A 90-second hold before a withdrawal request is confirmed reduced session-18 reversal rates by 22% in a controlled test of 4,180 U.S.-facing online casino accounts, according to operator-side data collected between March and August 2024. The effect appeared only above the 90-second threshold; holds of 30 and 60 seconds produced reversal reductions of 4% and 9% respectively, both inside the confidence interval of the no-hold control group. The finding matters less for the headline number than for what it implies about the mechanism: reversals are not primarily a liquidity problem, and the interventions that treat them as one are mispriced.
What the Study Measured and Why Session 18
The dataset came from a mid-size operator running a slots-and-table mix, roughly 61% slots handle by volume, with a median deposit of $40 and a median session length of 27 minutes. "Session 18" is not a marketing term. It refers to the 18th distinct login session after a player's first deposit, a point at which the cohort has separated cleanly into two populations: players who have settled into a routine stake size, and players who are escalating. In this operator's data, 71.3% of first-time reversal events — a withdrawal request cancelled and returned to the balance — occurred at or after session 18. That concentration is what made it the natural test window.
The test was a straightforward A/B allocation. Newly eligible accounts were randomized at the point of withdrawal request into four arms: no hold (control), 30-second hold, 60-second hold, and 90-second hold. "Hold" here means a mandatory confirmation delay between the request and the balance leaving the withdrawable state. It is not a pending period in the regulatory sense; the funds remain withdrawable throughout, and the player can cancel at any point. The primary outcome was reversal within 24 hours. Secondary outcomes were reversal within 72 hours, net deposit within 7 days, and 30-day retention.
The 22% figure is the relative reduction in 24-hour reversals in the 90-second arm versus control, from a base rate of 11.4% to 8.9%. Absolute reduction: 2.5 percentage points. With 4,180 accounts split across four arms, the 90-second result cleared the pre-registered significance threshold; the 30- and 60-second arms did not.
Why the Threshold Is Not Linear
The non-linearity is the interesting part and the part most operators will get wrong if they read only the headline. A 60-second hold cut reversals by 9%; a 90-second hold cut them by 22%. If the mechanism were simple friction — make the cancel button harder to reach in time — you would expect roughly linear returns, or diminishing ones. Instead the curve is flat, then steep.
The most plausible reading is that reversals cluster into two behavioral types. The first is reflexive: a player requests a withdrawal, sees the balance drop from the display, and cancels within seconds because the number on screen changed. A 30-second hold catches most of these. The second type is deliberative and takes longer to surface: the player requests the withdrawal, sits with the decision, and cancels somewhere between 60 and 120 seconds later, often after re-entering the lobby. The 90-second hold is the first arm long enough to interrupt that second window before the cancel action completes.
This is consistent with the 72-hour secondary outcome. In the 90-second arm, 72-hour reversals fell 17% versus control, a smaller relative drop than the 24-hour figure. Some of the 24-hour reversals were deferred rather than prevented, which is exactly what you would expect if the hold is buying deliberation time rather than eliminating the underlying urge.
The Revenue Question Operators Actually Ask
Any intervention that reduces reversals reduces something else, and the honest version of this finding has to name it. Reversals are, in accounting terms, retained handle. A player who cancels a withdrawal and keeps playing generates additional wagering, and a portion of that wagering is house revenue. Cutting reversals by 22% in the 90-second arm is, on its face, a revenue-negative change.
The data does not support that conclusion cleanly. Net deposit within 7 days in the 90-second arm was 6.1% higher than control, not lower. Thirty-day retention was 3.8 percentage points higher. The plausible explanation is that forced reversal — a player who cancels a withdrawal and then loses the balance — is a churn event. The player who wanted $400 out, cancelled, and lost it does not reliably deposit again. The player who successfully withdraws $400 and returns a week later does.
If that holds, the operator math is not "reversals foregone" versus "reversals retained." It is the lifetime value of a player who trusts the withdrawal process against the short-term handle from a player who does not. This operator's 90-day cohort data, which the study authors flag as preliminary, suggests the trust effect dominates after roughly 45 days. Before that, the 90-second arm looks revenue-negative.
Where the 90-Second Number Could Fail to Replicate
Three conditions in this dataset deserve scrutiny before anyone treats 90 seconds as a standard.
First, the payment rail. This operator processed withdrawals through a method with a median settlement of 11 hours. In jurisdictions or operators where settlement is under two hours, the hold may be redundant — the player's money is already effectively gone, and the cancel window that the hold is designed to interrupt does not exist in the same form.
Second, the stake distribution. The median deposit was $40. The 22% effect was concentrated in accounts with deposits between $100 and $600; below $100 the effect was 7%, and above $600 the sample was too thin to report. A hold that adds 90 seconds to a $20 withdrawal is a different product experience than one applied to a $500 withdrawal, and players respond accordingly.
Third, the framing. The hold in this test was presented as a confirmation step with a countdown, not as a cooldown or a "cooling-off" period. Language matters here in a way that is easy to underestimate. A countdown reads as a system process. A cooling-off label reads as a judgment about the player, and players who feel judged cancel the withdrawal and often the relationship with the operator.
Regulatory Overlap and the Responsible Gambling Frame
Several U.S. state regimes already mandate some form of withdrawal delay, though the rationales differ. Some are anti-fraud holds tied to deposit-source verification; others are problem-gambling provisions. The 90-second hold tested here is short enough to sit underneath most of those requirements rather than conflict with them, which is part of why it is operationally attractive. It does not require a regulatory change to deploy.
That said, the responsible-gambling case for a hold is not the same as the retention case, and conflating them produces bad policy. A 90-second confirmation step is a mild friction with a measurable behavioral effect on a specific population. It is not a treatment intervention, and it should not be marketed internally as one. Operators who frame it as player protection while measuring it purely on reversal reduction will eventually discover they have built a feature with no owner.
The more defensible framing is narrower: a short confirmation delay changes the default from "cancel is easy and immediate" to "cancel is easy and takes 90 seconds." Defaults move behavior. That is the entire finding, and it is enough.
What the Next Test Should Isolate
The open question is not whether 90 seconds works better than 60. It is whether the effect survives when the player knows the hold is coming. Every account in this study encountered the hold for the first time at the withdrawal screen. If the second withdrawal request is faster because the player has learned to expect the delay and plans around it, the 22% is a novelty effect with a decay curve, and the operator has 90 days of goodwill before it reverts.
A follow-up that randomizes hold length within accounts across successive withdrawal requests would answer this. Until someone runs it, the 90-second number should be treated as a first-encounter estimate, not a steady-state one.