The average return-to-player on a sample of 240 UK-licensed video slots fell from 96.1% at 8pm to 94.0% by 11pm, a 2.1 percentage point drop that appeared within the same games rather than across a different game mix. Session telemetry from the same panel put median play time at 31 minutes before 9pm and 36.3 minutes after 11pm — a 17% extension — which is the sort of figure that looks trivial in isolation and rather less trivial once you multiply it by stake and spin rate.

That pairing is the whole story. Not that slots get worse at night, which sounds like a conspiracy, but that the two numbers are probably not causally linked in the direction most people assume, and the second one is the one operators can actually influence.

What the RTP figure does and doesn't show

Start with the obvious objection: a slot's published RTP is fixed in the paytable. A game configured at 96.1% returns 96.1% over its full cycle — millions of spins, not an evening's worth. It does not ratchet down when the clock strikes eleven.

So a measured drop has to come from somewhere else. Three candidates, in rough order of how much they explain.

Game mix. Late-night lobbies skew toward higher-variance, lower-RTP titles. A 94.2% Megaways-style game with 30,000x max win gets more traffic at 11pm than a 96.5% low-volatility pub-style slot does. If your sample weights by session rather than by spin, the mix shift alone can account for most of a two-point move.

Bonus-buy and feature entry. Players who buy into free spins or hold-and-win features late at night are, in effect, paying a premium for variance. The RTP on the buy is usually a little lower than the base game, sometimes by 1–2 points. Concentrate those purchases in a two-hour window and the aggregate number sags.

Sample noise. Genuinely. A 94.0% reading at 11pm over, say, 4 million spins has a standard error wide enough that the true figure might be 94.6% or 93.4%. Anyone quoting a single evening's number to two decimal places without a confidence interval is selling something.

The honest reading: RTP doesn't drop at 11pm. The population of games being played changes, and the population of players changes with it.

Where the 17% comes from

The session-length figure is more interesting because it survives the mix-shift explanation less comfortably. A 17% extension is large. It holds when you control for game type in most of the datasets I've seen, though the effect shrinks to somewhere around 9–12% once you strip out players who were already on long sessions earlier in the day.

What's left is a set of well-documented behavioural effects:

• Loss-chasing after a bad run. Evening sessions that start badly run longer than those that start well. This is not a controversial finding; it's been replicated in behavioural research for two decades.
• Reduced friction. No commute, no closing time, no reason to stop at a round number. The 11pm–1am window has fewer natural session boundaries than 6pm–8pm.
• Alcohol. Self-reported intake correlates with session length in most player surveys, and UK licensing already treats alcohol-adjacent environments as higher risk for a reason.
• Deposit-limit resets. A meaningful minority of UK players set daily limits that reset at midnight. Players who hit their cap at 10:45pm and wait it out are, by definition, still in session at 11pm.

That last one is a design artefact, not a player failing, and it's the one operators control outright.

The regulatory backdrop

The Gambling Commission's remote gambling and software technical standards already require licensees to monitor for markers of harm and intervene. Since the 2020 credit card ban and the 2021 strengthening of the customer interaction rules, the expectation has been that operators act on session data — not just deposit velocity.

A 17% session extension concentrated in a two-hour window is exactly the kind of pattern those rules were written for. Whether it's being acted on is a different question. The Commission's own 2023–24 enforcement caseload included several settlements where the failure was not "didn't have the data" but "had the data and didn't use it."

There's also the GamCare angle. The National Gambling Helpline's call volumes show a consistent late-evening peak, roughly between 10pm and 1am, which lines up almost exactly with the session-length data. That's correlation, not proof, but it's the same shape.

The 94% number is a distraction

Here's the part that gets lost. If RTP genuinely fell to 94% at 11pm, that would be a product-design scandal — operators quietly serving worse-value games to a captive late-night audience. It would be worth a Select Committee hearing.

What's actually happening is duller and harder to fix. Players at 11pm are playing different games, for longer, with weaker stopping cues, and a higher share of them are in the loss-chasing phase of a session. The RTP figure is a symptom of who's playing, not a cause of anything.

That matters because the two problems have different solutions. If RTP were dropping, you'd regulate paytables. Because session length is stretching, you regulate — or at least interrogate — the interaction model.

What operators could actually do

Three interventions that would move the 17% figure, in descending order of how likely they are to be adopted.

Shift deposit-limit resets. A limit that resets at 4am rather than midnight removes the "wait it out" incentive without reducing the limit's value. Some operators already stagger reset times by account creation date. There's no technical barrier; there's a commercial one.

Treat 11pm as a checkpoint, not a threshold. Most interaction models trigger on cumulative loss or deposit count within a session. A time-based check — a soft prompt at the 90-minute mark for sessions that started after 9pm — catches the specific pattern the data shows. The industry's objection is that prompts annoy low-risk players. That's true, and it's a cost worth measuring rather than assuming.

Publish session-length data by hour. No UK operator does this. A few publish average session length as a single number, which is close to useless. Hourly distributions would let researchers, and the Commission, see whether the pattern holds across the market or is concentrated in a handful of brands.

The third one is the interesting test. If the 17% extension is a market-wide behavioural fact, publishing it costs nothing and helps everyone. If it's concentrated in specific operators with specific design choices, publishing it is expensive — which is precisely why it hasn't happened.

So the open question isn't whether RTP drops at 11pm. It doesn't. The question is whether an industry that can measure a 17% session extension to the decimal place, and knows the helpline peaks in the same window, has any intention of doing something about it — or whether "we monitor for markers of harm" is going to keep meaning "we have the dashboard open."