Product
Platform AI agents Security & governance
Who it's for
Asset management Acquisitions Finance & IR
Company
Resources PricingTrust Center Read the thesis Book a demo →
Analysis

Covenant risk and the cost of finding out late

A covenant breach is rarely a surprise in hindsight. The numbers were drifting for two quarters. The surprise is the timing of the discovery, at the test date, when the cure options have narrowed and the call to the lender is no longer a heads-up but a confession.

Built AI · May 2026 · ~7 min read
Built AI
Analysis · May 2026 · 7 min read
CovenantsRisk
Key takeaways
  • Covenant risk is a timing problem, not a measurement problem. The breach is rarely the surprise; the timing of the discovery is.
  • Breaches are found late because testing is bolted to a quarterly, backward-looking cycle, on inputs that are themselves lagged.
  • Late discovery narrows cure options, strains the lender relationship, and turns a manageable risk into a fire drill.
  • Continuous, forward monitoring on a deterministic engine forecasts breaches before the test and drafts the lender heads-up for approval.

There are two ways to learn that an asset is going to breach a covenant. The first is to model it three months out, while the DSCR is still 1.24x and drifting, with enough runway to accelerate lease-up, top up a reserve, or open a calm, early conversation with the lender. The second is to find out on the test date, when the number prints below the threshold, the cure window is days not months, and the lender's first question is why they are only hearing about it now. The math is identical in both cases. The outcomes are not remotely the same, and the only variable is when you knew.

Covenant risk is therefore less a measurement problem than a timing problem. Most institutional firms can calculate a DSCR or an LTV perfectly well. What they struggle to do is calculate it continuously, forward, across the whole book, early enough that the answer is still actionable. The cost of the gap is not the breach itself. It is the loss of every good option that existed while the breach was still in the future.

Why breaches are discovered late

The structural reason is that covenant testing is usually bolted to the reporting cycle, and the reporting cycle is quarterly and backward-looking. The covenant gets tested when the quarter closes, against figures that describe a period that has already ended. The test confirms what already happened rather than forecasting what is about to. By the time the DSCR is computed for the period, the period is over and the lever that might have moved it, the lease that did not sign, the expense that ran hot, is in the past.

The inputs make it worse. A DSCR test depends on net operating income and debt service, and NOI depends on in-place rent, which depends on a lease portfolio that, as anyone who has abstracted one knows, frequently disagrees with the rent roll and the model. An LTV test depends on a valuation that may be months stale. So even the quarterly number is often computed from inputs that are themselves lagged and partly unreconciled. The firm is testing late, against data that is also late, on a cadence that cannot see forward. Breaches do not sneak up. The measurement system simply is not pointed at the future.

DSCR forecast
The drift was visible two quarters before the test
1.33x 1.24x 1.15x Covenant 1.20x Today 1.24x Forecast 1.18x Q-2 Now Test date
Solid indigo is the measured DSCR; dashed grey is the forecast to the next test. The line crosses the 1.20x covenant before the test date, so the breach is knowable a quarter out, not on the day it prints.
The covenant did not breach without warning. The warning arrived two quarters early. There was just no system watching for it, so nobody read it until the test date made it impossible to ignore.

What being late actually costs

The penalty for late discovery is paid in three currencies, and the breach is only the first.

Narrowed cure options

A breach seen three months out is a menu. You can accelerate leasing, fund a reserve, restructure a small piece of debt, adjust a distribution, or simply pre-agree a waiver from a position of control. A breach discovered at the test date is a much shorter menu, often just a hurried waiver request or an emergency equity cure, negotiated from weakness. The earlier you see it, the more of the cheap options are still open.

Lender relationship

Lenders price trust. A borrower who calls early, with a modeled breach, a proposed cure plan, and an updated schedule, is managing a known risk like a professional, and the conversation stays collaborative. A borrower who surprises the lender at the test date has, from the lender's seat, either missed the drift or hidden it, and neither is reassuring. The same breach handled early versus late produces very different terms on the waiver and a very different relationship for the next financing.

The fire drill

Late discovery also imposes an operational cost that is easy to underweight. A breach found at the test date becomes an all-hands scramble: re-running the numbers under deadline, assembling a cure plan overnight, drafting lender correspondence in a panic, pulling senior people off other work. The same breach found early is a line item on a watchlist, handled in the normal course. The drift was the same. The chaos was a choice the timing made for you.

The asymmetry

Three months of warning costs almost nothing to produce and preserves every cure option. Zero warning costs nothing to produce either, and forecloses most of them. The only difference between the two is whether something was continuously watching the forward number. The cheap thing and the expensive thing are separated entirely by when you looked.

What continuous covenant monitoring does instead

The fix is to detach covenant testing from the quarterly rear-view and run it continuously and forward, across the entire book, on inputs that reconcile themselves as they arrive. That is a capability, not a report, and it has three parts.

It forecasts, it does not just measure

Instead of computing the DSCR for the closed quarter, the system projects it to the next test date, drawing on the live debt schedule and the actual lease-up pipeline rather than a stale assumption. When a forecast crosses the threshold before the test, that is the alert. The number you act on is the one in front of you, not the one behind you.

It runs on a deterministic engine, so the forecast is defensible

A covenant forecast is only useful if you can stand behind it with a lender and an investment committee. So the calculation runs on a deterministic engine, not in a language model's prose: the same inputs always produce the same DSCR, every figure traces to the debt schedule, the rent roll, the loan agreement clause it came from, and the whole thing can be replayed. A forecast you cannot audit is not a forecast you can take to a lender. This one is.

It drafts the heads-up, and waits for a human

When a breach is projected, a covenant agent does the rote work that a late discovery never leaves time for: it models a cure plan against the live lease-up schedule, and it drafts the lender heads-up in your template, with the numbers cited. Then it stops. Nothing is sent. The controller reviews the forecast, confirms or adjusts the cure plan, and approves the communication. The system turns a future breach into a prepared, sourced, human-approved conversation that happens on your timeline instead of the test date's.

Covenant agent run
From a forward forecast to a human-approved lender note
Triggered
Continuous monitoring runs the forward DSCR across the book on every new input.
Pulled sources
Reads the live debt schedule, the rent roll, and the lease-up pipeline as they reconcile.
Computed forecast
Projects DSCR to the next test on a deterministic engine: 1.24x now, 1.18x forecast.
Flagged breach
Forecast crosses the 1.20x covenant before the test date. Alert raised early.
Drafted lender note
Models a cure plan and drafts the heads-up in your template, every figure cited.
Awaiting approval
Nothing is sent. The controller reviews, adjusts the cure, and approves the send.
The agent does the rote work a late discovery never leaves time for, then stops. The judgment and the send stay human.
The goal is not to prevent every breach. Markets breach covenants. The goal is to never be surprised by one, so that every breach is a managed conversation instead of an emergency.

From fire drill to managed conversation

Run covenant monitoring this way and the entire posture changes. The watchlist shows the assets trending toward a test, ranked by headroom and time, long before any of them prints a breach. The drift that used to be invisible until quarter-end is visible the week it starts. And when a breach genuinely is coming, the firm meets it with a modeled cure, a drafted lender note, and weeks of runway, rather than a scramble and an apology.

  • Earlier action. A breach forecast a quarter out leaves the cheap cures, leasing, reserves, small restructurings, still on the table.
  • Stronger lender conversations. Calling early with a plan keeps the relationship collaborative and the waiver terms reasonable.
  • No fire drills. Covenant risk becomes a standing line on a watchlist, handled in the normal course, instead of a monthly possibility of chaos.
  • Defensible numbers. Because the forecast is deterministic and cited, it holds up in front of a lender, an auditor, and an IC.

The expensive thing about a covenant breach is almost never the breach. It is finding out at the test date, when the good options have closed and the lender call has turned adversarial. Built AI runs covenant testing continuously and forward, on a deterministic engine with every figure cited, forecasts breaches before the test, models the cure against the live pipeline, and drafts the lender heads-up for a human to approve. The drift becomes visible the week it starts, not the day it is too late. To watch the covenant agent forecast a breach on your own book, see how it works for asset management or book a walkthrough.