Geo-Located Odds Feeds Shift In-Play Margins 12% in 400 Events

A 400-event study of in-play football and basketball markets has found that routing prices through geo-located odds feeds—rather than a single centralised feed—widened average in-play margins by roughly 12% across the sample, with the effect concentrated in the first 90 seconds after a scoring event. The finding, drawn from a dataset spanning 11 months and six jurisdictions, suggests that latency arbitrage and jurisdiction-specific liability management are now material determinants of the price a bettor actually sees, not marginal technical details.

Methodology and the 400-Event Sample

The study tracked 400 fixtures across football (287), basketball (94) and tennis (19) between March 2024 and January 2025. For each event, two parallel price streams were captured: a centralised feed sourced from a single Dublin-based aggregation point, and a geo-located feed with edge nodes in London, Malta, Gibraltar, Manila and São Paulo. Both streams were timestamped to the millisecond, and the resulting 2.14 million price observations were matched against settlement data to compute realised margin rather than posted margin.

The distinction matters. Posted margin—the overround visible on screen—is a bookmaker's intention. Realised margin is what survives after bets are accepted at stale prices. Much of the 12% gap in the sample comes from the second category: geo-located feeds cut the window in which a bettor can hit a price that has already moved elsewhere. The average stale-price acceptance rate fell from 3.8% of in-play bets on the centralised feed to 1.9% on the geo-located one, and that single change accounts for an estimated 7.1 of the 12 percentage points.

Why Margin Widened Rather Than Narrowed

A naive expectation would be that faster feeds compress margins through tighter competition. The data shows the opposite, and the mechanism is liability. When a feed updates in 180ms rather than 640ms, the trading desk can reprice before accumulated exposure on one side of a market becomes dangerous. The desk responds by offering shorter prices on the side it fears, which raises the effective overround. Faster information does not automatically mean better prices for the customer; it means the operator's risk window is smaller, and operators charge for the residual risk they still carry.

Where the 12% Concentrates

The aggregate figure hides a highly uneven distribution. Margins on pre-match markets barely moved—0.4% across the sample, well inside noise. The effect is almost entirely an in-play phenomenon, and within in-play it clusters around discrete events.

The 0–90-second window is where the two feeds diverge most sharply. On the centralised feed, a goal in a low-scoring football match produced a median repricing delay of 610ms; on the geo-located feed, 170ms. In that 440ms gap, roughly 2.3% of all in-play stakes in the sample were matched at prices that the geo-located feed had already superseded. For a mid-tier operator writing €40m a month in in-play turnover, that gap is worth somewhere between €180,000 and €260,000 in monthly margin—enough to justify the infrastructure cost several times over.

Sport-Specific Variance

Basketball showed the largest aggregate shift, at 15.2%, because scoring is frequent and each possession resets the market. Tennis was the outlier at just 4.8%, likely because point-level repricing is already fast on most feeds and because the two-outcome structure limits how much liability can accumulate on one side. Football sat between the two at 11.4%, but with the widest variance—some fixtures showed margin compression rather than widening, particularly low-liquidity matches where the geo-located feed had thinner price discovery and occasionally lagged the centralised one.

That last point deserves emphasis. Geo-located feeds are not uniformly faster. In the 38 fixtures with the lowest pre-match liquidity, the geo-located feed produced wider spreads without corresponding latency gains, and realised margin fell slightly. The technology rewards markets with enough depth to price efficiently at the edge; it punishes thin ones.

Regulatory Fragmentation as a Driver

Six jurisdictions were represented, and the margin effect correlated with regulatory divergence more than with geography. Jurisdictions with distinct liability rules, advertising restrictions or mandatory payout caps showed a 14.8% margin shift; those with harmonised rules showed 8.1%. This is consistent with a straightforward explanation: where a single operator must maintain different risk positions for different regulators, a geo-located feed lets it price each jurisdiction separately. The customer in a stricter market pays for the compliance burden embedded in that market's book.

There is a responsible-gambling dimension here that operators rarely discuss publicly. Wider in-play margins on fast-moving markets fall hardest on recreational bettors who chase live prices without understanding that the number they see is already stale. The 21.6% margin widening in the 0–90-second window is, in effect, a tax on reaction speed, and it is paid disproportionately by customers least able to price it. Several regulators have begun asking whether in-play markets should carry mandatory delay mechanisms—a 5-second bet-acceptance delay, for instance—which would neutralise much of the latency advantage and, by the study's logic, compress margins in exactly the window where they widened most.

The Infrastructure Cost Question

Geo-located feeds are not free. Edge nodes, redundant connectivity and the trading staff to monitor jurisdiction-specific books add up; operators in the sample reported incremental costs between €14,000 and €31,000 per month depending on the number of jurisdictions served. Against the margin uplift, the payback period for a mid-tier operator is under two months on in-play-heavy books, and considerably longer for operators whose revenue is predominantly pre-match. That asymmetry may accelerate consolidation: the technology favours scale, and smaller operators without the turnover to amortise edge infrastructure may find themselves pricing against competitors who can.

The 12% figure is also a snapshot of a moving equilibrium. If geo-located feeds become standard, the latency advantage erodes, and the margin uplift should compress toward the cost of the infrastructure itself. The study's authors estimate that full adoption across the six jurisdictions would reduce the effect to 3–4% within 24 months—at which point the differentiator shifts from feed architecture to pricing model.

The open question is not whether geo-located feeds improve margins, but who ultimately absorbs the difference. If the answer is recreational in-play bettors in fragmented jurisdictions, regulators will eventually notice, and the next 400-event study may be measuring the effect of mandatory delays rather than feed topology.