The CLO Market Is Outgrowing Its Tools

$35 Billion and Counting

Six weeks into 2026, CLO ETFs have pulled in $4 billion in net inflows. Total assets have crossed $35 billion, more than doubling in just over a year. Janus Henderson's JAAA alone sits at $26 billion. On February 12, Fidelity entered the race with two new ETFs (FAAA and FCLO), waiving management fees for 12 months to capture share.

CLOs are no longer an institutional-only asset class. They are going mainstream.

That shift changes the economics, the expectations, and the operational demands placed on every CLO manager, originator, and investor in the market.

The Second-Order Problem Nobody Is Talking About

Most of the commentary around CLO ETFs focuses on spreads, demand dynamics, and whether retail investors understand what they are buying. Those are fair questions. But there is a more immediate one for the people who actually manage these portfolios:

Does your team have the capacity to keep pace with a market that moves this fast? And the confidence that nothing slips through?

The investor base is widening. Reporting requirements are increasing. Regulatory scrutiny is intensifying on multiple fronts simultaneously.

The ECB published Guideline ECB 2026/1 on January 27, redefining residual value risk for ABS collateral eligibility. Effective March 30. No grandfathering clause. For deal teams holding auto ABS with balloon loans or return options, this is not abstract policy. It is a direct question about which positions in your book may lose ECB collateral eligibility, and you need that answer before the effective date, not after.

On February 17, the UK's FCA and PRA published consultation papers (CP26/6 and CP2/26) proposing a wholesale simplification of the securitisation framework. The proposed changes are extensive: new risk retention options, simplified due diligence requirements, and a collapsed distinction between public and private securitisations. If you manage cross-border portfolios, the operational implications of aligning UK and EU compliance are real and they are coming fast.

Meanwhile, UBS strategist Matthew Mish is projecting $75-120 billion in fresh defaults across leveraged loans and private credit by year-end. That estimate, based on default rate increases of up to 2.5% for leveraged loans and up to 4% for private credit, is driven partly by AI-led disruption in software and business services, the very sectors that populate CLO collateral pools. By late February, UBS raised its worst-case private credit default scenario to 15%.

The market is moving on multiple fronts simultaneously. Large US institutions are starting to adopt AI extraction tools for their deal pipelines. But most mid-market European buy-side teams are still running their analysis the same way they did in 2018.

The Real Bottleneck in Structured Credit

Here is what a typical deal analysis workflow looks like today at most firms. These teams already have Intex for cashflow modeling, Bloomberg for pricing, and often Moody's or Trepp for surveillance. None of those tools reads an offering circular. The analyst is still the data ingestion layer.

1. An offering circular arrives. 150-300 pages.
2. An analyst spends 2-4 hours extracting parameters manually: tranche structures, fee mechanics, waterfall priorities, swap terms, pool characteristics. Less if the deal fits an existing internal template. More if the structure is unfamiliar.
3. Those parameters get keyed into a spreadsheet model. Another 2-3 hours if the template fits. Longer if it doesn't.
4. The model gets reviewed, debugged, and iterated. More hours.
5. The analyst runs a base case. Maybe one or two stress scenarios if there is time.

Total elapsed time from document to decision-ready model: one to three days. Sometimes longer.

And every step in that chain is a place where something can go wrong quietly. A wrong date. A miskeyed spread. An assumption someone changed without telling anyone. The deal team knows this. They carry it every day: the awareness that their credibility rests on a stack of spreadsheets, manual inputs, and hand-rolled logic that only they can fully explain. One missed assumption in front of the investment committee, and the trust they spent months building evaporates.

That timeline was acceptable when the market was smaller, the investor base was institutional, and deal flow was manageable. It is not acceptable when:

• Your competitor just launched a CLO ETF and needs to screen every new deal in the market within hours, not days.
• Your portfolio has 80+ positions and the ECB just altered collateral eligibility for a subset of them.
• Your LP base now includes retail investors who expect the same reporting cadence they get from their equity ETFs.
• A UBS research note just flagged that 25-35% of private credit portfolios face elevated AI disruption risk, and your CIO wants exposure mapped before the next committee meeting.

The bottleneck was never the math. It was the time cost of implementing the math correctly for each deal, every time.

What Changes When the Bottleneck Breaks

Consider what an AI agent does when you hand it a 179-page offering circular and nothing else. No templates, no parameter sheets, no manual extraction. The agent indexes the document, identifies every structural parameter, extracts and cross-verifies the data, then runs deterministic cashflow tools to build a complete 60-period waterfall model with a full payment schedule.

Take Mercedes-Benz Silver Arrow Compartment 20, a EUR 744M German auto ABS. An agent produced the model in minutes, not days. And it caught a nuance that would have broken a manual model: the distinction between Issue Date and Closing Date for first-period interest accrual. A two-month difference that would have thrown off every downstream calculation. That is exactly the kind of quiet error that surfaces in an investment committee and costs someone their credibility.

The technical details of how that pipeline works are in the companion piece. This one is about what it means for your business.

When modeling time drops from days to minutes, the question is not "can we build models faster?" The question is what becomes possible that was not possible before:

You screen the full market, not a filtered subset. Most teams look at 20-30% of available deals because they only have bandwidth to model that many. When the modeling bottleneck breaks, you can screen everything and make exclusion decisions based on analysis, not capacity constraints.

You stress test before you commit, not after. Running 10 scenarios instead of 2 is not a luxury. It is the difference between understanding a deal's behavior under stress and hoping it behaves. When scenarios are cheap to run, your default should be to run more of them.

You stop saying no to bespoke structures because of the extraction cost. Private credit ABF is the fastest-growing segment of structured credit, and every deal arrives with a non-standard structure. No existing cashflow library has pre-built models for these. Every offering circular has to be read from scratch. That is exactly the category where manual extraction costs the most time and where the competitive advantage of faster analysis is largest.

You respond to market events in hours, and your team does the work that matters. When the ECB redefines collateral eligibility, you should not need a week to figure out which deals in your book are affected. When UBS publishes a default forecast, you should be able to map that to your portfolio before the next morning's meeting. Every hour an analyst spends extracting parameters from a PDF is an hour they are not spending on assumption setting, relative value, or structural risk assessment. The goal is not to replace the team. It is to stop burying them in extraction and reconciliation so they can do the work they were hired to do, with confidence that the inputs underneath them are solid.

The Trust Question

Every executive in structured finance has the same reasonable objection: how do I defend these numbers to my investment committee?

Nobody wants opaque outputs from a black box. That is not what this is.

The systems that work in regulated finance separate reading from calculating. The language model reads the document, identifies relevant information, and reasons about deal structures. Deterministic tools handle the math: precise, auditable, reproducible computation. No language model is doing arithmetic.

Every number traces back to a deterministic function, not a token prediction. Every extracted parameter carries a page citation from the source document. Every assumption that could not be verified is explicitly flagged as an estimate, not silently filled in.

The result is a model you can defend. Not because you trust the system, but because every output is transparent enough to verify. That is the standard your team already holds themselves to. The difference is that the verification takes minutes instead of days.

The Window

The CLO market is entering a new phase. Broader investor access means more deals to screen, more positions to monitor, more scenarios to model, and more reporting to produce. The firms that absorb that growth will be the ones whose teams have the capacity to analyze more deals without losing confidence in their outputs. The firms still running manual workflows will hire more analysts, burn more hours, and still fall behind.

UBS is already warning that disruption risk is not priced into credit markets. The irony is that the same technology creating disruption risk in CLO collateral pools can also help you measure, monitor, and manage that risk faster than any manual workflow, with outputs you can defend.

The question is not whether your team needs more capacity. It is whether you build that capacity now, while you have the luxury of choosing how, or later, when the market is choosing for you.

For the technical deep-dive on the agent pipeline, read I Watched an AI Agent Model a EUR 744M Auto ABS Deal from Scratch, or if you want to see this on one of your deals, let's talk.