected toward roughly $4.5 trillion by 2030, capacity to underwrite well at pace becomes the binding constraint, and discipline gets expensive exactly when deployment pressure peaks.

Each is a different symptom of the same condition.

So what makes a data layer defensible? Four things, and none of them arrive as a purchase.

It is yours. Built around how you originate, what you lend against, and the borrowers you actually see.

It is live. Continuous rather than periodic, which is what makes monitoring economically viable instead of aspirational.

It is configurable to your credit policy. Your credit box encoded as policy your officers can write and change, not logic a vendor owns.

It is explainable. Whatever the model does, you remain accountable for the reasons behind an adverse decision. Governance and clean structured outcome data are, by most accounts, the real constraint on advanced AI in this market, not model access.

And it compounds. Every file you process well makes the next one faster and the portfolio more legible. A competitor starting today starts at zero on that curve, and they cannot buy their way to your position, because the asset is the accumulated structure of your own lending.

That is the moat. Not a model. The infrastructure that decides what your model, and your people, ever get to see.

For the lenders reading this: if a competitor bought identical technology tomorrow, what would still be hard for them to copy about how you underwrite?

Why the model cannot be the moat

Model capability is converging and its cost is falling. Whatever edge a lender gets from model selection this year, a competitor can typically access next year for less. An advantage anyone can purchase on a short lag is a feature, not a moat. That is not an argument against using capable models; it is an argument about where durable advantage actually lives. It lives one layer down, in the data infrastructure the model depends on.

The four properties of defensible lending data infrastructure

A defensible underwriting data layer has four properties, and none of them arrives as a purchase. It is owned, built around how you originate and what you lend against. It is live, continuous rather than periodic, which is what makes monitoring viable instead of aspirational. It is configurable to your credit policy, written and changed by your own officers rather than a vendor. And it is explainable, so every adverse decision carries a clear reason. Together these compound: every file processed well makes the next decision faster and the portfolio more legible.

Compounding is the whole point

Most technology advantages depreciate as the tools commoditize. Structured lending data does the opposite. Its value grows with volume and history, because the accumulated, well-structured record of how you lend is not something a competitor can buy. A firm starting today starts at zero on that curve. That is why the moat is the pipeline, not the model: the model is a subscription, and the data infrastructure is an asset that appreciates.

What lending data infrastructure means in practice

For a lender, the practical test is a single question: if a competitor bought identical technology tomorrow, what would still be hard for them to copy about how you underwrite? The honest answer is rarely the model. It is the accumulated, structured record of how you originate, the credit policy encoded in your own officers’ language, the live monitoring built on your book, and the explainability that satisfies a regulator. Those assets are earned over time and grow with volume, which is what makes them defensible. The strategic implication is to stop treating data infrastructure as plumbing to be minimized and start treating it as the asset that compounds. Buy capable models, by all means, but do not mistake them for the moat. The firms that win the next decade of private credit will be the ones that built the layer underneath, because that is the part no one can purchase, and the part that makes every future decision faster, cheaper, and more legible than the last.

Related reading

→ Why AI is not the bottleneck in lending

→ Private credit’s path to $4.5T

Frequently asked questions

What is an underwriting moat?

An underwriting moat is a durable, hard-to-copy advantage in how a lender makes credit decisions. The argument here is that the moat is not the model, which any competitor can buy, but the data infrastructure underneath it: owned, live, configurable to a credit policy, explainable, and compounding with every file processed.

Why is model choice not a durable advantage in lending?

Model capability is converging and its cost is falling, so whatever model advantage a lender has this year, a competitor can typically access on a short lag. An advantage anyone can purchase is a feature, not a moat. The defensible layer is the proprietary, accumulated data infrastructure a model depends on.

What makes a lending data layer defensible?

Four properties: it is owned and built around how you originate, it is live rather than periodic, it is configurable to your credit policy by your own officers, and it is explainable to regulators. Together these compound over time, because every file processed well makes the next decision faster and the portfolio more legible.

What does explainability require of lenders using AI?

Regardless of model complexity, lenders remain accountable for providing clear, specific reasons for adverse credit decisions. In practice the data and logic behind a decision must be traceable and explainable. Governance and clean, structured outcome data are widely cited as the real constraints on advanced AI in lending, more than access to models.

Can a lender build this advantage quickly?

Not instantly, and that is the point. A defensible data layer compounds with volume and history, so a firm starting today starts at zero on that curve. The advantage comes from consistently structuring how you lend over time, which is why it is durable rather than something a competitor can buy in a quarter.

Sources: Preqin

Intrepid is the lending data infrastructure that compounds with every file. See how at intrepidfinance.io.

Published by Intrepid. Democratizing Access to Capital. intrepidfinance.io