Unconditionally-cancellable lines barely move your CECL reserve—so the discipline on $60 billion of fresh headroom has to live in signals and cadence, tier by tier, not on the balance sheet.

The scissors

Something quietly diverged in the first quarter. Aggregate card limits rose about $60 billion, while card balances actually fell roughly $25 billion to $1.25 trillion, per the NY Fed’s Q1 2026 Household Debt and Credit report (May 12). Balances are still up 5.9% year over year, and total household debt nudged up 0.1% to $18.8 trillion—but the gap between what consumers can draw and what they’re choosing to draw widened.

That gap is open-to-buy. And it’s not spread evenly. TransUnion’s Q1 CIIR (April 30) shows a K-shaped market underneath the aggregate: new-account limits for super-prime borrowers hit $12,511, up 11.5% year over year, while deep-subprime opened at $678 (high-subprime at $1,034). Lenders, in TransUnion’s words, “actively managed exposure by limiting credit line growth, particularly below prime.” Super-prime is now more than 40% of the credit-active population—roughly 15 million more consumers than in 2019.

So the headline “$60B in new limits” is really two stories. One book is getting handed real room to spend. The other is getting metered. The discipline question is different for each.

The CECL paradox

Here’s the part that makes line management easy to under-manage: most card lines are unconditionally cancellable, which means the undrawn portion carries limited CECL reserve. You can extend a customer another $5,000 of open-to-buy and book almost nothing against it, because—on paper—you retain the right to cut it on demand.

That’s a real accounting fact, and it’s worth saying plainly. But “limited reserve” is not “no consequence.” The right to cut on demand runs into UDAAP scrutiny, complaint volume, and the score impact a borrower feels when their available credit drops. Reserve relief is genuine; the operational and reputational cost of exercising that right is not free.

Which leads to the uncomfortable conclusion: if the balance sheet doesn’t force discipline on undrawn lines, then the discipline has to come from somewhere else. It has to live in the signals you watch and the cadence at which you act on them. That’s a process control, not a capital control.

The tier-aware grid

The aggregate book is benign right now. NY Fed early-delinquency transitions on cards actually ticked down, from 8.7% to 8.6% annualized. The FDIC’s Q1 QBP (May 27) put the industry net charge-off rate at 0.59%—down 4bps for the quarter and 8bps year over year, with asset quality stable to improving. Worth noting: that 0.59% is the all-bank figure, not a card-specific number; card stress is showing up in pockets at specific institutions rather than across the industry. Bankcard 90+ DPD did rise to 2.53% (up 10bps YoY), so the picture is “stable with a warm spot,” not “all clear.”

In a K-shaped book, a single portfolio rule averages those realities into something that’s wrong for everyone. A tier-aware grid fits the action to the segment:

TierSignals to watchReview cadenceDefault actionAdverse-action notice?Super-primeUtilization velocity, spend trend, trended bureau improvementEvent-triggered; lean-in CLI on positive signalProactive CLI (within §1026.51 ability-to-pay policy)No—an increase isn’t adverse actionNear-primeUtilization velocity, paycheck-deposit stability, trended bureauQuarterly + event triggersHold; CLI only through a cash-flow gateNo on hold; yes if you decreaseSub-prime / thin-fileCash-flow/deposit data, minimum-payment coverage, early-delinquency markersMonthly or event-triggeredHold and right-size to spending capacity; limit growthYes if you decrease on unfavorable reviewDeteriorating (any tier)Deposit drop, utilization spike, missed-payment markers, new derogatoriesReal-time / event-triggeredEvent-triggered CLD with notice preparedYes—decrease on unfavorable review = adverse action

Two guardrails sit on top of this grid.

First, on the super-prime “lean-in”: a proactive batch CLI program isn’t a free lever. Reg Z §1026.51(a) requires you to consider the consumer’s ability to make the required minimum payments—from income, assets, and current obligations—before you increase a limit, not only when a consumer asks. (This trips people up: the ability-to-pay rule applies to issuer-initiated increases too.) Proactive CLI programs need reasonable written policies and a sound method for estimating income or assets. Leaning in is fine; leaning in without that documented method is not.

Second, the “deteriorating” row is where notice burden bites, and we’ll come back to it.

Two ways to be wrong

Every line decision can fail in two directions, and the two failures don’t cost the same way—or show up the same way.

Over-extend. You give a borrower more room than their capacity supports, they draw it, and it goes bad. This loss is visible. It lands in charge-offs, it’s attributable, someone owns it. The whole apparatus of underwriting exists to keep this number down, and it’s the failure mode everyone instinctively manages against.

Over-limit. You hold a good customer too tight—undersize the line on someone who’d have spent responsibly and paid—and they put their spend on a competitor’s card, or churn. This loss is invisible. It never hits a loss column. It shows up as foregone interchange, foregone revolving balances, and attrition you can’t cleanly attribute. Under-limiting good customers is a silent retention drain, and because it’s silent, it’s chronically under-weighted.

The asymmetry matters in a K-shaped book specifically. Right now the temptation is to manage the whole portfolio toward the over-extend fear—tighten everywhere. But the super-prime tier, with limits up 11.5% and delinquency transitions falling, is precisely where over-limiting costs you the most and risks you the least. Treating line management as a single score cutoff buries this. Treating it as decision optimization—maximize profit per account subject to a loss constraint, an efficient-frontier framing rather than one threshold—surfaces it. (To be clear, “the efficient frontier” here is an internal modeling convention, not a regulatory standard.)

The line you say to leadership: we are not just avoiding over-extension; we are also paying, invisibly, for over-limiting our best customers—and our reserves don’t see either cost.

From annual batch to event-triggered—and the ceiling on it

If discipline lives in cadence, the upgrade is moving from an annual batch review toward event-triggered signals: trended bureau data, paycheck-deposit or cash-flow drops, utilization velocity, behavioral triggers. Cash-flow data is especially useful for right-sizing limits to actual spending capacity on thin-file and non-prime accounts, where a static score tells you little. Average revolving utilization is running roughly 29% (a Sept 2025, Experian-cited figure—directional, not a clean Q1 2026 print), which is low enough that velocity changes carry more information than the level.

But “real-time” has a regulatory ceiling, and it’s worth being exact about where.

A credit-limit decrease triggered by an unfavorable review is adverse action. It requires FCRA/ECOA notice, including the credit-score disclosure added under Dodd-Frank. You can detect deterioration in near-real-time; you cannot cut in near-real-time without the notice machinery firing. That’s the rate limiter on the deteriorating tier—not your data latency, but your notice obligation.

Two precision points so you don’t over- or under-paper this:

• Denying a consumer’s request for more credit on an existing account is generally not adverse action. Changing the existing terms—cutting the line on review—is. Same account, different trigger, different obligation.
• Combined ECOA/FCRA notices using the Reg B model forms cut the operational burden of doing this at volume. If you’re going to run an event-triggered CLD program, build the combined-notice workflow before you need it, not during the first stress event.

Takeaway

The $60 billion of new open-to-buy won’t show up in your reserve in any way proportional to the risk it carries. That’s exactly why it needs a different kind of discipline. Do one thing this quarter: pull your line-management rules apart by tier and check whether you’re running one cadence and one set of triggers across a book that has clearly split in two. Lean in where the signal earns it (with your §1026.51 income-estimation method documented), hold and cash-flow-gate where it doesn’t, and build the combined adverse-action notice workflow now—so that when an event trigger says “cut,” the notice is the thing that’s ready, not the thing that slows you down.