A/B routing is the practice of splitting live payment traffic between two processing routes, usually two acquirers or two configurations of one acquirer, to measure which performs better on approval rate, cost or latency. A rules engine or orchestration layer assigns a percentage of eligible transactions to each route, often by hashing a customer or order identifier, so seasonality and issuer outages hit both groups equally. Routes can differ by provider, network tokens, merchant account location or descriptor format. Approval rate alone misleads: sound tests also track net revenue per attempt, fraud and dispute rates, and exclude cascaded retries. For a buyer, documented routing tests show whether approval-rate claims rest on evidence.
What is A/B routing in payments
Merchants with more than one acquiring relationship rarely know which is better for a given slice of traffic. Published approval rates are averages across thousands of merchants and card mixes. A/B routing produces a merchant-specific answer by sending comparable transactions down each route at the same time, so seasonality, promotions and issuer outages affect both groups equally.
The routes being compared do not have to be different providers. A merchant can test the same acquirer with and without network tokens, with a local versus a cross-border merchant account, or with two different payment descriptor formats.
How does A/B payment routing work
The split is configured in a routing rules engine or an orchestration layer. A rule assigns a percentage of eligible transactions to route A and the rest to route B, usually by random allocation or by hashing a customer or order identifier so the same shopper always lands on the same route. The engine tags each transaction with its test group, and results are compared once each group has enough volume.
A typical test might send 50% of Visa and Mastercard debit transactions from German cardholders to each of two acquirers for four weeks. If route A approves 91.4% of 12,000 attempts and route B approves 89.1% of 12,000, the gap of 2.3 points is large enough to be meaningful, and on €2 million of monthly volume it represents roughly €46,000 of payments that would otherwise fail.
Reading the results correctly
Approval rate alone misleads. A route that approves more transactions can also generate more fraud chargebacks, charge higher interchange-plus markups, or send more payments to 3DS challenges that customers abandon. A sound test tracks net revenue per attempt, fraud rate and dispute rate alongside approvals.
Retries distort the numbers too. If a declined payment on route B is retried on route A through cascading, the retry has to be excluded or counted separately, otherwise route A looks better than it is.
From A/B routing to permanent rules
Once a test settles, the winning configuration becomes a standing rule, and the losing route often stays live for a small share of traffic as a benchmark and backup. Merchants that run tests continuously are effectively doing manual smart routing, and many later automate the allocation by BIN, country and card type. For a buyer of a payment company, documented routing tests show whether its approval-rate claims come from evidence or from a single acquirer's averages.