Operators who pushed jackpot-win notifications into a three-per-hour feed saw average stakes in the final quarter-hour of a session fall roughly 9% against a matched control cohort, according to an internal A/B test run across four U.S.-facing casino brands over 11 weeks ending March 2025. The effect was measured on settled rounds only, excluded bonus-fund wagers, and held after controlling for game mix, deposit tier, and time of day. The mechanism is not a mystery: a visible win feed converts a private, solitary session into a comparative one, and players respond to that comparison by trimming exposure.
The Study Design and Why the 9% Figure Survives Scrutiny
The test split roughly 214,000 active accounts into treatment and control at the account level rather than the session level, which matters. Session-level randomization leaks, because a player who sees the feed on Tuesday behaves differently on Thursday even if Thursday's session is nominally "control." Account-level assignment avoids that contamination.
Three design choices give the 9% number its weight:
- Matched cohorts. Treatment and control were balanced on 90-day historical stake, game-category concentration, and deposit frequency before the test opened. Post-hoc reweighting moved the estimate by less than 0.4 percentage points.
- A defined window. The 9% is specifically a session-75 metric — average stake per round in the final 25% of a session, measured from first spin to cash-out or idle timeout. Early-session stakes moved only 1.8%, which is inside noise. The compression is a late-session phenomenon.
- Exclusion of promotional rounds. Free spins and bonus-fund wagers were stripped out. Bonus play has its own staking logic and would have muddied the read.
The feed itself was capped at three posts per hour per player, drawn from the operator's live jackpot events. The cap is the interesting variable. Uncapped feeds in earlier, unpublished tests produced a smaller and noisier effect — plausibly because a firehose of wins reads as background noise, while three an hour reads as signal.
What "Session-75 Stakes" Actually Measures
Session-75 stakes are not total session spend. They are the mean wager per settled round across the last quarter of a session. A player who opens at $2.00 a spin, drifts to $2.50 mid-session, then closes at $1.80 has a session-75 figure near $1.80. The metric isolates the tail of a session, where most of the money actually moves and where behavioral interventions tend to bite. A 9% cut there is worth more in dollar terms than the same cut applied to opening stakes, because late-session rounds carry higher average handle.
Why a Win Feed Changes Staking Behavior
The dominant explanation in the operator's own write-up is social comparison, and the data are consistent with it. Treatment players did not reduce session length — median session duration rose 4%, if anything. They reduced stake per round while staying in the chair. That pattern fits a player who is now measuring their own result against a visible stream of other people's wins and adjusting downward to manage the gap.
Two competing explanations deserve attention.
The novelty-and-fatigue account. A win feed is new, and new UI elements depress engagement briefly before habituation. If the effect were pure novelty, it should decay across the 11 weeks. It did not: the week-10 estimate was within 0.7 points of week-2. That argues against a simple habituation story, though 11 weeks is short enough that a longer decay can't be ruled out.
The loss-salience account. Jackpot posts are, by construction, wins. A feed of wins makes losses feel more anomalous, and players may stake more conservatively to avoid the felt contrast. This predicts the effect should be strongest among players who have just lost, which the segment data weakly support — the 9% figure was 11.4% among players in a drawdown at the time of feed exposure and 6.9% among players in profit.
| Segment | Session-75 stake change |
|---|---|
| All treatment accounts | −9.0% |
| In drawdown at exposure | −11.4% |
| In profit at exposure | −6.9% |
| High-variance slot players | −12.1% |
| Low-variance table players | −4.3% |
The high-variance split is the one to watch. Slot players chasing volatile titles are the most responsive to a jackpot feed, which is unsurprising — the feed is showing them exactly the outcomes their game type produces, at a frequency their own play rarely matches.
The Operator Incentive Problem
Here is where the finding gets uncomfortable for the people who commissioned it. A 9% reduction in late-session stakes is a revenue headwind, not a tailwind. Operators do not typically ship features that cut handle, and the fact that this test was run at all suggests someone believed the feed would do the opposite — that visible jackpots would drive engagement and larger bets.
The result inverts that assumption. The feed does appear to hold players in session slightly longer, but at materially lower stakes, and the stake effect dominates. Net gaming revenue in the treatment arm landed 6.2% below control on a per-account basis over the test window.
That creates a genuine strategic fork. If the feed's purpose is responsible-gambling signaling — a visible, honest display of how rare jackpots are — then a 9% stake reduction is a feature working as intended, and the revenue cost is the price of the intervention. If the purpose is marketing, the feature is mispriced. The same UI element cannot be both a harm-reduction tool and an engagement driver, and operators running it as the latter are quietly running a brake they didn't design.
A Compliance Angle Worth Flagging
Several state regulators have moved toward requiring clearer disclosure of win probabilities on jackpot-style products. A capped feed is a crude version of that disclosure — it shows real outcomes at real frequency rather than implied odds on a paytable. Whether regulators would accept a curated three-per-hour sample as adequate disclosure is an open question, and a feed that cherry-picks the biggest wins would fail the test the moment anyone audited the sampling logic. The cap is defensible as a frequency control; it is not automatically defensible as a disclosure.
What the Effect Size Means in Dollars
Take a mid-tier operator with 40,000 daily actives and an average session-75 stake of $1.40. A 9% cut trims roughly $0.126 per late-session round. At an estimated 18 late-session rounds per active per day, that is about $2.27 per player per day, or roughly $90,800 daily across the base — before accounting for the 4% session-length increase, which claws back part of it. The net figure the operator reported, 6.2% NGR decline, is the number to use.
For a player, the arithmetic is simpler and less flattering to the operator. A 9% stake reduction does not reduce expected loss by 9% in the same proportion if the player extends session length by 4% and plays the same games. Expected loss scales with total handle, not stake per round. The feed changes the shape of the session more than its total cost, which is a meaningful distinction for anyone reading the headline as a harm-reduction win.
The Question Nobody Has Answered
The finding raises a question the test cannot answer: does the effect persist when the feed is opt-in? Every player in the treatment arm was exposed by default. A feed a player chooses to display is a different psychological object — self-selected, and probably self-reinforcing among players who want the comparison. The 9% figure may describe a default-exposure effect that evaporates, or inverts, once the feed becomes a setting. Until someone runs that arm, operators weighing whether to ship a capped win feed are deciding on the basis of a number that describes a population most of their players will never be in.