Live dealer tables lose 12% of hands to connection lag

Live dealer blackjack, roulette and baccarat tables are quietly shedding roughly one hand in eight to connection problems rather than to the house edge. That figure — 12% — comes from a six-month internal audit of session logs at three Central European operators, covering 4.1 million live-dealer rounds played between January and June 2025. It is not a measure of how often players lose. It is a measure of how often the round they were dealt never reached a conclusion at all: the stream froze, the bet timed out, the dealer's card stalled mid-flip, or the player was bounced to a reconnecting screen and returned to find the hand voided.

For anyone who has spent an evening on a Polish-facing live table, the number is unlikely to feel exaggerated. What is more interesting is where the losses land, and who absorbs them.

What the 12% actually contains

The audit split dropped hands into four buckets, and they are not equally distributed.

Cause	Share of dropped hands
Video stream stall over 2 seconds	41%
Betting-phase timeout (clock ran out)	27%
Dealer-side terminal or scanner fault	19%
Player-side disconnect	13%

The first bucket is the largest and the least visible. A live table is not one stream; it is typically three — a wide camera, a card or wheel close-up, and an overlay feed carrying the game state. When the overlay and the video fall out of sync by more than a few hundred milliseconds, the software cannot verify that the player saw the same card the dealer turned. Most operators void the hand rather than risk settling a disputed result. The player gets their stake back, which sounds neutral, but a voided hand is not a neutral event if you were holding a 20 against a dealer's 6.

The second bucket is behavioural as much as technical. Live blackjack tables run a 12-to-15-second betting window in most lobbies; some turbo variants cut it to 8. On a mobile connection with 180ms of round-trip latency — not unusual on Polish LTE outside the major cities — the clock can expire while the player is still tapping chips. The audit found timeout rates 2.6 times higher on mobile sessions than on desktop over the same period.

The latency arithmetic nobody publishes

Operators rarely disclose round-trip latency to their live studios, because the studios are third parties and the numbers are unflattering. The audit's own measurements put median latency from a Warsaw connection to a Malta-hosted live studio at 74ms, with a 95th percentile of 310ms. That tail is where the problems live. A 310ms round trip, plus client-side rendering, plus the operator's own risk checks, can push total input-to-confirmation time past 900ms on a bad evening.

That matters most in games where decisions are time-boxed. Live baccarat and roulette are comparatively forgiving — the betting window is longer and the outcome does not depend on the player acting. Live blackjack, and especially live poker variants where players act in sequence, is where the 12% concentrates. The audit's own breakdown puts dropped-hand rates at 8.4% for live roulette, 11.1% for live baccarat, and 17.3% for live blackjack. The headline average hides a game that is meaningfully worse than the rest.

There is a second-order effect that the raw figure misses. A dropped hand still consumed a seat. On tables with limited positions — seven seats is common for live blackjack — a frozen player occupies a spot that a waiting player cannot take. The audit estimated that on peak Friday and Saturday evenings, between 4% and 6% of live blackjack seats at any moment were held by sessions that were technically disconnected but not yet reaped by the operator's timeout logic.

Why operators do not treat this as a bug

Voided hands are, from a compliance standpoint, the safe outcome. Polish law under the amended Gambling Act of 2009, and the licensing conditions attached to it by the Ministry of Finance, puts the burden on the operator to demonstrate that the game was conducted fairly and that the player saw what the dealer saw. When the evidence is ambiguous, voiding the hand and returning the stake is the defensible choice. Settling it either way invites a complaint to the regulator, and the regulator does not accept "the stream probably showed the right card" as a defence.

So the incentive runs the other way. A voided hand costs the operator nothing in expected value — the stake goes back — but it costs the player the opportunity. Over 4.1 million rounds, 12% is roughly 492,000 hands that were dealt, interrupted, and unwound. If those hands had been played to completion at the table's normal edge, the arithmetic is small per hand but not trivial in aggregate. More importantly, the player experience is not neutral: a session where one hand in eight evaporates is a session that feels broken, and players who feel a table is broken leave it.

There is also a subtle asymmetry in how voids are handled. The audit noted that when a hand is voided after the player has already doubled or split, the additional stake is returned along with the original — but the time invested is not, and neither is the position in the shoe. Players who were counting on a favourable count in live blackjack find the shoe has moved on by the time they reconnect.

The mobile gap and the Polish connection

Poland's live-dealer audience skews mobile more heavily than its sportsbook audience, and the audit data reflects it. Mobile sessions accounted for 61% of live-dealer rounds in the sample but 74% of dropped hands. That gap is not explained by device performance alone — modern phones handle the streams fine. It is explained by network variance. A desktop on fibre has a stable path; a phone on a train between Kraków and Katowice does not.

Some operators have responded by shortening the betting clock on mobile, on the theory that a faster game reduces dead time. The audit suggests the opposite: shorter windows correlate with more timeouts, not fewer. The better-performing configuration in the sample used a longer window on mobile with a visible countdown and an audible cue, and it cut timeout-driven drops by roughly a third. That is an interface decision, not a network one, and it is cheap to make.

None of this is a reason to avoid live dealer tables, and the audit's authors are careful to say so. The 12% is a failure rate, not a loss rate. But it is a failure rate that operators have largely chosen not to compete on, because the cost of a dropped hand is borne mostly by the player's patience rather than the operator's margin. The open question is whether that changes when a regulator starts asking operators to publish their void rates alongside their RTP figures — or when a studio decides that a table which loses one hand in eight to lag is a marketing problem rather than an engineering one. Until then, the number stays where it is: invisible in the lobby, and 12% in the logs.