Arb bots: how automated arbitrageurs exploit stale odds across books

How the arb pipeline works

The operation starts with data: odds feeds from every book that matters, ingested and normalized into a single model of the market. The bot continuously compares each book's price against the consensus. When one book's line drifts far enough from the pack, the bot calculates the stake sizes for both sides and fires the bets, often through multiple accounts to stay under per-account limits.

The window is measured in seconds. Books with sharp risk teams correct stale lines quickly, so the bot has to detect, size, and place before the correction lands. This is why arb operations invest in low-latency infrastructure and direct API access wherever they can get it. A bot that places in 800 milliseconds beats one that places in three seconds, every time.

Why books hate it more than they fear sharp bettors

A sharp bettor is a cost of doing business; every book expects to lose to the best-informed players sometimes. Arb bots are different because they never take a position. They extract the disagreement itself, which means the book pays out on a structural inefficiency rather than losing a fair bet. There is no handicapping skill to respect, just latency to exploit.

Worse, arb activity distorts the book's own risk signals. A flood of arb-driven bets on one side of a market looks like informed money, which can trigger the risk team to move the line in the wrong direction. The bot does not just take the stale price; it can cause the book to misread its own market.

The fingerprints of arb automation

Arb bots leave a distinctive trail. Look for accounts that bet both sides of the same event across correlated markets, wager sizing that matches Kelly-optimal arb stakes rather than round numbers, and bet timing clustered within seconds of a line move at another book. Human bettors do not behave this way; their stakes are round, their timing is scattered, and they rarely hedge perfectly.

Cross-account patterns are the giveaway at scale. The operation needs many accounts to spread volume, so look for clusters: accounts created in batches, funded from the same sources, betting the same events within the same seconds. No single account looks abusive. The cluster does.

Defenses that raise the cost of arb

The structural fix is faster line management: the shorter the window your odds disagree with the market, the less there is to arb. Investing in automated line monitoring that flags your own stale prices is defense and good trading in one. Every stale line you correct before the bots find it is margin kept.

Account-level defenses include bet delays on suspicious patterns, reduced limits for arb-profile accounts, and requiring manual review for correlated multi-leg activity. The goal is not to eliminate arbitrage entirely, which is probably impossible, but to make your book the expensive one to arb against. Bots are rational; they migrate to softer targets.

Where the line sits on enforcement

Books have to be careful here. Aggressive arb enforcement can catch legitimate bettors: the fan who hedges a futures bet, or the line shopper who genuinely found a better price. The distinction is in the pattern, not the single bet. One hedged wager is normal behavior. Fifty perfectly sized cross-book hedges in a week from a cluster of related accounts is an operation.

Document the pattern before acting, and make the enforcement proportional. Limit reductions and bet delays are reversible; account closures are not. The books that handle arb best treat it as a risk-management problem with a technical solution, not a moral crusade. Keep the lines sharp, watch the clusters, and let the bots go eat someone else's stale odds.